# Technique 02: Anti-AI-Slop Defense System

## What It Is

A multi-layered detection and rewriting system that identifies and eliminates AI writing patterns from generated content. Based on research from 6+ GitHub anti-slop skills, Wikipedia's WikiProject AI Cleanup (24+ patterns), academic research on semantic ablation, and StyloAI's 31+ stylometric markers. This is not a post-processing "humanizer" -- it is a quality standard built into the writing process from the start.

**Framing note:** Never label this as "humanization" or "AI detection avoidance." Frame it as quality writing standards. The goal is not to trick detection -- it is to produce writing that is genuinely better than default AI output.

## Why It Works

AI content is detectable because of **semantic ablation** -- RLHF training systematically strips high-entropy information (unique metaphors, specific details, strong opinions), producing text with low burstiness (uniform sentence rhythm) and low perplexity (predictable word choice). These are the exact signals that both human readers and detection algorithms use to identify AI content.

Google's NavBoost system (confirmed in antitrust trial) demotes content that fails to engage users. Bland, detectable AI content drives pogo-sticking (users bouncing back to SERP), which directly hurts rankings.

**Key insight:** Post-processing (running text through a "humanizer") is detectable itself -- it creates a different statistical signature that is distinct from both pure AI and pure human writing. The solution is to generate human-like text from the start, not to fix it after.

## The Science of Detection

### Perplexity (How predictable each word is)
- **AI text**: Low perplexity -- the model picks the statistically most likely next word. "The results clearly demonstrate the significant impact..."
- **Human text**: Higher perplexity -- humans choose unexpected words, use slang, make idiosyncratic choices. "The numbers surprised us. Not because they were good -- they weren't -- but because they were exactly wrong."
- **Fix**: Instruct the model to use less common phrasings, domain-specific jargon, and unexpected vocabulary choices

### Burstiness (Sentence length variation)
- **AI text**: Remarkably uniform -- 15-25 words per sentence, consistent complexity
- **Human text**: Wild variation. Like this. And then suddenly a sentence that stretches across multiple clauses, connecting ideas that might not obviously belong together but make sense when you follow the writer's train of thought. Then short again.
- **Fix**: Explicitly vary sentence length between 3 and 40+ words per section

### Stylometric Patterns (31+ markers per StyloAI research)
- Function word frequency, sentence complexity distribution, vocabulary richness, syntactic diversity
- **AI text**: Clusters tightly in a predictable zone across all markers
- **Human text**: Much wider distribution with more variation

## The Tiered Banned Phrase System

### Tier 1: Always Remove (+3 points each)

These words and phrases are so strongly associated with AI that their presence alone flags content:

**Vocabulary:**
delve, tapestry, realm, landscape (metaphorical), multifaceted, nuanced, testament, cutting-edge, revolutionary, comprehensive (as adjective), crucial, compelling, vibrant, game-changer, leverage (verb), unlock potential, harness, endeavour, embark, navigate (metaphorical), pivotal, intricate, innovative, seamless, robust, transformative, meticulous, facilitate, utilize, commence, paramount, plethora, myriad, culminate, underscore, bolster, spearhead

**Phrases:**
- "It's worth noting that..."
- "In today's [anything] landscape"
- "Moreover" / "Furthermore" / "Additionally" (as sentence starters)
- "Let's dive in" / "Without further ado"
- "In conclusion" / "In summary" / "In essence"
- "It's important to note that..."
- "Stands as a testament to..."
- "Plays a vital/crucial/pivotal role"
- "In the realm of..."
- "It goes without saying"
- "Buckle up" / "Strap in"
- "The landscape is ever-evolving"
- "Marking a pivotal moment in the evolution of..."

**Replacements:**
| AI Word | Human Alternative |
|---------|------------------|
| leverage | use |
| utilize | use |
| commence | start, begin |
| facilitate | help, enable |
| endeavor | try, attempt |
| robust | strong, solid |
| seamless | smooth, easy |
| innovative | new, original |
| comprehensive | full, complete, thorough |
| paramount | critical, key |
| plethora | many, lots of |
| culminate | end in, lead to |
| underscore | show, highlight |
| delve | dig into, look at |
| testament | proof, sign |
| landscape | space, market, field |
| showcase | show |

### Tier 2: Flag When Clustered (+2 points each, flag at 3+ per article)

These are acceptable in isolation but signal AI when they cluster:

- however (more than 2x per article)
- here's the thing
- at the end of the day
- paired adjectives ("comprehensive and thorough", "innovative and cutting-edge")
- template openings ("In this article, we'll cover...")
- firstly / secondly / thirdly
- moving forward
- arguably
- notably
- essentially
- that said
- on the other hand
- it's no secret that

### Tier 3: Flag at High Density (+2 points per cluster of 3+)

- Transition words when overused (however, therefore, consequently, meanwhile)
- Corporate buzzwords (scalable, streamlined, optimize, synergy)
- Excessive qualifiers (quite, rather, somewhat, fairly, relatively)
- Hedging language (perhaps, might, could potentially, to some extent)

## Structural Pattern Detection

### Patterns to Flag and Rewrite

1. **Uniform sentence length** (+4 points): Three or more consecutive sentences within 3 words of each other in length. Fix: vary dramatically -- mix 5-word punches with 30-word complexes.

2. **Rule-of-three addiction** (+3 points): Habitual grouping of three examples, adjectives, or list items. Fix: use 2, 4, or 5 items instead.

3. **Topic-sentence-support pattern** (+3 points): Every paragraph starting with a topic sentence followed by supporting detail. Fix: start some paragraphs with an example, a question, or a statement that only makes sense after reading on.

4. **Binary contrast pattern** (+3 points): "It's not X, it's Y" or "No X, no Y, just Z." Fix: make the actual argument without the rhetorical frame.

5. **Staccato fragments** (+3 points): "Short. Punchy. Exhausting." -- dramatic fragmentation that feels manufactured. Fix: use fragments sparingly, for genuine emphasis only.

6. **Compulsive summarizing** (+2 points): Summary paragraph at the end of every section. Fix: only summarize if the section is genuinely complex (3+ subsections).

7. **Copula avoidance** (+2 points): "serves as" instead of "is", "functions as" instead of "is". Fix: just say "is" when that's what you mean.

8. **Synonym cycling** (+2 points): Using different words for the same concept in adjacent sentences to appear varied. Fix: repeat the same word if clarity demands it.

9. **Participial tack-ons** (+2 points): "...highlighting the importance of X" or "...underscoring the need for Y" at the end of sentences. Fix: delete or make a separate sentence.

10. **Em dash overuse** (+1 point per em dash beyond 1 per 500 words): AI uses em dashes far more frequently than humans. Fix: use commas, parentheses, or separate sentences.

### Additional Structural Patterns

11. **The hourglass structure**: AI structures text as synthesis -> detail -> synthesis. Humans start with a specific story, zoom out, dive into details in uneven bursts, and end with something specific again. **Break the hourglass.**

12. **Asymmetric coverage absence**: AI distributes word count evenly across all subtopics. Humans spend 500 words on the interesting part and 50 on the boring-but-necessary part. If every section is roughly the same length, flag it.

## The Two-Pass Audit System

### Pass 1: Detection and Rewrite

Scan the complete text against all three tiers plus structural patterns. For each detection:
1. Quote the flagged text
2. Explain why it's flagged (tier + specific pattern)
3. Provide a rewrite that preserves meaning while eliminating the pattern
4. Apply the rewrite

### Pass 2: Surviving Pattern Check

After Pass 1 rewrites, re-scan the full text. First-pass rewrites sometimes introduce new AI patterns (the model's natural tendency reasserts). Specifically check for:
- Recycled transitions that survived the first edit
- Lingering vocabulary inflation (replaced one AI word with another)
- Copula swaps that crept back in
- New structural uniformity introduced by the rewrites

### The Horoscope Test (Final Check)

For each paragraph, ask: "Could anyone have written this, for anyone, about anything?" If the answer is yes, the paragraph fails regardless of passing the phrase checks. Every paragraph must contain at least one element that is specific to THIS topic, THIS audience, THIS moment.

Score: +5 points for each paragraph failing the Horoscope Test.

## Risk Scoring

| Score | Risk Level | Action |
|-------|-----------|--------|
| 0-5 | Low | Content passes |
| 6-12 | Medium | Flag specific issues for human review |
| 13-20 | High | Auto-revise flagged sections |
| 21+ | Critical | Full rewrite required |

## Content Patterns to Eliminate

Beyond vocabulary and structure, eliminate these content-level patterns:

1. **No formulaic challenges**: "Despite facing challenges, the company continues to thrive" -- name the actual challenge or cut it
2. **No vague attributions**: "Studies show..." -> "A 2024 Ahrefs study of 100K pages found..."
3. **No significance inflation**: "Marking a pivotal moment in the evolution of..." -> Cut the entire phrase. Start with what actually happened.
4. **Use contractions**: "doesn't" not "does not." "We've found" not "We have found."
5. **Use colloquial transitions**: "Thing is," "Here's the catch," "The weird part?" instead of "Furthermore," "Moreover," "Additionally"
6. **Repeat the clearest word** instead of cycling synonyms: If the subject is "customers," don't alternate between "clients," "users," "patrons," and "consumers"

## Tips

- **The "bar conversation" test**: Would you say this sentence to a colleague at a bar? If not, it's too formal or AI-sounding. "The implementation demonstrated significant improvements in key performance indicators" -> "The thing actually worked. Revenue went up 18%."

- **The Horoscope Test**: Could this paragraph appear in any article, about any topic, for any audience? If yes, it has zero specificity. Every paragraph needs at least one element unique to THIS topic.

- **Self-correction is human**: "At first I thought this was a branding problem -- turns out it was pricing all along" shows the writer's actual thought process. AI does not self-correct mid-argument.

- **Tangential value**: Brief tangents that add context but aren't strictly necessary: "Side note: we tested this during Black Friday, which might have skewed results upward." AI stays rigidly on-topic. Humans digress when the digression adds value.

- **Asymmetric coverage**: Spend 500 words on the interesting part and 50 on the boring-but-necessary part. AI distributes word count evenly. Humans allocate attention where it matters.

- **Imperfections are features**: One slightly awkward phrasing, one mild redundancy, one sentence that could be tighter -- these are signals of human writing. Perfect prose = AI prose.

- **Domain convention awareness**: Academic writing IS more formal than blog writing. Match the expected register for the content type. Don't make an academic paper sound like a blog post, and don't make a blog post sound like a dissertation.

- **The rhythm check**: Count sentence lengths in each paragraph. If standard deviation is < 5 words, rewrite for more variation.

## Common Mistakes

1. **Thinking "more editing" = more human** -- Over-humanizing creates its own detectable pattern. Don't make every sentence a fragment or start every paragraph with "Look,". One technique used excessively becomes its own signature.

2. **Using "humanizer" tools** -- These are detectable as post-processed text. The statistical signature is different from both pure AI and pure human writing. They solve the wrong problem.

3. **Thinking typos = human** -- Intentional typos are a different pattern than natural typos. Don't add them deliberately.

4. **Ignoring domain conventions** -- Academic writing IS more formal than blog writing. A technical whitepaper should not read like a casual blog post. Match the expected register for the content type.

5. **Removing patterns without adding personality** -- Anti-slop auditing alone produces sterile text. Voice injection (Technique 03) must happen BEFORE anti-slop auditing.

6. **Thinking "different words" = fixed** -- Replacing one AI word with another AI word (e.g., "leverage" -> "harness") accomplishes nothing. Replace with genuinely plain language ("use").

## Implementation Notes

- The tiered system allows the writing phase to use Tier 2/3 words occasionally (natural) while the audit phase catches clusters
- Pass 2 is essential -- research shows first-pass rewrites retain 15-20% of AI patterns
- The Horoscope Test catches generic content that avoids banned words but remains soulless
- Voice injection (Technique 03) must happen BEFORE anti-slop auditing -- removing patterns without adding personality produces sterile text
- Anti-detection instructions should be included IN the writing prompt, not as a separate post-processing step

## Key Sources

- [blader/humanizer](https://github.com/blader/humanizer) -- 1,600+ stars, based on Wikipedia's 24 patterns
- [conorbronsdon/avoid-ai-writing](https://github.com/conorbronsdon/avoid-ai-writing) -- 102-entry tiered replacement table
- [hardikpandya/stop-slop](https://github.com/hardikpandya/stop-slop) -- 8 rules, 50-point scoring
- [aplaceforallmystuff/the-antislop](https://github.com/aplaceforallmystuff/the-antislop) -- Horoscope Test
- [DonAldente-AI/anti-slop-system](https://github.com/DonAldente-AI/anti-slop-system) -- Naive drafter + ruthless critic pattern
- [Wikipedia: Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing)
- [Semantic Ablation (The Register)](https://www.theregister.com/2026/02/16/semantic_ablation_ai_writing/)

# Content Types Overview

Quick-reference index covering all 23 content-type templates. Use this file to pick the right content type for a SERP, check its target word count, dominant schema, snippet format, and H1/H2 skeleton without loading the full per-type file. Load the individual `content-types/<type>.md` only when you need the depth (process detail, anti-AI focus, common mistakes, example topics).

## Decision table

| Content type | Dominant intent | Word count | Primary schema | Snippet format | H1 pattern | Key H2s (first 3-4) |
|---|---|---|---|---|---|---|
| how-to | Informational | 1,500-3,000 | HowTo + FAQ | Ordered list | "How to [Action] [Qualifier]" | Quick answer / Prerequisites / Step-by-step / Common mistakes |
| definition | Informational | 1,000-2,000 | DefinedTerm + FAQ | Paragraph | "What Is [Term]? [Subtitle]" | Quick definition / How it works / Types / Benefits |
| pillar-page | Informational + navigational | 3,000-5,000+ | Article + FAQ | Paragraph (list per section) | "The Complete Guide to [Topic] [Year]" | Table of contents / Overview / Core subtopics / Practical steps |
| faq-page | Informational | 1,500-2,500 | FAQPage + BreadcrumbList | Paragraph (per Q) | "[Topic] FAQ: Veelgestelde vragen over [topic]" | Intro / Category 1 / Category 2 / Costs |
| statistics | Informational | 2,000-3,500 | Article + Dataset | List | "[Topic] Statistics [Year]: [X] Key Facts and Trends" | Key highlights / Category stats / Regional data / Trends |
| news | Informational + navigational | 800-1,500 | NewsArticle + FAQ | Paragraph | "[News event]: [Impact or what it means]" | Key takeaways / What happened / Why it matters / What to expect |
| thought-leadership | Informational | 1,200-2,500 | Article + Person | Paragraph (low priority) | "[Strong opinion or provocative question]" | Conventional wisdom / Why it's wrong / Supporting arguments / Counterarguments |
| glossary | Informational | 5,000-15,000 (200-400/term) | DefinedTermSet + BreadcrumbList | Paragraph (per term) | "[Industry] Woordenlijst: [X] Begrippen Uitgelegd" | Intro / A / B / C ... |
| comparison | Commercial investigation | 1,200-2,500 | FAQ + Product | Table | "[Product A] vs [Product B]: [Outcome promise]" | Quick verdict / At-a-glance table / How we compared / Head-to-head |
| listicle | Commercial investigation | 2,000-4,000 | ItemList + FAQ | List | "Best [Category] for [Audience/Use Case] in [Year]" | Top picks at a glance / How we evaluated / 1. Top pick / 2. Second pick |
| alternatives | Commercial investigation | 1,500-3,000 | ItemList + FAQ | List | "[X] Alternatieven: [N] Beste Opties in [Year]" | Why look for alternatives / Comparison table / Alternative 1 / Alternative 2 |
| product-review | Commercial investigation | 1,500-2,500 | Review + Product | Paragraph | "[Product] Review [Year]: [Honest verdict]" | Quick verdict / What is [Product] / Our experience / Pros |
| buying-guide | Commercial investigation | 2,000-3,500 | Article + FAQ | List | "[Category] Kopen: Waar Moet Je Op Letten?" | Quick checklist / Criterion 1 / Criterion 2 / Budget guide |
| product-page | Transactional | 1,000-1,500 | Product + Offer + AggregateRating | Product rich result | "[Product Name] — [Key Differentiator]" | Description / Features and specs / What's included / Who is this for |
| category-page | Commercial + navigational | 300-1,000 (editorial) | CollectionPage + ItemList | List | "[Category Name]" | Intro / Product grid / Koopgids / Subcategories |
| landing-page | Transactional | 800-1,200 | WebPage or Service + FAQ | Not primary | "[Clear value proposition in one line]" | Problem statement / Solution / How it works / Social proof |
| pricing-page | Transactional | 500-1,500 | Product + Offer | Table | "Prijzen" or "[Product] Pricing" | Pricing tiers / Feature comparison table / Which plan / What's included |
| integration-page | Commercial + navigational | 800-1,500 | SoftwareApplication + HowTo | List | "[Your Product] + [Partner Tool] Integratie" | What it does / Key features / How to set it up / Use cases |
| service-page | Commercial + transactional | 1,200-2,000 | Service + LocalBusiness + FAQ | List | "[Service Name] — [Key Benefit or Location]" | What we do / Who this is for / How it works / Deliverables |
| location-page | Local + transactional | 1,200-1,500 | LocalBusiness + Service + FAQ | Paragraph | "[Service] in [Location]" | Local context / Services in [Location] / Why choose us / Recent projects |
| case-study | Commercial investigation | 1,500-2,500 | Article + Organization | Paragraph | "[Client/Industry]: [Key Result in Numbers]" | Overview / The challenge / Our approach / The results |
| about-page | Navigational + trust | 500-1,500 | Organization + Person | Knowledge Panel (not snippet) | "Over [Brand]" or "About [Brand]" | Mission / Our story / Our team / Expertise |
| programmatic-page | Informational + commercial | 500-1,200 | Depends on type + BreadcrumbList | Table | "[Variable A] [Connector] [Variable B]" | Data summary / Key data / How [B] compares / Trends over time |

## Intent → content type mapping

Search intent is the primary filter. Pick based on what the top 5 SERP results are actually doing, not what you think they should be doing.

**Informational.** User wants to understand something. Not buying yet.
- "how to [x]" / "[x] tutorial" → how-to
- "what is [x]" / "[x] meaning" → definition
- "[topic] explained" / "guide to [topic]" → pillar-page
- "[topic] questions" / PAA cluster → faq-page
- "[topic] statistics" / "[topic] data [year]" → statistics
- "[term] vocabulary" / 30+ terms in niche → glossary
- "[event] [year]" / breaking topic → news
- "why [industry belief]" / contrarian query → thought-leadership

**Commercial investigation.** User is shopping but hasn't picked a product.
- "best [category] for [audience]" → listicle
- "[A] vs [B]" → comparison
- "[product] alternatives" → alternatives
- "[product] review" → product-review
- "[category] buying guide" / "what to look for in [x]" → buying-guide
- "[service] results" / "[industry] case study" → case-study

**Transactional.** User is ready to convert.
- "[product name]" / "[product] kopen" → product-page
- "[product] pricing" / "[product] kosten" → pricing-page
- "[service] aanvragen" / "[tool] free trial" → landing-page
- "[service] inhuren" / "[service] kosten" → service-page
- "[tool A] + [tool B] integration" → integration-page
- "[service] in [city]" / "[service] near me" → location-page

**Navigational + trust.** User is verifying who you are.
- "[brand] about" / "who is behind [brand]" → about-page
- "[category]" root browse → category-page
- Long-tail pattern at scale with structured data → programmatic-page

Note: several content types span two intent classes. Pillar pages are informational but also navigational (hub for the cluster). Category pages are commercial but also navigational (people type the category name to browse). Pick the dominant intent, but don't be surprised when a page has to satisfy both.

A few more overlaps worth flagging. Integration pages and category pages both straddle commercial and navigational. Programmatic pages can skew either informational or commercial depending on the variable. Service pages are commercial in SERP but behave like transactional once the visitor lands.

## When to load the full per-type file

The decision table gives you enough to pick the type and draft the skeleton. Load the full `content-types/<type>.md` when:

- **pillar-page**: you need the hub-and-spoke linking architecture, the visual-break rhythm (image/table every 300-500 words), and the regular-update cadence rules. The overview doesn't capture the "every cluster article links back" rule, which is the whole point of the pattern.
- **comparison**: you need the full 10-section framework, especially the "verdict first per criterion" rule at the H3 level. Comparison articles fail hard if you just know the H2 skeleton.
- **how-to**: you need the quick-answer targeting (40-60 words, ordered list after the most relevant H2) and the step-by-step "one action per step" constraint. The anti-AI section on error messages and version numbers is the highest-signal part.
- **case-study**: you need the PAS-adjacent structure (challenge → approach → results → takeaways) and the "lead with the number in H1 and first 50 words" rule. The overview skips the client-permission and visual-proof requirements.
- **product-review**: you need the "who should skip" framing and the Google review-update requirements around hands-on evidence.
- **programmatic-page**: you need the quality-gate logic (noindex thin pages, conditional template copy, sibling linking). Shipping programmatic without reading the full file gets you penalized.
- **listicle**: you need the "methodology section + updated quarterly" rule and the 7-10 item sweet spot.
- **glossary**: you need the A-Z navigation pattern, anchor-link scheme, and the cross-reference style between terms.
- **location-page**: you need the doorway-page warning. Location pages without genuinely unique per-city content get penalized. The overview doesn't make this loud enough.

For everything else the overview table is usually enough to draft the first pass. Go deeper if the draft feels generic or if you're unsure about a specific structural element.

## Cross-reference: YMYL-sensitive types

YMYL ("your money or your life") topics trigger elevated E-E-A-T and Trust scrutiny from Google's quality raters. The content types that live closest to YMYL need named authors, visible credentials, transparent methodology, and verifiable evidence. That list: **pricing-page**, **service-page**, **product-review** (Google's review update explicitly targets this type), **case-study**, **about-page** (the page raters read to assess the whole site's trust), **thought-leadership**, and any **news** article about health, finance, legal, or safety topics. For these, skip stock photos, add real author bylines with linked credentials, include dated evidence (screenshots, numbers, timelines), and publish visible update logs. If you're writing in a regulated niche (medical, financial advice, legal), the Trust bar is higher than any of the per-type files capture. Add a subject-matter expert reviewer byline and cite primary sources (regulators, peer-reviewed studies) rather than secondary blogs.

## Word count targets by intent

The overriding rule: match the average word count of the top 5 ranking results, then add roughly 10%. Don't pad to hit a number. Don't under-write to stay "concise" when the SERP is hungry for depth.

| Intent class | Typical range | Rule of thumb |
|---|---|---|
| Informational (simple) | 800-1,500 | Match SERP average; news and short definitions fall here |
| Informational (deep) | 1,500-3,500 | How-to, statistics, FAQ, buying guide: depth signals expertise |
| Informational (hub) | 3,000-5,000+ | Pillar pages only; every section earns its place |
| Commercial investigation | 1,200-3,000 | Comparison, listicle, alternatives, review: enough to build confidence before conversion |
| Transactional | 500-1,500 | Pricing, product, landing, integration: short and scannable, conversion over coverage |
| Trust / navigational | 500-1,500 | About, category: authentic over long, unique content over padding |
| Programmatic at scale | 500-1,200 per page | Quality gates first; noindex anything thin |

When the SERP has mixed formats (one pillar, three listicles, one FAQ), write for the dominant format but cover the angles the minority formats introduce. If the top 5 are all 3,500 words and yours is 1,200, you're not competing. You're just hoping.

## Related reference files

If you need the underlying methodologies rather than the content-type skeletons, load these from the same `references/` directory:

- `intent-matching.md`: the full intent-classification logic and SERP-reading playbook
- `serp-driven-writing.md`: how to read the top 5 and extract the winning format
- `structured-data-snippets.md`: the full schema-markup catalog with examples
- `anti-slop-ruleset.md`: the banned-vocabulary list and structural tells
- `voice-injection-playbook.md`: how to inject first-person experience into any type
- `eeat-signal-embedding.md`: where and how to place author, credentials, and evidence

# Technique 04: EEAT Signal Embedding

## What It Is
Systematically embedding Experience, Expertise, Authoritativeness, and Trustworthiness signals into content — not as claims ("we're experts") but as demonstrated proof through content structure, source attribution, and writing patterns.

## Why It Works
Google's algorithm leak confirmed that EEAT is evaluated through proxy signals. Google cannot independently verify if content is accurate — it uses structural indicators that correlate with knowledgeable sources. Content that SHOWS expertise through depth, specificity, and honest nuance outranks content that merely CLAIMS expertise.

The December 2025 Core Update specifically intensified EEAT evaluation, particularly the "Experience" component — first-hand involvement signals.

## The Four EEAT Components — What They Actually Mean

### Experience (Most underrated)
- **What Google says**: Content created by someone with first-hand experience
- **What it actually means**: Specific details only someone who DID the thing would know
- **Signal strength**: Very high since the "first E" was added in December 2022

**Weak experience signals (don't work):**
- "I have 15 years of experience in marketing"
- "As an expert in this field..."
- Author bio listing credentials

**Strong experience signals (actually work):**
- "When we ran this campaign for [Client], the Facebook CPC averaged €2.40 — until we changed the creative on day 7 and it dropped to €0.85"
- "The first time I set this up, I forgot to configure the DNS TXT record and spent 3 hours debugging. Save yourself the trouble: do it first."
- Mentioning specific tool versions, error messages, unexpected results
- Describing what DIDN'T work before finding what did

### Expertise (Depth over breadth)
- **What it actually means**: Content that demonstrates understanding of nuance, edge cases, and tradeoffs — not surface-level coverage of everything
- **The expertise paradox**: Real experts narrow their scope. Beginners try to cover everything. Content that says "this works for B2B SaaS with >€5K ACV but NOT for consumer apps" demonstrates more expertise than content covering all business types generically.

**Expertise signals:**
- Discussing when advice DOESN'T apply
- Explaining the "why" behind each recommendation
- Acknowledging tradeoffs ("this increases conversion but may increase support tickets")
- Using precise terminology correctly (not keyword-stuffing but natural domain language)
- Addressing edge cases and exceptions

### Authoritativeness (Being cited, not claiming)
- **What it actually means**: Other people and sites reference you as a source
- **This is mostly a domain-level signal**, not a content-level one
- Content can support it through: accurate citations of others (builds reciprocal authority), original research that others want to cite, named author with verifiable credentials

### Trustworthiness (Transparency and accuracy)
- **What it actually means**: The content is verifiable, transparent about limitations, and doesn't mislead
- Every statistic has a named source with date
- Affiliate relationships or biases are disclosed
- Claims are hedged when appropriate ("in our experience" vs "always")
- Errors are corrected (dated correction notes build trust)

## Step-by-Step Process

### Step 1: Experience Signal Planning
1. Before writing, identify 3-5 experience markers to embed:
   - A specific project/client/implementation story
   - A mistake made and lesson learned
   - A specific tool, metric, or process detail
   - An unexpected result or counterintuitive finding
   - A comparison between expected and actual outcome
2. These CANNOT be generated by AI — they must come from the human author

### Step 2: Expertise Signal Architecture
3. Structure the article to demonstrate depth:
   - Cover the "advanced" angle, not the "intro" angle (unless intent is beginner)
   - Include "when NOT to do this" sections
   - Discuss tradeoffs explicitly
   - Address 2-3 edge cases or exceptions
4. Use precise domain terminology naturally (not forced)

### Step 3: Source Attribution (Trustworthiness)
5. Every factual claim must have a source:
   - Academic papers → Author, year, journal
   - Industry reports → Organization, year, specific finding
   - Statistics → Source name, year, methodology note
   - Expert opinions → Named person, credentials, context
6. Replace ALL instances of:
   - "Studies show..." → "[Organization]'s [year] study of [N] [subjects] found..."
   - "Experts agree..." → "[Named expert], [credentials], argues that..."
   - "Research suggests..." → "A [year] [journal] paper by [author] demonstrated..."

### Step 4: Author Authority Integration
7. Include author byline with specific credentials relevant to THIS topic
8. Link to author's other published work on the topic
9. Reference personal involvement: "In my role as [specific title] at [company], I..."
10. If applicable, reference speaking engagements, publications, or patents

### Step 5: Trust Signals
11. Date the content prominently
12. Note when information was last verified
13. Disclose any affiliations, sponsorships, or biases
14. Include a methodology section for any original research/data
15. Add a corrections section if updating previously published content

## Tips

- **The "screenshot proof" technique**: Including actual screenshots of dashboards, results, or tools is one of the strongest experience signals. It's nearly impossible to fake and instantly demonstrates first-hand involvement.
- **Date your screenshots**: Screenshots with visible dates prove recency and experience
- **Name specific dollar amounts**: "$2,400/month" is more trustworthy than "thousands per month." Real practitioners know exact numbers.
- **Cite yourself**: If you've written about this topic before, reference your own earlier work with what you've learned since. This creates a trail of genuine expertise.
- **Use first-person sparingly but specifically**: Not "I think" for opinions, but "I tested" for experience and "I found" for results.

## Common Mistakes

1. **EEAT as a checklist**: Adding an author bio and citing 3 sources doesn't make content trustworthy. The EEAT signals must be woven through the content, not bolted on.
2. **Fake experience markers**: "Imagine you're building a website..." is not an experience signal. "When I built [specific site]..." is. AI defaults to hypotheticals; real EEAT uses specifics.
3. **Over-attributing to build trust**: Every sentence having a citation reads like an academic paper, not expert content. Cite where it matters; assert where you have authority.
4. **Confusing EEAT with keywords**: Adding "expert" and "authoritative" into your content doesn't improve EEAT. Demonstrating expertise through depth and specificity does.

## Research Evidence

### Kyle Roof's Testing-Based Position
E-E-A-T elements will NOT help you rank, but they help you KEEP your rank once evaluated. Roof's controlled experiments (ranking a Lorem Ipsum page for "Rhinoplasty Plano") proved on-page keyword signals alone can achieve rankings. E-E-A-T is defensive, not offensive.

**The priority:** Trust > Expertise > Experience > Authoritativeness. Kyle Roof: "Who cares about the E or the A -- it's all about the T. Trust."

### Cyrus Shepard's 50-Site Case Study (4,000+ Websites)
Across 2023 Google updates:
- **Winners:** Used first-person pronouns, demonstrated first-hand experience
- **Losers:** Excessive ads (14.01 per page vs. 6.32 on winners), over-optimized anchor text
- 17 on-page features showed statistically significant correlations with ranking gains/losses

### December 2025 Core Update
Specifically intensified EEAT evaluation, particularly the "Experience" component -- first-hand involvement signals became stronger ranking factors.

# Technique 08: Fact-Checking & Verification

## What It Is
A systematic process to verify every factual claim, statistic, quote, and attribution in AI-assisted content — because AI models confidently fabricate data, invent sources, and present plausible-sounding but incorrect information.

## Why It Works
One fabricated statistic destroys an article's credibility. Google's EEAT framework heavily weights Trustworthiness, and factual accuracy is its foundation. More practically: if a reader fact-checks ONE claim and it's wrong, they dismiss the entire article — and bounce back to search results (NavBoost penalty).

AI hallucination rates for factual claims vary by model and domain, but even the best models fabricate 5-15% of specific statistics and quotes. For an article with 20 factual claims, that's 1-3 fabricated "facts" by default.

## Step-by-Step Process

### Step 1: Claim Extraction
1. Read the draft and highlight every factual claim:
   - Statistics ("63% of marketers...")
   - Quotes ("According to [name]...")
   - Named studies or reports
   - Historical facts ("Founded in 2019...")
   - Tool capabilities ("Supports up to 100 users...")
   - Comparisons ("3x faster than...")
2. Create a verification checklist with each claim

### Step 2: Source Verification
3. For each statistic:
   - Can you find the ORIGINAL source (not another blog citing it)?
   - Is the source reputable (academic, government, major research firm)?
   - Is the year current? (Statistics from 2020 may not apply in 2026)
   - Does the exact number match? (AI often rounds or inverts percentages)
4. For each quote:
   - Did this person actually say this?
   - Is it attributed to the right person?
   - Is it in context? (AI often paraphrases and presents as direct quotes)
5. For each named study:
   - Does the study exist?
   - Does it say what the article claims it says?
   - Is the author/organization correct?

### Step 3: Claim Classification
6. Classify each claim:
   - **Verified**: Source found, claim matches
   - **Partially correct**: Source exists but numbers/context differ
   - **Fabricated**: No source found, likely hallucinated
   - **Unverifiable**: First-party data that only the author can confirm
7. Fix or remove all fabricated claims
8. Correct all partially correct claims to match actual sources
9. Flag unverifiable claims for human verification

### Step 4: Attribution Standards
10. Replace all vague attributions:
    - "Studies show..." -> "[Organization]'s [year] study of [N subjects] found..."
    - "Experts agree..." -> "[Name], [title] at [organization], states..."
    - "Research suggests..." -> "A [year] paper in [journal] by [author] found..."
    - "According to industry data..." -> "According to [specific report name]..."
11. Every statistic must have: source name, year, and ideally a link

### Step 5: Currency Check
12. Flag statistics older than 2 years for review
13. Note when data was last updated
14. Add "as of [date]" qualifiers for time-sensitive data
15. Check if more recent data is available

## Tips

- **The "Google Scholar" test**: If you can't find a study on Google Scholar, it probably doesn't exist. AI loves citing plausible-sounding studies that were never published.
- **Reverse-search quotes**: Paste any direct quote into Google with quotation marks. If it only appears in AI-generated content, it's fabricated.
- **Watch for "statistic laundering"**: AI often cites a real organization but invents the specific number. "According to HubSpot, 78% of..." — HubSpot exists, but they never published that 78% figure.
- **Common AI fabrication patterns**:
  - Round numbers (AI loves 67%, 83%, 91% — exact round numbers)
  - Recent years (AI often cites "[current year - 1]" studies that don't exist)
  - Well-known organizations (AI cites McKinsey, HubSpot, Forrester with invented numbers)

## Common Mistakes

1. **Trusting AI citations**: Even when AI provides a URL, the page often doesn't exist or doesn't contain the cited information
2. **Skipping first-party data verification**: When the human provides data, still verify it makes sense (basic sanity checks)
3. **Verifying against other AI content**: If your "source" is another AI-generated article that cites the same fake statistic, you've verified nothing
4. **Removing instead of fixing**: Sometimes the claim is directionally correct but the specific number is wrong. Find the real number rather than cutting valuable data points

# Technique 06: GEO (Generative Engine Optimization)

## What It Is

Optimizing content to be cited by AI systems -- ChatGPT, Claude, Perplexity, Google AI Overviews. The foundational academic work is [GEO: Generative Engine Optimization (Aggarwal et al., Princeton / Georgia Tech / Allen Institute for AI / IIT Delhi, KDD 2024)](https://arxiv.org/abs/2311.09735), which found 40% visibility improvements in generative engine responses using core GEO techniques. This is the emerging complement to traditional SEO.

## Why It Works

AI systems cite content that is: easy to extract, verifiable, authoritative, and structured for natural language queries. Content optimized for AI citation also tends to rank well in traditional search because the same qualities (clarity, specificity, authority) that AI systems value are what Google's quality signals reward.

## Key Statistics

- 38% of AI Overview citations come from top-10 ranking pages in early 2026, down from 76% in mid-2025 ([Ahrefs](https://ahrefs.com/blog/ai-overview-citations-top-10/)). AI Overviews and classical ranking are decoupling fast.
- FAQ schema materially improves citation rates in AI Overviews per vendor studies (Relixir's 50-site study reports 41% vs 15% with/without proper FAQPage schema).
- Authors with presence across multiple platforms (Wikipedia, Reddit, LinkedIn) get cited more often than isolated profiles.
- Brand search volume correlates more strongly with AI citations than raw backlink count.
- ChatGPT shows the strongest recency preference of any engine — most-cited pages tend to be updated within the past 30 days.
- Over 53% of pages cited in AI Overviews are under 1,000 words ([Ahrefs](https://ahrefs.com/blog/short-vs-long-content-in-ai-overviews/)).

## Step-by-Step Process

### Step 1: Answer-First Formatting (+340% AI citations per vendor study)

Lead every major section with a clear, extractable answer before expanding with detail.

**Pattern:**
```
## [Question as H2]

[Direct answer in 40-60 words -- clear, factual, citable]

[Expanded explanation with context, evidence, and nuance]
```

**Why it works:** AI systems scan for concise, authoritative statements they can extract and cite. An answer buried in paragraph 3 of a section is less likely to be cited than one at the top.

### Step 2: FAQ Schema Implementation (+28% AI citations per vendor study)

FAQPage schema makes content materially more likely to appear in AI Overviews — see the Relixir study cited in Key Statistics.

**Implementation:**
- Include 3-5 FAQ items per page
- Use actual questions people search for (from PAA data in the brief)
- Answer in 40-60 words per question
- Implement FAQPage JSON-LD schema

### Step 3: Claim-Evidence Pairs

Structure content as verifiable claims followed by evidence. AI systems prefer content they can confidently cite.

**Pattern:**
```
[Factual claim with specific data point]. [Source attribution].
[Supporting evidence or context].
```

**Example:**
"Email marketing generates an average ROI of EUR 36 for every EUR 1 spent (DKG, 2025). This makes it the highest-ROI digital channel, outperforming social media (EUR 12.71) and paid search (EUR 8.14)."

### Step 4: Entity-Rich Writing

Explicitly name entities and their relationships. AI systems build knowledge graphs from entity-rich content.

**Implementation:**
- Name specific companies, people, tools, frameworks (not "many companies" but "Coolblue, Bol.com, and Zalando")
- Define relationships between entities explicitly
- Use structured data (Organization, Person, Product schema) to reinforce entity markup
- Link to authoritative external sources for entity validation

### Step 5: Structured Q&A Format

Clear question-answer patterns that AI can extract and cite.

**Implementation:**
- Use questions as H2/H3 headings where natural
- Answer immediately in extractable format
- Include "who/what/when/where/why/how" question variants
- Structure complex topics as series of questions

### Step 6: Source Attribution

Include verifiable statistics with explicit source citations.

**Implementation:**
- Cite specific sources by name and year: "(Ahrefs, 2025)" not "according to research"
- Include URLs for original data where possible
- Prefer recent sources (within 12 months)
- Link to original research, not secondary summaries

## Six Trust Signals for AI Citation

1. **Authorship attribution** -- clear author name with verifiable credentials
2. **Structured data** -- Schema.org/JSON-LD markup for content type
3. **Verifiable references** -- named sources with dates
4. **Consistent brand identity** -- consistent entity signals across platforms
5. **Recency** -- updated within 30 days for maximum ChatGPT citation chance
6. **Transparent data provenance** -- methodology disclosed for original claims

## GEO Measurement Metrics

| Metric | What It Measures |
|--------|-----------------|
| Prompt Coverage | How often your content appears in AI responses for target queries |
| Citation Share | Your percentage of total citations vs. competitors |
| Visibility Score | Prominence of your citations (position, context) |
| Authority Weight | How AI systems rate your domain's reliability |
| Freshness Ratio | Recency of your most-cited content |

## Dual Optimization: Google + AI

The writing skill should optimize for BOTH simultaneously. The overlap is large:

| Signal | Google Ranking | AI Citation |
|--------|---------------|-------------|
| Clear, extractable answers | Featured snippets | Direct citation |
| FAQ schema | Entity understanding | 3.2x citation rate |
| E-E-A-T signals | Quality score | Authority weight |
| Structured headings | Crawlability | Parsability |
| Specific data with sources | Information gain | Verifiability |
| Recency/freshness | QDF signal | Recency preference |
| Entity-rich content | Knowledge Graph | Entity linking |

**What differs:**
- Google rewards engagement signals (NavBoost) -- AI systems don't (yet)
- AI systems prefer concise, extractable text -- Google rewards comprehensive coverage
- AI systems heavily weight recency -- Google balances recency with authority

**Resolution:** Write comprehensive content (for Google) with clear, extractable summary answers at the start of each section (for AI). Keep content fresh (for both).

## Tips

- Test your content against actual AI systems: paste your target query into ChatGPT, Claude, and Perplexity and see if your content gets cited
- Prioritize FAQ schema -- the 3.2x citation rate is the highest-leverage GEO technique
- Keep extractable answers under 60 words -- longer answers are less likely to be cited verbatim
- Update content within 30-day cycles for maximum ChatGPT citation probability

## Common Mistakes

1. **Optimizing for AI only** -- classical ranking still correlates with AI citations (38% of AI Overview citations come from top-10 pages per Ahrefs 2026). Traditional SEO is still the foundation.
2. **Burying answers in paragraphs** -- AI systems need clear, extractable statements at section tops
3. **Vague attributions** -- "research shows" is not citable; "(Ahrefs, 2025)" is
4. **Ignoring structured data** -- FAQ schema provides a measurable citation lift

## Key Sources

- [krillinai/GEO (GitHub)](https://github.com/krillinai/GEO)
- [awesome-generative-engine-optimization (GitHub)](https://github.com/amplifying-ai/awesome-generative-engine-optimization)
- [AgriciDaniel/claude-blog (GitHub)](https://github.com/AgriciDaniel/claude-blog) -- +340% citations from answer-first, +28% from FAQ schema
- [aaron-he-zhu/seo-geo-claude-skills (GitHub)](https://github.com/aaron-he-zhu/seo-geo-claude-skills) -- CORE-EEAT framework
- [Ahrefs: Short vs Long Content in AI Overviews](https://ahrefs.com/blog/short-vs-long-content-in-ai-overviews/)

# Technique 17: Human Input Framework -- The Critical Differentiator

## What It Is
A structured system for collecting and integrating human-provided content into AI-assisted articles -- because AI without human input produces high-quality slop. The human's experiences, data, opinions, and case studies are what transform generic AI content into genuinely valuable, ranking-worthy articles.

## Why It Works
Every technique in this research converges on one truth: **AI content that ranks is content that includes things only a human could provide.** Information Gain requires unique data. EEAT requires demonstrated experience. Anti-AI-detection requires genuine opinions and idiosyncratic phrasing. NavBoost requires content that actually satisfies users.

AI alone cannot provide:
- First-party data from your business
- Case studies with real clients, real numbers, real timelines
- Genuine opinions and contrarian perspectives based on experience
- Specific tool interactions, error messages, and unexpected results
- The "what went wrong" stories that build trust

## The Human Input Hierarchy

### Tier 1: Essential (Content won't rank without these)
1. **First-party data**: Numbers from your own business, research, or analysis
2. **Case study details**: Client name (or anonymized), problem, solution, specific results
3. **Genuine opinion**: What the author actually thinks about the topic, including disagreements with conventional wisdom

### Tier 2: High-Value (Significantly improves quality)
4. **Specific experience details**: Tool versions, error messages, unexpected behaviors encountered
5. **Process documentation**: The exact steps the author follows, with reasoning for each
6. **Failure stories**: What was tried and didn't work, and why
7. **Comparisons from actual usage**: "I used tool A and tool B for 3 months each. Here's what I found."

### Tier 3: Enhancing (Adds polish and authenticity)
8. **Analogies and metaphors**: The author's unique way of explaining a concept
9. **Predictions/opinions**: Where the author thinks the industry is heading
10. **Behind-the-scenes context**: Why certain decisions were made, what alternatives were considered
11. **Personal anecdotes**: Brief relevant stories that illustrate points

## Step-by-Step Process

### Step 1: Pre-Writing Interview (5-10 questions)
Before the AI pipeline starts, collect human input via structured questions:

**For blog posts / thought leadership:**
1. "What's YOUR take on [topic]? What do you disagree with that most people believe?"
2. "Can you share a specific example or case study related to [topic]?"
3. "What numbers or data do you have from your own experience?"
4. "What's the most common mistake you see people make with [topic]?"
5. "If you had to give one piece of counterintuitive advice on [topic], what would it be?"

**For how-to guides / tutorials:**
1. "When you do [process], what's the exact sequence of steps?"
2. "What usually goes wrong? What error messages do people see?"
3. "What prerequisite do beginners always forget?"
4. "Is there a shortcut or trick that makes this significantly easier?"
5. "What tools/versions do you use, and does it matter?"

**For product/comparison content:**
1. "Which option do you actually recommend and why?"
2. "What's the biggest hidden drawback that nobody mentions?"
3. "Who should NOT use this product/approach?"
4. "What's changed about this product/market in the last 6 months?"
5. "Can you share specific metrics from your usage?"

### Step 2: Input Integration
6. Map each human input to a specific section of the outline
7. Use direct quotes where the human's phrasing is distinctive
8. Weave data points into arguments (don't dump them in a "data" section)
9. Use case studies as section anchors (start sections with the story)
10. Let opinions drive the article's angle, not just flavor paragraphs

### Step 3: Input Verification
11. If the human provides statistics, verify they're reasonable
12. If the human names tools or products, verify they exist and are current
13. If the human references clients, confirm disclosure is appropriate
14. Flag any claims that need supporting evidence

### Step 4: Quality Gate
15. Before finalizing, check: "What in this article could ONLY come from this specific human?"
16. If the answer is "nothing" or "just the case study in paragraph 7," more human input is needed
17. Target: at least 20% of the article's value should be human-sourced

## The Input Collection Interface (MCP Skill Design)

A well-designed workflow should implement a structured input collection:

```
CONTENT WRITING REQUEST
========================
Topic: [user provides]
Content Type: [user selects from 10 types]
Target Keyword: [user provides]

HUMAN INPUT REQUIRED
========================
DATA: Do you have any data or numbers to include?
   [text input -- first-party data, metrics, percentages]

CASE STUDY: Can you share a specific case study or example?
   [text input -- client story, project outcome, before/after]

OPINION: What's YOUR opinion on this topic? Any disagreements with conventional wisdom?
   [text input -- personal take, contrarian view]

CONTRARIAN: What common advice on this topic do you think is WRONG?
   [text input -- things that don't work, outdated advice]

PROCESS: Any specific tools, processes, or methods you use?
   [text input -- exact steps, tool names, configurations]

FAILURES: What usually goes wrong with this? Any failure stories?
   [text input -- mistakes, unexpected results, lessons learned]

[Optional] Upload voice samples for brand voice matching
[Optional] Paste 2-3 example articles in your style
```

## Tips

- **Audio input is often better than text**: People share more detailed, natural-sounding experiences when speaking vs typing. If possible, accept voice memos and transcribe them.
- **The "why" follow-up**: When the human says "we use tool X," follow up with "why did you choose X over alternatives?" -- the reasoning is the information gain, not the tool name.
- **Anonymization template**: Provide a simple template for case studies that need anonymization: "A [industry] company with [X employees] in [region]..."
- **Capture the messy version**: Raw, unpolished human input often sounds MORE authentic than cleaned-up versions. Preserve the original voice.

## Common Mistakes

1. **Making human input optional**: If the pipeline allows "skip" on all human input fields, users will skip everything -- and the content will be generic
2. **Collecting input after writing**: Human input should shape the OUTLINE and WRITING, not be bolted on as quotes after the fact
3. **Over-processing human input**: Don't AI-rewrite the human's case study. Their natural phrasing is an anti-detection signal.
4. **One input, many articles**: Each article should have UNIQUE human input. Reusing the same case study across 10 articles weakens information gain.

# Technique 01: Information Gain Writing

## What It Is
Writing content that adds genuinely NEW information compared to what already ranks for a keyword. Based on Google's Information Gain patent (granted June 2024), which scores how much additional value a page provides beyond existing search results.

## Why It Works
Google's algorithm compares your content against other pages the user has already seen on the same topic. Pages that say the same thing as the top 10 get low Information Gain scores. Pages that add unique data, perspectives, or insights get high scores — and rank higher.

The API leak confirmed this via `OriginalContentScore` — a separate metric that evaluates content uniqueness across the index.

**Algorithm reasoning:** If 10 pages all explain "how to do X" the same way, Google gains nothing by showing an 11th identical page. But if page 11 adds a case study, original data, or a contrarian perspective? That's information gain — value the user can only get from YOUR page.

## Step-by-Step Process

### Step 1: SERP Gap Analysis
1. Search the target keyword and read the top 10 results fully
2. Create a spreadsheet: rows = topics covered, columns = each competitor
3. Mark what each competitor covers and — critically — what they DON'T cover
4. Identify patterns: where do all 10 say the same thing? Where do they disagree?
5. Note the "missing angles": perspectives, data types, or use cases nobody addresses

### Step 2: Unique Value Identification
6. Ask: "What do I know about this topic that these 10 pages don't include?"
7. Sources of unique value:
   - **First-party data**: "We analyzed 500 customer accounts..."
   - **Original case studies**: "Client X tried this and here's what happened..."
   - **Expert interviews**: "I spoke with [Name], who said..."
   - **Contrarian perspective**: "Most guides say X. In our experience, Y works better because..."
   - **Process documentation**: "Here's the exact 9-step process we use internally..."
   - **Failure stories**: "We tried the common approach and it failed because..."
   - **Tool comparison**: "We tested 4 tools and measured actual results..."
8. Select 3-5 information gain elements to include

### Step 3: Content Architecture
9. Structure the article to lead with unique insights, not rehashed basics
10. Place information gain elements in the first 30% of the content (Google evaluates engagement early)
11. Use unique headings that signal novel content (not generic "What is X?" and "Benefits of X")
12. Plan specific data points, quotes, and examples for each section

### Step 4: Writing with Information Gain
13. Every section must answer: "What can the reader ONLY learn here?"
14. Replace generic statements with specific ones:
    - Bad: "Many companies have seen success with this approach"
    - Good: "We implemented this for 12 e-commerce clients in 2025. Average conversion improvement was 23%, but 3 clients saw no change — all in the B2B space"
15. Add "not found elsewhere" sections: edge cases, failure modes, advanced tips

## Hidden Tips & Tricks

- **The "So What?" test**: After every paragraph, ask "So what? Where can I ONLY read this?" If the answer is "anywhere," the paragraph has zero information gain.
- **Use your analytics**: Your GSC data, your customer data, your A/B test results — these are information gain goldmines that competitors literally cannot replicate.
- **Cite non-obvious sources**: Everyone cites HubSpot and Ahrefs. Cite academic papers, industry reports from niche organizations, or government data nobody else uses.
- **The 10-10-80 rule**: 10% covering basics (for context), 10% discussing what the competition says, 80% unique content. Most AI content is 80-10-10 — the opposite.

## Common Mistakes

1. **Thinking "more words" = information gain** — A 5,000-word article repeating the same points as competitors has zero information gain. A 1,500-word article with original data has high gain.
2. **Adding information gain at the end** — Google evaluates engagement early. Put unique insights in the first 500 words, not the conclusion.
3. **Fabricating data** — Never make up statistics for information gain. One fabricated stat that gets fact-checked destroys all credibility.
4. **Confusing "different format" with "different information"** — Putting the same information in a table instead of paragraphs isn't information gain.

## When to Use This Technique

- **Always** for competitive keywords (keyword difficulty > 30)
- For any content where you have access to unique data or experiences
- When updating content that lost rankings (likely lost due to competitors with higher information gain)
- Critical for pillar/hub content that anchors a topic cluster

## Algorithm Confirmation (from Google Leak)

The Information Gain patent (granted June 2024) explicitly describes the scoring system:

> "Information gain scores indicate how much more information one source may bring to a person who has seen other sources on the same topic. Pages with higher information gain scores may be ranked higher."

This means Google is not just comparing your page to competitors -- it's comparing it to pages the *specific user* has already seen in their search session. Returning to search after reading 3 similar articles? The 4th needs to add something NEW.

**Tools Used:**
- keyword search — identify target keyword landscape
- competitor data — analyze who ranks for this topic
- opportunity detection — find keywords where we rank 4-20
- SERP feature detection — identify featured snippet / PAA opportunities
- page-level SEO data — analyze current page performance if updating

# Technique 10: Search Intent Matching

## What It Is
Writing content that matches the exact type of information a user expects when they search a keyword -- not just covering the topic, but delivering it in the FORMAT and DEPTH the user needs at that moment in their journey.

## Why It Works
Google's NavBoost system (confirmed by the 2024 API leak) tracks what happens after a user clicks a search result: do they stay, engage, and convert -- or do they bounce back to try another result? Content that mismatches intent causes pogo-sticking, which directly harms rankings.

The December 2025 Core Update doubled down on intent matching: pages optimized for the wrong intent dropped 30-90% regardless of content quality or domain authority.

## The Four Intent Types

### 1. Informational ("I want to learn")
- **Keywords**: "how to...", "what is...", "why does...", "guide to..."
- **User expectation**: Comprehensive answer, step-by-step instructions, explanations
- **Format**: Long-form articles, how-to guides, explainers, tutorials
- **Critical first 100 words**: Must signal "this page teaches you [exact topic]"
- **SEO signals**: HowTo schema, FAQ schema, table of contents

### 2. Commercial Investigation ("I want to compare")
- **Keywords**: "best...", "vs...", "review...", "top 10...", "comparison..."
- **User expectation**: Hands-on comparison, recommendations, honest assessment
- **Format**: Comparison tables, review articles, listicles with scoring
- **Critical first 100 words**: Must mention the things being compared and signal real testing
- **SEO signals**: Product schema, Review schema, comparison tables

### 3. Transactional ("I want to buy/do")
- **Keywords**: "buy...", "price...", "sign up...", "download...", "[product name]"
- **User expectation**: Clear pricing, easy purchase path, conversion-focused
- **Format**: Product pages, landing pages, pricing pages
- **Critical first 100 words**: Value proposition + CTA visibility
- **SEO signals**: Product schema, Offer schema, clear CTAs

### 4. Navigational ("I want to find a specific page")
- **Keywords**: "[brand name]", "[brand] login", "[product] pricing"
- **User expectation**: The exact page they're looking for
- **Format**: The actual page (homepage, login, pricing)
- **SEO signals**: Brand entity signals, sitelinks

## Step-by-Step Process

### Step 1: Intent Classification
1. Search the target keyword in Google
2. Look at the TOP 3 results -- Google has already classified intent
3. Note the content TYPE ranking: are they blog posts? Product pages? Comparison articles? YouTube videos?
4. If top 3 are all comparison articles and you write a how-to guide, you WILL NOT rank

### Step 2: Format Matching
5. Match the dominant format in the SERP:
   - All listicles -> write a listicle
   - All how-to guides -> write a how-to guide
   - Mix of formats -> the intent is fragmented; pick the one that aligns with your content type
6. Exception: if ALL top 10 use the same format, there's an opportunity to differentiate IF you still match intent (e.g., a video among all text results)

### Step 3: Depth Calibration
7. Measure competitor word counts for the top 5 results
8. Your content should match or slightly exceed the depth level
9. **But**: more words does not equal better depth. A 5,000-word article for "what time is it in Tokyo" is absurd. Match the COMPLEXITY of the intent.

### Step 4: First-100-Words Optimization
10. The first 100 words must immediately signal intent match
11. **Informational**: Answer the core question within the first paragraph, then expand
12. **Commercial**: Name the products/options being compared and establish credibility
13. **Transactional**: State the value proposition and make the CTA visible
14. Users who don't see intent match in 5 seconds bounce -- NavBoost penalizes this

### Step 5: Content Structure for Intent
15. Structure the article to match how users consume content for each intent:
    - **Informational**: TL;DR -> detailed sections -> summary/checklist
    - **Commercial**: Quick comparison table -> detailed reviews -> recommendation
    - **Transactional**: Benefit headline -> features -> social proof -> CTA
    - **Mixed**: Address primary intent first, secondary intents in supporting sections

## Content Length by Intent

Ahrefs found near-zero correlation (0.04) between word count and ranking. Match the SERP, not a word count target.

| Intent | Optimal Length | Rationale |
|--------|---------------|-----------|
| Informational (comprehensive guide) | 1,500-3,000+ | Needs to cover the full topic |
| Informational (specific question) | 800-1,500 | Answer directly, don't pad |
| Commercial investigation (comparison) | 1,200-2,500 | Depth in evaluation criteria |
| Commercial investigation (roundup) | 2,000-4,000 | Multiple items need coverage |
| Transactional | Minimal needed | No fluff; specs, pricing, CTA |
| Local service | 300-500 unique words | Must be genuinely unique per location |

**Key rule:** Match the SERP. If top 5 results average 2,000 words, write 2,000-2,500. If they average 800, writing 3,000 wastes effort and may signal wrong intent.

## Tips

- **Check "People Also Ask"**: These reveal the JOURNEY. If PAA for "best CRM" includes "is Salesforce worth it?" and "CRM pricing comparison," users are in commercial investigation -- not informational.
- **Search intent can shift**: "AI writing tools" in 2023 = informational ("what are they?"). In 2026 = commercial investigation ("which one should I use?"). Always verify current SERP, don't assume.
- **Mobile vs desktop intent**: The same keyword can have different intent on mobile (quick answer) vs desktop (deep research). Optimize for both, but prioritize based on traffic split.
- **Featured snippet = informational confirmation**: If Google shows a featured snippet, the intent is informational. Structure your content to capture it (direct answer in 40-60 words).

## Common Mistakes

1. **Writing a blog post for a transactional keyword**: "Buy running shoes" -> users want a product page, not a 2,000-word guide
2. **Writing a product page for an informational keyword**: "How to choose running shoes" -> users want guidance, not a buy button
3. **Ignoring mixed intent**: Some keywords have genuinely mixed intent. Your page must address the primary intent fully and nod to secondary intents
4. **Over-optimizing for one intent type**: Don't force transactional CTAs into informational content. It feels manipulative and users bounce.

# Technique 07: NavBoost Engagement Optimization

## What It Is
Writing content optimized for Google's NavBoost system — the confirmed (via 2024 API leak) user behavior tracking that measures clicks, dwell time, pogo-sticking, and engagement to influence rankings.

## Why It Works
NavBoost was described in the antitrust trial as "one of the most important signals" Google uses for ranking. It tracks: click-through rate from SERPs, time on page, scroll depth, whether users return to search results (pogo-sticking), and what they do next. Content that satisfies users on the first click gets rewarded. Content that causes users to bounce back gets demoted.

This is why a perfectly SEO-optimized page with correct keywords can still drop in rankings — if users don't engage with it, NavBoost pushes it down.

## Step-by-Step Process

### Step 1: First-Click Satisfaction (The Critical First 10 Seconds)
1. **Meta title must promise what the page delivers** — no clickbait
2. **First paragraph must match search intent** within 2 sentences
3. **Answer the core question immediately** — don't make users scroll to find what they searched for
4. Informational: "Here's how to [do X]..." or "The answer is [X]. Here's why..."
5. Commercial: Show the comparison/recommendation immediately
6. **Visual above the fold**: Table of contents, key statistic, or comparison table — give users a reason to stay

### Step 2: Dwell Time Optimization
7. **Progressive depth**: Start with the answer, then go deeper. Users who got their quick answer may leave (that's fine — they were satisfied). Users who want depth will keep scrolling.
8. **Engaging hooks per section**: Each H2 should make the user want to read that section. "The counterintuitive part" > "Key Findings"
9. **Break up long sections**: No wall of text longer than 150 words without a visual break (subheading, image, list, quote block, or callout)
10. **Add interactive elements descriptions**: Tables, comparison charts, calculators — these increase time on page

### Step 3: Anti-Pogo-Sticking
11. **Match the search intent exactly**: If someone searches "how to fix 404 errors" and your first 300 words are about what 404 errors are, they'll bounce
12. **Address secondary intents**: After answering the primary question, address the logical next question. This prevents users from going back to search for follow-up info
13. **"Is this what you were looking for?" check**: Read your first paragraph. If you had searched the target keyword, would you feel you're in the right place?

### Step 4: Scroll Depth Signals
14. **Valuable content throughout**: Don't front-load all value. Include surprising insights, data points, or practical tips in the latter half
15. **Use pattern interrupts**: After every 300-400 words, include something visually different — a callout box, a quote, a comparison table, an example
16. **End with value, not filler**: The conclusion should add something new (a checklist, a template, a specific recommendation), not summarize what was already said

### Step 5: Click Satisfaction Signals
17. **Fulfill the promise**: If the meta title says "7 Steps," deliver exactly 7 actionable steps
18. **No false scarcity**: Don't gate content behind email signups when the meta description implies the answer is on the page
19. **Outbound links that work**: Dead links = frustration = bounce
20. **Page speed**: Pages that take > 3s to load lose 53% of visitors (confirmed by Google). LCP < 2.5s is the target

## The NavBoost Feedback Loop

```
Good content -> Users stay -> NavBoost up -> Rankings up -> More traffic -> More engagement data -> Rankings up further
Bad content -> Users bounce -> NavBoost down -> Rankings down -> Less traffic -> Less data -> Stagnation
```

This is a compounding effect. Content that performs well on NavBoost gets MORE traffic, which provides MORE positive signals, which further improves rankings. The opposite is also true — once content starts losing NavBoost signals, it spirals down.

## Tips

- **The "back button test"**: After reading your article, would you hit the back button? If yes, the content didn't fully satisfy your intent. Fix it.
- **Branded search is the ultimate NavBoost signal**: The SerpMonsters case study showed that a clinic jumped from position 28 to top 3 when their branded search volume increased 300% (from an offline campaign). Google interpreted the branded searches as trust signals.
- **CTR optimization is NavBoost optimization**: A higher CTR from SERPs means more clicks, more engagement data, stronger NavBoost signal. Optimize meta titles for clicks, not just keywords.
- **Long dwell time does not equal good signal**: If users spend 10 minutes on your page but then search the same thing again, that's a NEGATIVE signal. Dwell time only matters if the user is SATISFIED.

## Common Mistakes

1. **Burying the answer**: SEOs trained in the "skyscraper" era write 500-word introductions before getting to the point. Modern NavBoost rewards immediate answers.
2. **Clickbait meta titles**: High CTR but high bounce rate = net negative NavBoost signal. The title and content must match.
3. **Ignoring mobile UX**: 60%+ of searches are mobile. If your content doesn't render well on mobile (horizontal scroll, tiny text, intrusive popups), users bounce regardless of content quality.
4. **Too many CTAs**: Aggressive conversion attempts mid-content interrupt reading flow and increase bounce rate.

# Technique 16: Quality Scoring System

## What It Is
A 100-point composite scoring system that evaluates AI-generated content across five dimensions before output. Content scoring below threshold is automatically revised. This is the final quality gate preventing bad content from being published.

## Why It Works
Organizations using structured quality gates report 45% fewer post-publication content issues. The scoring system provides objective, repeatable assessment and identifies WHICH areas need improvement rather than a binary pass/fail.

## Scoring Breakdown (100 Points Total)

### Content Quality (30 points)

**Topical Completeness (0-10):**
- Are all SERP-common topics covered? (from brief's competitor analysis)
- Are PAA questions from the brief addressed?
- Scoring: 10 = all topics covered, 7 = 80%+, 5 = 60%+, 3 = 40%+, 0 = below 40%

**Depth of Coverage (0-10):**
- Does each major section contain specific examples, data points, or analysis?
- Are claims supported with evidence rather than stated as fact?
- Scoring: count sections with at least one specific example/data point. 10 = all sections, proportional down

**Originality (0-10):**
- Does the content contain at least one section adding information not found in the top 10 SERP results?
- Are there unique insights, original analysis, contrarian perspectives, or proprietary data?
- Scoring: 10 = multiple unique sections, 7 = one strong unique section, 4 = unique framing of known info, 0 = pure rehash

### SEO Optimization (25 points)

**Keyword Placement (0-8):**
- Primary keyword in meta title: +2
- Primary keyword in H1: +2
- Primary keyword in first 100 words: +2
- Primary keyword or variation in 2+ H2s: +2

**Internal Links (0-5):**
- 3-5 verified internal links per 1,000 words: +3
- All links validated against real sitemap: +2
- Deduct 2 points per hallucinated link

**Meta Quality (0-4):**
- Meta title under 60 chars with keyword + hook: +2
- Meta description under 155 chars with CTA: +2

**Schema Markup (0-4):**
- Correct primary schema for content type: +2
- FAQ schema present (if FAQ section exists): +2

**Featured Snippet Targeting (0-4):**
- Target snippet format identified and implemented: +2
- 40-60 word direct answer at correct position: +2

### E-E-A-T Signals (15 points)

**First-Person Experience (0-5):**
- Uses "I/we tested," "in my experience," "our data shows": +2
- Contains specific case study or personal testing data: +3
- Scoring: presence of genuine experience signals, not just the words

**Specific Data & Examples (0-5):**
- Named companies, specific numbers, dated events: +2
- At least one example that is not a common/obvious example: +3
- Deduct points for "many companies" or "experts agree" without names

**Source Attribution (0-5):**
- Claims backed by named sources with dates: +3
- External links to authoritative references: +2

### Anti-Slop Score (15 points)

**Zero Tier 1 Phrases (0-5):**
- 5 = zero Tier 1 words/phrases detected
- Deduct 1 point per Tier 1 detection (minimum 0)

**Burstiness (0-3):**
- Measure sentence length standard deviation
- 3 = high variation (SD > 8 words), 2 = moderate (SD 5-8), 1 = low (SD 3-5), 0 = uniform

**Voice Consistency (0-3):**
- Content matches the loaded voice document characteristics
- 3 = strong match, 2 = moderate, 1 = weak, 0 = no voice detected

**Horoscope Test (0-4):**
- Count paragraphs that pass the specificity check (topic-specific, audience-specific)
- 4 = all pass, 3 = 80%+, 2 = 60%+, 1 = 40%+, 0 = below 40%

### AI Citation Readiness (15 points)

**Answer-First Formatting (0-5):**
- Key sections lead with extractable 40-60 word answers: +3
- Direct answer appears within first 100 words of the article: +2

**FAQ Schema (0-3):**
- FAQPage schema present with 3+ properly formatted Q&As: +3

**Claim-Evidence Pairs (0-4):**
- Data claims include named source + date: +2
- Statistics include specific numbers (not "many" or "most"): +2

**Entity Richness (0-3):**
- Named entities (companies, people, tools) used instead of generic references: +2
- Entity relationships explicitly stated: +1

## Readability Targets

| Audience | Flesch Reading Ease | Grade Level | When to Use |
|----------|-------------------|-------------|-------------|
| General consumer (NL) | 60-70 | Grade 7-9 | Comparison sites, how-tos, consumer guides |
| Professional / B2B | 50-60 | Grade 9-12 | SaaS content, industry analysis |
| Technical | 40-50 | Grade 12+ | Developer docs, technical guides |

Readability is assessed but NOT part of the 100-point score (since optimal readability varies by audience and content type).

## Quality Gate Actions

| Score Range | Risk Level | Action |
|-------------|-----------|--------|
| 85-100 | Ready | Output for publication |
| 70-84 | Minor issues | Output with annotations on weak areas for human review |
| 60-69 | Needs revision | Auto-revise weak sections (re-enter Phase 4), maximum 2 cycles |
| 0-59 | Major issues | Full rewrite (re-enter Phase 3), alert human reviewer |

## Auto-Revision Strategy

When score is 60-69, target the weakest category first:

1. **Low Content Quality:** Add missing topics from brief, inject specific examples into thin sections
2. **Low SEO:** Fix keyword placement, add missing internal links, generate meta tags
3. **Low E-E-A-T:** Add first-person experience language, inject specific data points, add source citations
4. **Low Anti-Slop:** Run additional editing passes against banned phrase list, increase sentence length variation
5. **Low GEO:** Restructure key sections with answer-first formatting, add FAQ schema, add source attributions

**Maximum 2 revision cycles.** Research (Self-Refine) shows diminishing returns after 2 cycles, and quality can actually degrade with over-revision.

## Reporting Format

```markdown
## Quality Report

**Overall Score: XX/100** [Ready | Minor Issues | Needs Revision | Major Issues]

| Category | Score | Status |
|----------|-------|--------|
| Content Quality | XX/30 | [pass/flag] |
| SEO Optimization | XX/25 | [pass/flag] |
| E-E-A-T Signals | XX/15 | [pass/flag] |
| Anti-Slop | XX/15 | [pass/flag] |
| AI Citation Ready | XX/15 | [pass/flag] |

**Readability:** Flesch XX (Grade Level X) -- [appropriate/too complex/too simple for target audience]

**Issues Found:**
1. [Specific issue + location in content + suggested fix]
2. [...]

**Strengths:**
1. [What scored well]
2. [...]
```

## Key Sources

- [AgriciDaniel/claude-blog](https://github.com/AgriciDaniel/claude-blog) -- 100-point system across 5 dimensions
- [aaron-he-zhu/seo-geo-claude-skills](https://github.com/aaron-he-zhu/seo-geo-claude-skills) -- 80-item CORE-EEAT framework
- [hardikpandya/stop-slop](https://github.com/hardikpandya/stop-slop) -- 50-point scoring across 5 dimensions
- [AI Content Quality Control Guide 2026 (Koanthic)](https://koanthic.com/en/ai-content-quality-control-complete-guide-for-2026-2/)
- [Self-Refine: Iterative Refinement (Learn Prompting)](https://learnprompting.org/docs/advanced/self_criticism/self_refine)

# Technique 14: SEO Optimization Layer

## What It Is
The technical SEO optimization applied after content is written and humanized -- keyword placement, meta elements, structured data, internal linking, and readability scoring. This is the mechanical layer that ensures good content is actually FOUND by search engines.

## Why It Works
Quality content without SEO optimization is invisible. SEO optimization without quality content is empty. This layer bridges the gap -- it takes content that's already good (information gain, EEAT, anti-detection) and makes it technically discoverable.

The key insight: SEO optimization should be the LAST step, not the first. Writing "for SEO" produces stiff, keyword-stuffed content. Writing for humans THEN optimizing for SEO produces content that ranks AND reads well.

## Step-by-Step Process

### Step 1: Keyword Placement Verification
1. **H1**: Contains primary keyword (naturally, not forced)
2. **First 100 words**: Primary keyword appears at least once
3. **H2 headings**: 2-3 of them include secondary keywords
4. **Keyword density**: 1-2% for primary keyword (NOT higher)
5. **LSI/related terms**: Naturally distributed throughout (topic coverage, not keyword stuffing)
6. **Alt text**: If images are present, describe them with relevant keywords

### Step 2: Meta Elements
7. **Meta title** (50-60 characters):
   - Contains primary keyword near the beginning
   - Includes a value proposition or differentiator
   - Uses power words for CTR (how, best, guide, [year])
   - Format: `[Primary Keyword]: [Value Proposition] | [Brand]`
   - Generate 3-5 options for A/B testing
8. **Meta description** (150-160 characters):
   - Summarizes the page's unique value
   - Contains primary keyword (bolded in SERPs)
   - Includes a call-to-action or hook
   - Does NOT over-promise (reduces bounce rate)
   - Generate 3 options

### Step 3: Content Structure Optimization
9. **Table of contents**: Add for articles > 1,500 words (jump links)
10. **Header hierarchy**: H1 -> H2 -> H3 (no skipping levels)
11. **Short paragraphs**: Max 3-4 sentences (improves mobile readability)
12. **Scannable elements**: Bold key phrases, use bullet lists for quick-reference items
13. **Above-the-fold content**: Primary keyword and value proposition visible without scrolling

### Step 4: Internal Linking
14. Query internal link data for link opportunities
15. Add 3-5 internal links to related content:
   - Link FROM the new page TO established pages (passes topic relevance)
   - Link TO the new page FROM established pages (passes authority)
16. Use descriptive anchor text (not "click here" or "read more")
17. Link to cornerstone/pillar content where relevant

### Step 5: External Linking
18. Include 2-3 external links to authoritative sources:
   - Government sites, academic papers, industry reports
   - These signal trust and support EEAT
19. Avoid linking to direct competitors
20. Use `rel="noopener"` for new-tab links (not nofollow for legitimate citations)

### Step 6: Structured Data
21. Select schema type based on content type:
   - Blog post -> `Article` schema with `author`, `datePublished`, `dateModified`
   - How-to guide -> `HowTo` schema with steps
   - FAQ -> `FAQPage` schema
   - Product page -> `Product` schema with `offers`, `review`
   - Comparison -> `Article` + individual `Product` schemas
   - Local content -> `LocalBusiness` schema
22. Generate JSON-LD snippet for implementation
23. Validate with Google's Rich Results Test

### Step 7: Readability Scoring
24. Target metrics:
   - Flesch Reading Ease: 60-70 (8th-10th grade level)
   - Average sentence length: 15-20 words
   - Average paragraph length: 3-4 sentences
   - Passive voice: < 10% of sentences
25. Flag sections that exceed complexity thresholds

### Step 8: Technical Checks
26. URL slug: Short, contains primary keyword, no stop words
27. Image optimization: Compressed, descriptive filenames, alt text
28. Mobile readability: No wide tables that require horizontal scroll
29. Page speed considerations: Recommend lazy loading for images below fold

## Content Scoring Tools: Reality Check

Surfer SEO, Clearscope, Frase analyze top-ranking pages and identify patterns. Ahrefs tested correlation with actual rankings:
- NeuronWriter + Ahrefs: Strongest (but modest) correlation
- Clearscope: ~0.30
- Surfer SEO: ~0.27

**The fundamental limitation:** These tools measure FIRST-STAGE retrieval signals only (lexical matching, BM25-style). After that:
1. RankEmbed adds candidates keyword matching missed
2. Mustang applies ~100+ signals including NavBoost, quality scores
3. DeepRank applies BERT-based understanding to final 20-30 results

**Bottom line:** Use SERP-derived topical completeness as a baseline check, not a ranking guarantee. The writing skill should ensure all SERP-common topics are covered, but the real ranking signals come from engagement, uniqueness, and authority.

## Tips

- **Featured snippet optimization**: For informational queries with existing snippets, format your answer in the snippet's format (paragraph, list, or table) within the first 300 words. Target 40-60 words for paragraph snippets.
- **PAA (People Also Ask) optimization**: Answer PAA questions within your content using the exact question as an H2, followed by a direct 40-60 word answer, then expanded content.
- **Keyword cannibalization check**: Before publishing, verify no existing page targets the same primary keyword. If it does, either merge or differentiate clearly.
- **Meta title CTR optimization**: Include brackets [2026], parentheses (with examples), or numbers (7 steps) -- these consistently increase CTR in A/B tests.

## Common Mistakes

1. **Optimizing before writing**: Writing around keywords produces unnatural content. Write naturally, then optimize.
2. **Keyword density > 2%**: Over-optimization triggers Google's keyword stuffing filter. 1-2% is the sweet spot.
3. **Exact-match keyword forcing**: "Best SEO content writing tools 2026 guide" as a sentence is not natural. Partial matches and variations work fine.
4. **Ignoring internal linking**: New content without internal links is an island. Google struggles to discover and contextualize it.
5. **Skipping structured data**: Free SERP real estate (rich snippets, FAQ dropdowns) is left on the table.

## Tools Used
- keyword search -- keyword targets and secondaries
- internal link data -- existing internal link structure
- page-level SEO data -- check for keyword cannibalization
- SERP feature detection -- identify featured snippet / PAA opportunities
- keyword cluster data -- ensure topic coverage completeness
- opportunity detection -- find keywords where pages rank 4-20 for quick wins

# Technique 05: SERP-Driven Writing

## What It Is

Writing content based on what ACTUALLY ranks in Google, not what theoretically should work. Every structural decision -- headings, length, format, keyword placement -- is derived from analysis of the current top 10 results for the target keyword.

## Why It Works

Google's ranking system is empirical. What ranks IS what Google rewards. SERP analysis reverse-engineers the patterns that currently win for a specific query. This approach beats theoretical best practices because it accounts for query-specific factors: intent, competition level, SERP features, and user expectations.

Kyle Roof's 400+ controlled experiments at PageOptimizer Pro proved that on-page signals alone can achieve rankings -- but only when those signals match what Google expects for that specific query.

## E-E-A-T: What Actually Matters

### The Reality (vs. What Google Says)

**Kyle Roof's tested position:** E-E-A-T elements will NOT help you rank, but they help you KEEP your rank once evaluated. He ranked a Lorem Ipsum page for "Rhinoplasty Plano" using on-page signals alone. E-E-A-T is defensive, not offensive.

**The priority:** Trust > Expertise > Experience > Authoritativeness

**Cyrus Shepard's 50-site case study (4,000+ websites across 2023 updates):**
- Winners: used first-person pronouns, demonstrated first-hand experience
- Losers: excessive ads (14.01 per page vs. 6.32 on winners), over-optimized anchor text
- 17 on-page features showed statistically significant correlations

### E-E-A-T Signals the Writing Skill Should Produce

**Experience signals (the new "E"):**
- First-person language: "in my testing," "when I tried," "our data shows"
- Specific case studies with real numbers and dates
- References to original photos, screenshots, or videos
- Practical tips that only come from doing the thing

**Expertise signals:**
- Depth of coverage (not just mentioning a topic, but explaining the WHY)
- Correct use of industry terminology
- Nuanced positions (not just repeating the obvious)
- Citing specific sources (not "experts say")

**Trust signals:**
- Author attribution with verifiable credentials
- Transparent methodology ("here's how we tested this")
- Honest limitations ("this approach doesn't work for X")
- Clear source attribution for claims

## Keyword Placement (Kyle Roof's Tested Hierarchy)

Based on 400+ controlled experiments:

**Group A (Highest weight):**
- Meta title -- the undisputed #1 on-page signal

**Group B:**
- URL path
- H1 heading
- Body content (~2% density, naturally distributed)
- H2 headings (primary keyword or close variation in 2-3 H2s)
- H3/H4 headings
- Anchor text of internal links TO this page

**Group C:**
- Bold text
- Italic text
- Image alt text

**Group D (Lowest):**
- Schema markup text
- HTML tags
- Open graph text

**Not indexed at all:**
- Meta description (doesn't affect rankings, but affects CTR)
- Meta keyword tag

**Kyle Roof's evolved position:** Stop counting exact-match keywords. Instead, ensure the page includes the FULL VOCABULARY of the topic. Topical completeness matters more than keyword density.

## Content Length by Intent

Ahrefs found near-zero correlation (0.04) between word count and ranking. The data by intent:

| Intent | Optimal Length | Rationale |
|--------|---------------|-----------|
| Informational (comprehensive guide) | 1,500-3,000+ | Needs to cover the full topic |
| Informational (specific question) | 800-1,500 | Answer directly, don't pad |
| Commercial investigation (comparison) | 1,200-2,500 | Depth in evaluation criteria |
| Commercial investigation (roundup) | 2,000-4,000 | Multiple items need coverage |
| Transactional | Minimal needed | No fluff; specs, pricing, CTA |
| Local service | 300-500 unique words | Must be genuinely unique per location |

**Key rule:** Match the SERP. If top 5 results average 2,000 words, write 2,000-2,500. If they average 800, writing 3,000 wastes effort and may signal wrong intent.

## Featured Snippet Optimization

**Format distribution:** Paragraph (70%), List, Table, Video

**Optimal answer length:** 40-50 words (293 characters average)

**Query-format mapping:**
- "How" queries -> ordered list snippets (46.91%)
- "Does/Are/Is/Can/Should/Will" queries -> paragraph snippets
- Comparison queries -> table snippets
- "What is" queries -> paragraph snippets (definition format)

**Implementation:** Place a 40-60 word direct answer immediately after the relevant H2. Format it as the snippet type Google expects for that query. Then expand below.

## People Also Ask Coverage

PAA prevalence spiked 34.7% on mobile and 37.5% on desktop (2024-2025). Your site does NOT need to rank #1 to appear in PAA.

**Strategy:**
- Include 3-5 PAA questions as H2 or H3 headings
- Answer each in 40-60 words immediately after the heading
- Then expand with detail below the direct answer
- Use the exact question phrasing when natural

## Algorithm Signals That Matter (From the Leak)

### NavBoost (Confirmed)
Uses rolling 13-month window of click data. Tracks:
- **goodClicks** -- positive engagement
- **badClicks** -- negative engagement
- **lastLongestClicks** -- where the user's search ENDED (strongest signal)

**Writing implication:** Content must immediately engage. The first screen of content determines whether users stay or pogo-stick back. Front-load value.

### Information Gain (Patent Granted June 2024)
Measures how much UNIQUE information a document provides beyond what already exists. NOT about length -- about saying something competitors don't.

**Writing implication:** Every article must contain at least one section that adds genuinely new information: original data, unique analysis, contrarian perspective, or specific case study.

### Site Authority (Confirmed in Leak)
Despite public denials, Google has a site-level authority metric. Homepage authority influences the entire domain.

**Writing implication:** Each piece of content contributes to site-level authority. Thin, low-quality content hurts the whole domain (HCU classifier operates site-wide).

## Content Scoring Tools: Reality Check

Surfer SEO, Clearscope, Frase analyze top-ranking pages and identify patterns. Ahrefs tested correlation with actual rankings:
- NeuronWriter + Ahrefs: Strongest (but modest) correlation
- Clearscope: ~0.30
- Surfer SEO: ~0.27

**The fundamental limitation:** These tools measure FIRST-STAGE retrieval signals only (lexical matching, BM25-style). After that:
1. RankEmbed adds candidates keyword matching missed
2. Mustang applies ~100+ signals including NavBoost, quality scores
3. DeepRank applies BERT-based understanding to final 20-30 results

**Bottom line:** Use SERP-derived topical completeness as a baseline check, not a ranking guarantee. The writing skill should ensure all SERP-common topics are covered, but the real ranking signals come from engagement, uniqueness, and authority.

## Content Freshness

### QDF (Query Deserves Freshness)
Activates for: breaking news, recurring events, topics with frequent updates. When active, newer content temporarily outranks stronger traditional signals.

### AI Citation Freshness
ChatGPT shows strongest recency preference: 76.4% of most-cited pages were updated within 30 days.

### Practical Rules
- Date-stamp all content with publication and last-updated dates
- Update high-traffic content every 3-6 months
- Don't just change the date -- Google detects fake freshness
- 51% of companies report updating content is more effective than creating new

## Tips

- Match SERP format exactly: if the top 5 all use comparison tables, include a comparison table
- Check SERP features before writing: featured snippets, PAA, video carousels each require different content structures
- Use SERP feature detection to identify which SERP features are available for target keywords
- Monitor ranking changes after publishing to validate SERP-driven decisions

## Common Mistakes

1. **Ignoring the SERP entirely** -- writing based on assumptions instead of what currently ranks
2. **Over-optimizing for content scoring tools** -- these only measure first-stage retrieval signals
3. **Writing longer than needed** -- if the SERP rewards 800-word answers, a 3,000-word article signals wrong intent
4. **Copying competitor structure without adding value** -- matching format is necessary but insufficient; you still need information gain

## Key Sources

- Kyle Roof -- PageOptimizer Pro: [nichepursuits.com](https://www.nichepursuits.com/kyle-roof-hcu/), [pageoptimizer.pro](https://www.pageoptimizer.pro/bestplacestoputakeyword)
- Cyrus Shepard -- [zyppy.com](https://zyppy.com/seo/google-update-case-study/)
- [Content Scoring: First Gate in Google's Pipeline (Search Engine Land)](https://searchengineland.com/content-scoring-tools-work-but-only-for-the-first-gate-in-googles-pipeline-469871)
- [Ahrefs Content Score Study](https://ahrefs.com/blog/seo-content-score-study/)
- [NavBoost Analysis (Hobo)](https://www.hobo-web.co.uk/navboost-how-google-uses-large-scale-user-interaction-data-to-rank-websites/)
- [Google Leak Analysis (Hobo)](https://www.hobo-web.co.uk/the-google-content-warehouse-leak-2024/)
- [Information Gain Patent (SEJ)](https://www.searchenginejournal.com/googles-information-gain-patent-for-ranking-web-pages/524464/)
- [Featured Snippet Statistics 2025](https://mycodelesswebsite.com/featured-snippet-statistics/)

# Technique 13: Structured Data & SERP Feature Capture

## What It Is
Using structured data (JSON-LD schema markup) and content formatting to capture SERP features -- featured snippets, FAQ dropdowns, How-To cards, and rich results that increase visibility, CTR, and NavBoost signals.

## Why It Works
SERP features occupy prime real estate above or within organic results. Pages with rich results get 20-40% higher CTR than plain blue links. Higher CTR feeds NavBoost (Google's click-based ranking signal), creating a compound ranking advantage.

Additionally, structured data is how content gets cited by AI Overviews and AI search engines. Without it, your content is less likely to be selected as a source for AI-generated answers.

## SERP Features by Content Type

| Content Type | Target SERP Features | Schema Types |
|-------------|---------------------|-------------|
| Blog post | Featured snippet, PAA | Article, FAQPage |
| How-to guide | HowTo card, Featured snippet | HowTo, FAQPage |
| Product page | Product rich result, Review stars | Product, Review, Offer |
| FAQ page | AI Overview citation, PAA | FAQPage |
| Comparison | Comparison table, Review stars | Article, Product, Review |
| Recipe | Recipe card | Recipe |
| Event | Event listing | Event |
| Local business | Local pack, Knowledge panel | LocalBusiness |

## Step-by-Step Process

### Step 1: Identify Feature Opportunities
1. Use SERP feature detection from to see which features appear for target keywords
2. Search the keyword manually -- what features are currently showing?
3. Note: featured snippets, PAA boxes, HowTo cards, video carousels
4. Check if any competitor has a SERP feature you could capture

### Step 2: Content Formatting for Featured Snippets
5. **Paragraph snippets** (most common):
   - Place a direct answer in 40-60 words immediately after an H2 matching the query
   - Start with the definition/answer, then expand
   - Example: "What is information gain?" -> H2 -> 45-word direct answer -> detailed explanation
6. **List snippets**:
   - Use numbered or bulleted lists with clear H2 header
   - 5-8 items is optimal
   - Each item should be a clear, scannable point
7. **Table snippets**:
   - Use HTML tables with clear column headers
   - 3-5 columns, 4-10 rows
   - Comparison data is ideal for table snippets

### Step 3: FAQ Schema Implementation
8. Identify 3-5 questions users ask about the topic (use PAA boxes as source)
9. Include each question as an H2 or H3
10. Answer directly in 40-100 words
11. Implement FAQPage schema with each Q&A pair:
   ```json
   {
     "@context": "https://schema.org",
     "@type": "FAQPage",
     "mainEntity": [{
       "@type": "Question",
       "name": "What is information gain in SEO?",
       "acceptedAnswer": {
         "@type": "Answer",
         "text": "Information gain in SEO measures how much new, unique information your content provides compared to other pages on the same topic. Google's patented Information Gain Score rewards pages that add original data, perspectives, or insights not found in competing results."
       }
     }]
   }
   ```

### Step 4: HowTo Schema
12. For tutorial/guide content, implement HowTo schema:
    - Each step as a separate `HowToStep`
    - Include `estimatedCost` and `totalTime` when applicable
    - Add `tool` and `supply` arrays
    - Include images per step when possible

### Step 5: Article/Product Schema
13. **Article schema** (for blog posts, guides):
    - `author` with name and URL
    - `datePublished` and `dateModified`
    - `image` (required for rich results)
    - `headline` matching H1
14. **Product schema** (for product/review pages):
    - `offers` with price, availability, currency
    - `review` with `ratingValue` and `bestRating`
    - `brand` entity

### Step 6: Validation
15. Test all structured data with Google's Rich Results Test
16. Monitor Search Console for structured data errors
17. Verify rich results appear in SERPs after indexing

## Tips

- **PAA as keyword research**: "People Also Ask" questions are literally Google telling you what related queries users search. Answer them in your content -> PAA placement and AI Overview citations.
- **Featured snippet "sniping"**: Find a keyword where the current featured snippet is mediocre. Write a better, more direct answer in the same format. Google swaps snippets regularly.
- **Don't over-schema**: Adding 50 FAQ items or marking up content that doesn't qualify for a schema type can trigger manual actions. Only mark up content that genuinely matches the schema type.
- **AI Overview sourcing**: Structured data helps your content get selected as a source for AI Overviews. Content with clear schema markup is easier for AI systems to parse and cite.

## Common Mistakes

1. **FAQ schema for non-questions**: Only use FAQPage schema for actual Q&A content. Using it for bullet points or regular content is a misuse.
2. **Missing required fields**: Schema without required fields (like `image` for Article) won't generate rich results
3. **Duplicate schema**: Having both FAQPage and HowTo on the same page can conflict. Choose the primary schema type.
4. **Not monitoring**: Structured data can break (especially after CMS updates). Monitor Search Console regularly.

## GEO (AI Citation) Impact

Structured data directly impacts AI citation rates:
- **FAQ schema has 3.2x higher citation rate** in AI Overviews
- **92% of AI Overview citations** come from top-10 ranking pages -- SERP features help you get there
- **Answer-first formatting** (40-60 word direct answers under H2s) produces +340% AI citations
- Content with clear schema markup is easier for AI systems to parse and cite

Write comprehensive content (for Google) with clear, extractable summary answers at the start of each section (for AI). The overlap between SERP feature optimization and AI citation optimization is ~80%.

## Tools Used
- SERP feature detection -- identify which features appear for target keywords
- keyword search -- find keywords with featured snippet opportunities
- page-level SEO data -- check current SERP feature status for existing pages

# Technique 03: Voice & Personality Injection

## What It Is

A system for injecting genuine human voice, personality, and specificity into AI-generated content. Removing AI patterns (Technique 02) is necessary but insufficient -- "sterile, voiceless writing is just as obvious as slop" (blader/humanizer). This technique adds the human qualities that make content engaging, trustworthy, and undetectable.

## Why It Works

AI content is detectable primarily because it regresses to the mean of all writing. A specific voice has specific word preferences, sentence rhythms, opinions, and reference domains that break the low-perplexity pattern. Voice injection also directly addresses Google's E-E-A-T "Experience" signal -- content with genuine personality signals a real author with real opinions.

Voice consistency is one of the strongest anti-AI-detection signals AND brand trust signals. Readers recognize when content "sounds different" -- it breaks trust. AI-generated content without voice injection sounds like every other AI-generated piece, which is detectable by both humans and algorithms.

Google's EEAT framework implicitly rewards voice consistency: content from a recognizable, consistent voice signals a real author with real expertise. Generic AI voice signals generic AI.

## Step-by-Step Process

### Step 1: Voice Sample Collection
1. Gather 3-5 real articles written by the brand's actual writer(s)
2. Include variety: a formal piece, a casual piece, a technical piece
3. Each sample should be 500-1,500 words minimum
4. These samples are the SINGLE most important input for quality output
5. **Capture negative examples too**: "We NEVER say 'leverage'" or "We NEVER use bullet points for everything" is just as important as positive guidelines

### Step 2: Voice Analysis
6. Analyze the samples for:
   - **Sentence length patterns**: Does the writer use short, punchy sentences? Long, flowing ones? A mix?
   - **Vocabulary level**: Technical jargon? Plain language? Industry-specific terms?
   - **Tone markers**: Humorous? Direct? Academic? Conversational? Provocative?
   - **Structural preferences**: Lists vs prose? Subheadings frequency? Paragraph length?
   - **Personal pronouns**: "We" (corporate)? "I" (personal)? "You" (direct address)?
   - **Opening style**: Start with a story? A statistic? A question? A bold statement?
   - **Transition style**: Formal ("Furthermore")? Conversational ("But here's the thing")?
   - **Quirks**: Any distinctive patterns? Parenthetical asides? Rhetorical questions? Specific metaphors?

### Step 3: Voice Profile Creation
7. Create a voice profile document with this structure:

```
VOICE PROFILE: [Brand Name]

Tone: [e.g., "Direct and slightly irreverent. Not afraid to call out bad advice."]
Register: [e.g., "Professional but conversational -- like explaining to a smart colleague"]
Perspective: [e.g., "First person plural ('we') for company content, first person singular ('I') for author-attributed pieces"]

DO: [e.g., "Use short paragraphs, ask rhetorical questions, include specific numbers"]
DON'T: [e.g., "Don't use corporate jargon, don't hedge everything, don't use emojis"]

Vocabulary preferences:
- Say "customers" not "clients"
- Say "build" not "develop"
- Say "test" not "validate"

Example phrases that capture the voice:
- "Here's the thing nobody talks about..."
- "We tested this. The results surprised us."
- "Skip the theory. Here's what actually works."
```

### Step 4: Voice Integration in Prompts
8. Include the voice profile in every writing prompt
9. Include 1-2 voice sample excerpts as few-shot examples
10. Instruct the model: "Write as if you ARE [writer name]. Match their sentence patterns, vocabulary, and tone exactly."
11. After generation, compare output against samples for voice drift

### Step 5: Voice Consistency Audit
12. Read the draft next to a real sample -- do they sound like the same person?
13. Check for AI voice bleeding through (formality spikes, generic transitions)
14. Verify vocabulary consistency (using the brand's specific terms)
15. Check opening and closing style matches the brand pattern
16. **Update samples regularly**: If the brand's voice evolves (it will), update the samples. Stale samples produce dated-sounding content.

## The 10 Voice Injection Techniques

### 1. Named Persona

**What:** Give the AI a specific writer identity with documented quirks, opinions, and speech patterns.

**Mechanism:** Forces the model away from the statistical mean. A specific voice has specific word preferences, sentence rhythms, and opinions that break predictability.

**Implementation:**
```
You are writing as [Name], a [role] who has been doing this for [X years].
They are known for: [specific trait, e.g., "blunt honesty about what doesn't work"]
They frequently reference: [domain, e.g., "cooking analogies to explain technical concepts"]
They dislike: [specific pet peeve, e.g., "vague advice without data"]
They would never say: [banned phrases specific to this persona]
Example of their voice: "[2-3 example sentences]"
```

**Not:** "Write in an engaging, conversational tone." This is too vague and produces generic "friendly" AI writing.

### 2. Burstiness Enforcement

**What:** Explicitly vary sentence length dramatically.

**Mechanism:** Directly addresses the low-burstiness detection signal. Humans naturally produce "bursts" of complexity followed by simple punchy lines.

**Implementation:**
```
Vary sentence length dramatically. Follow a long, complex sentence with
a short punchy one. Some paragraphs should be a single sentence. Others
should be five or six sentences. Never let three consecutive sentences
be similar in length. Mix 5-word fragments with 35-word complex sentences.
```

**Example transformation:**
- AI: "Email marketing remains one of the most effective digital marketing strategies available to businesses today. It provides a direct line of communication with potential customers. The return on investment is consistently high across industries."
- Human: "Email marketing works. Not in the vague, 'it's part of a balanced strategy' way consultants love. I mean: for every euro you spend, you get 36 back. That's DKG's 2025 data across 4,000 campaigns, and it's been consistent for five years running."

### 3. Specificity Over Generality

**What:** Replace every general claim with a specific example, number, name, date, or anecdote.

**Mechanism:** Specific details are high-entropy tokens that break predictability. They also signal E-E-A-T (Experience, Expertise) to Google.

**Implementation:**
```
Never write a general claim without a specific example. Instead of
"many companies struggle with this," write "When Coolblue's marketing
team hit this wall in Q3 2025, they..." Every abstraction must be
grounded in a concrete particular. If you don't have a real example,
make the specificity about the scenario: exact numbers, exact steps,
exact consequences.
```

**The "30% Rule":** At least 30% of every page should contain details no generic AI could produce: proprietary data, original analysis, first-hand testing results, specific case studies, or named examples.

### 4. Contrarian/Opinionated Writing

**What:** Take clear stances. Disagree with conventional wisdom when appropriate.

**Mechanism:** AI is trained via RLHF to be balanced and inoffensive. Strong opinions are high-entropy by definition. They also drive engagement (comments, shares, backlinks) which are ranking signals via NavBoost.

**Implementation:**
```
Take a strong position on every topic. Don't hedge with "it depends"
unless you genuinely mean it. If something is bad, say it's bad. If
the conventional wisdom is wrong, say so and explain why with evidence.
You are allowed to be wrong -- that's what makes you interesting.

When appropriate, include phrases like:
- "Here's what most guides get wrong about this..."
- "I've tested this and the data says the opposite..."
- "Everyone recommends X but in my experience Y works better because..."
```

### 5. Cultural References & Temporal Anchoring

**What:** Reference specific cultural moments, current events, or shared experiences.

**Mechanism:** Real cultural references signal an author who lives in the world. They create "information gain" that generic content cannot replicate. They anchor content in a specific time and place.

**Implementation:**
```
Reference specific cultural touchstones where relevant -- recent news,
industry events, widely-known examples. Not as decoration but as genuine
analogies that illuminate your point. Use references your target audience
would know. For Dutch audiences: reference Prinsjesdag, Black Friday NL,
specific Dutch companies, known media personalities, etc.
```

### 6. Show Don't Tell

**What:** Demonstrate claims through mini-narratives and scenarios instead of stating them.

**Mechanism:** "Telling" produces low-perplexity text ("Page speed affects rankings"). "Showing" produces high-perplexity, engaging text ("You click a search result. Three seconds pass. Still loading. You hit back. Google noticed.").

**Implementation:**
```
Never just state a fact. Show it through a brief scenario. Instead of
"page speed affects rankings," write a mini-narrative: "You click a
search result. Three seconds pass. The page is still loading. You hit
back and click the next result. Google tracked every millisecond of that
interaction -- and it just cost the slow site a ranking position."
```

### 7. Read-Aloud Test (Paul Graham Method)

**What:** Every sentence should sound natural when spoken aloud to a friend.

**Mechanism:** Paul Graham claims this puts you "ahead of 95% of writers." Conversational language has naturally higher burstiness, uses contractions, and avoids the formal register AI defaults to.

**Implementation:**
```
Write as if talking to a smart friend over coffee. Use contractions
(don't, isn't, won't). Start sentences with "And" or "But" when natural.
Drop unnecessary qualifiers. If you wouldn't say it out loud, rewrite it.

Test: read every paragraph back. If it sounds like a textbook, rewrite
it until it sounds like a person.
```

### 8. Strategic Imperfection

**What:** Include deliberate human markers -- fragments, parenthetical asides, self-corrections.

**Mechanism:** AI produces unnaturally perfect grammar and structure. Human writing has quirks that increase both burstiness and perplexity.

**Implementation:**
```
You are allowed to be imperfect. Use sentence fragments for emphasis.
Start sentences with conjunctions. Use parenthetical asides (like this)
when you have a tangential thought. Occasionally correct yourself
mid-paragraph ("well, actually..." or "that's not quite right --").
Real writers do this naturally.
```

### 9. Register Shifts

**What:** Shift between formal and casual register within the same piece.

**Mechanism:** AI maintains a uniform register. Humans naturally shift -- a technical explanation followed by a casual aside, a formal point followed by a joke.

**Implementation:**
```
Shift between registers naturally. After a technical paragraph,
drop in something casual. After making a serious point, add a human
aside. The contrast is what makes writing feel alive.

Example: "The algorithm processes 12 trillion signals per query.
(Yes, trillion. With a T. And it does this in under 0.5 seconds.
Makes you feel productive, doesn't it?)"
```

### 10. The Gary Halbert "One Person" Principle

**What:** Write to one specific person, not an audience.

**Mechanism:** "Write your copy as if you're having a conversation with one reader." Using "I" and "you" creates intimacy. Specific, seemingly irrelevant details build trust.

**Implementation:**
```
Write to one specific person who already cares about this topic.
Use "you" and "I" (or "we" for brand content). Include specific,
concrete details even when they seem tangential -- they build trust.
Don't persuade; serve. Imagine the reader is sitting across from
you at a table, asking a real question.
```

## Voice Document Structure

The writing skill should load a Voice Document (from your saved business context or user-provided) with this structure:

```markdown
## Brand Voice Profile

### Identity
- Writer role/title:
- Years of experience:
- Known for:
- Industry perspective:

### Voice Characteristics
- Formality level: [casual / conversational / professional / formal]
- Uses contractions: [yes / no]
- Uses humor: [yes / sparingly / no]
- Takes strong positions: [yes / sometimes / rarely]
- Sentence length tendency: [short and punchy / mixed / flowing]

### Reference Domains
- Analogies drawn from: [e.g., cooking, sports, construction]
- Cultural references: [e.g., Dutch media, tech industry, specific era]

### Pet Peeves (things this voice NEVER does)
- [e.g., "never says 'at the end of the day'"]
- [e.g., "never hedges when they have data"]
- [e.g., "never uses business jargon"]

### Example Paragraphs (2-3 samples of the target voice)
[Actual writing samples that demonstrate the voice]

### Audience
- Writing for: [specific person description]
- Their knowledge level: [beginner / intermediate / expert]
- What they care about: [specific concerns]
```

**Voice varies by channel**: A brand's blog voice differs from their product page voice differs from their social media voice. Capture voice per content type if possible. The voice document should indicate which register applies to which content format.

## Copywriting Frameworks by Content Type

| Content Type | Best Framework | Why |
|-------------|---------------|-----|
| Service pages, landing pages | PAS (Problem-Agitate-Solve) | Forces emotional engagement |
| Commercial content, reviews | AIDA (Attention-Interest-Desire-Action) | Drives decision-making |
| Brand content, about pages | StoryBrand | Positions reader as hero |
| News, definitions, how-tos | Inverted Pyramid | Answer first, expand second |
| Email-style, newsletters | Gary Halbert "One Person" | Maximum intimacy |
| Data pages, comparisons | David Ogilvy "Facts First" | Specifics over superlatives |

## Tips

- **The "swap test"**: Put a paragraph from the AI draft next to a paragraph from a real sample. If a colleague could tell which is which, the voice is not matched yet.
- **Capture negative examples too**: "We NEVER say 'leverage'" or "We NEVER use bullet points for everything" is just as important as positive voice guidelines. What the brand avoids defines the voice as much as what it does.
- **Voice varies by channel**: A brand's blog voice differs from their product page voice differs from their social media voice. Capture voice per content type.
- **Update samples regularly**: If the brand's voice evolves (it will), update the samples. Stale samples produce dated-sounding content.
- **The "30% Rule"**: At least 30% of every page should contain details no generic AI could produce.
- **Voice before anti-slop**: Apply voice injection BEFORE running anti-slop audits (Technique 02). Removing patterns without adding personality produces sterile text.

## Common Mistakes

1. **"Write engagingly"** -- too vague, produces generic friendly AI. Specify exact voice characteristics.
2. **Using generic voice descriptions**: "Professional and friendly" describes 90% of brands. Be specific: "Slightly sarcastic, data-obsessed, uses 'actually' a lot"
3. **Adding personality to every sentence** -- exhausting; let some sentences be plain
4. **Forcing humor** -- bad jokes are worse than no jokes
5. **Overusing fragments** -- occasional fragments add punch; constant fragments are a new AI pattern
6. **Fake specificity** -- inventing plausible-sounding but false examples is worse than being general
7. **Voice inconsistency** -- voice should be consistent throughout; load voice document before EVERY section
8. **Over-constraining**: Too many rules make writing stiff. Focus on 5-7 key voice markers, not 50
9. **Ignoring audience awareness**: The writer's voice adjusts to audience. A piece for CTOs sounds different than a piece for marketers, even in the same brand voice
10. **One voice for all content**: Product descriptions, thought leadership, and technical docs require different registers within the same brand voice

## Key Sources

- [Paul Graham: Write Like You Talk](https://paulgraham.com/talk.html)
- [Gary Halbert: The Boron Letters (Enchanting Marketing)](https://www.enchantingmarketing.com/gary-halbert-boron-letters/)
- [David Ogilvy's 7 Principles (Cult Method)](https://cultmethod.com/articles/ogilvys-principles/)
- [StoryBrand Framework](https://storybrand.com/)
- [blader/humanizer](https://github.com/blader/humanizer) -- "Sterile, voiceless writing is just as obvious as slop"
- [haowjy/creative-writing-skills](https://github.com/haowjy/creative-writing-skills) -- Style learning from samples
- [viktorbezdek/definitive-llm-writing-style-guide](https://github.com/viktorbezdek/definitive-llm-writing-style-guide)

# Technique 15: Writing Pipeline Architecture

## What It Is

A multi-phase, multi-agent pipeline for producing high-quality SEO content. Based on analysis of 15+ agentic writing systems, this architecture represents the consensus best approach: specialized phases with quality gates, not a single monolithic generation.

## Why It Works

Research consistently shows that multi-agent pipelines outperform single-prompt generation:
- Section-by-section generation maintains quality over long articles
- Specialized agents (writer vs. critic) produce better results than a generalist
- One to two revision cycles capture 95% of improvement (Self-Refine research)
- Quality gates prevent bad content from being output

Human-written articles still generate 5.44x more traffic and hold attention 41% longer than pure AI. The pipeline must close this gap through specialization.

Single-prompt content generation fails because one prompt can't simultaneously optimize for research depth, writing quality, SEO mechanics, anti-detection, and fact accuracy. Each agent focuses on ONE thing well, and each stage builds on verified output from the previous stage.

## The 7-Agent Model

The pipeline uses seven specialized agents, each with a distinct role and optimal configuration:

| Agent | Role | Temperature | Tools |
|-------|------|-------------|---------------|
| Researcher | Analyze topic landscape, SERP competition, information gaps | Low (0.2-0.3) | keyword search, competitor data, SERP feature detection |
| Outliner | Create SERP-optimized content structure | Low-Medium (0.3-0.4) | opportunity detection, keyword cluster data |
| Writer | Generate first draft with anti-detection built in | Medium (0.5-0.7) | page content rendering (competitor analysis) |
| Humanizer | Audit and fix surviving AI patterns | Low (0.2-0.3) | None |
| Fact-Checker | Verify all claims, statistics, attributions | Low (0.1-0.2) | None |
| SEO Optimizer | Keyword placement, meta elements, schema, linking | Low (0.2-0.3) | internal link data, SERP feature detection |
| Quality Reviewer | Final holistic review and scoring | Low (0.2-0.3) | None |

## Implementation Architecture

```
+--------------+     +----------+     +--------+     +-----------+
| RESEARCHER   | --> | OUTLINER | --> | WRITER | --> | HUMANIZER |
+--------------+     +----------+     +--------+     +-----------+
                                                          |
+--------------+     +----------+     +----------------+  |
| QUALITY      | <-- | SEO      | <-- | FACT-CHECKER   | <+
| REVIEWER     |     | OPTIMIZER|     |                |
+--------------+     +----------+     +----------------+
      |
      v
  [FINAL OUTPUT]
  - Article (markdown)
  - Meta elements
  - Quality scorecard
  - SEO checklist
  - Schema JSON-LD
  - AI-detection risk score
```

## User Approval Workflow

The pipeline is NOT fully autonomous. Human checkpoints are mandatory:

1. User invokes "write content" -> triggers RESEARCHER
2. RESEARCHER output presented to user for review/approval
3. OUTLINER generates structure -> user approves or modifies
4. Pipeline PAUSES to collect human input (experiences, data, opinions)
5. WRITER generates with all context -> HUMANIZER audits
6. FACT-CHECKER flags issues -> user verifies first-party claims
7. SEO OPTIMIZER + QUALITY REVIEWER finalize
8. Final output with scorecard presented to user

At every human checkpoint, the user can modify, reject, or redirect. This is not optional overhead -- it is the mechanism that prevents high-quality slop.

## Pipeline Architecture (Detailed Phases)

### Phase 1: Preparation (Context Loading)

**Inputs loaded:**
1. Content Brief (from Content Brief Generator)
2. Voice Document (from your saved business context or user-provided)
3. Content Type Template (selected based on intent/type from brief)
4. Anti-Slop Rules (banned phrases, structural patterns)
5. Internal Link Map (from internal link data)
6. Competitor Content Analysis (from page content rendering on top competitor URLs)

**Critical:** All context must be loaded BEFORE writing begins. Loading mid-article breaks voice consistency.

**The RESEARCHER agent performs:**
1. Queries tools: keyword search, competitor data, SERP feature detection
2. Analyzes top 10 SERP results for the target keyword
3. Maps competitor content coverage (topics covered, depth, unique angles)
4. Identifies information gaps (topics nobody covers)
5. Notes SERP features present (PAA, featured snippets, video carousels)
6. Recommends target word count based on competitor analysis

**Output:** Research brief with SERP gap analysis, competitor coverage map, information gain opportunities.

### Phase 2: Outline Generation

**The OUTLINER agent generates a detailed outline from the content brief:**
1. SERP-derived H2/H3 structure (matching what ranks)
2. Keywords mapped to specific sections
3. Internal links planned for specific paragraphs
4. Data points / examples planned per section
5. Featured snippet target identified (format + section)
6. PAA questions assigned to sections
7. Word count allocation per section
8. "Human input required" points marked (case studies, data, opinions)

**Human approval gate:** Present outline for approval before proceeding. In automated mode, validate against brief structure. The user can add, remove, or reorder sections.

### Phase 3: Section-by-Section Writing

**Why section-by-section:** For articles over 1,500 words, full-article generation leads to quality drift, repetition, and context distraction. Section-by-section maintains focused context.

**The WRITER agent processes each section:**
1. Load section brief (H2/H3 structure, assigned keywords, planned links, planned examples)
2. Load voice document (reload for each section to maintain consistency)
3. Apply content type template rules
4. Inject human-provided content at designated points
5. Write the section with anti-detection built in (varied sentence length 3-40 words, Tier 1 word avoidance)
6. Place internal links at natural anchor points
7. Update scratchpad with key points covered (prevents repetition)
8. Compress section summary for context management

**Context management:** After each section, create a summary of what has been covered. Use the summary (not full text) as context for subsequent sections. This prevents the model from losing focus as context grows.

**Key finding from Anthropic:** Model performance drops around 32,000 tokens even with million-token windows due to context distraction. Use "just-in-time context" -- lightweight identifiers that dynamically load data at runtime, not monolithic context blocks.

**Section compression strategy:** Each completed section is reduced to a 2-3 sentence summary containing: key points made, data cited, keywords used, links placed. The full text is stored separately. The writer for the next section sees only these summaries plus the current section's brief.

### Phase 4: Anti-Slop Editing

**Separate from writing.** The writer agent should be "naive" about quality rules (DonAldente pattern) -- its job is to write naturally. The editor/critic is separate and ruthless.

**The HUMANIZER agent runs two passes:**

**Pass 1: Detection**
- Scan for Tier 1 banned phrases (always remove)
- Scan for Tier 2 clustering (flag if 3+)
- Scan for Tier 3 density (flag clusters)
- Check structural patterns (uniform sentence length, rule-of-three, etc.)
- Check burstiness (sentence length standard deviation must be > 5 words per paragraph)
- Check rhythm uniformity (flags sections where all sentences are 15-25 words)
- Score total points

**Rewrite:** For each flagged section, rewrite preserving meaning while eliminating the pattern. Do not just delete -- replace with human-sounding alternatives.

**Pass 2: Surviving Pattern Check**
- Re-scan the full text after Pass 1 rewrites
- First-pass rewrites retain 15-20% of AI patterns
- Specifically check for recycled transitions, lingering inflation, new uniformity

**Horoscope Test:** For each paragraph, ask: "Could anyone have written this, for anyone?" If yes, flag for specificity injection.

**Critical threshold:** If the Humanizer has to rewrite more than 20% of the content, the Writer prompt needs improvement. The Writer should produce mostly clean output; the Humanizer is a quality gate, not the primary fix.

### Phase 5: Fact-Checking

**The FACT-CHECKER agent verifies:**
1. All factual claims in the content
2. Statistics against named sources
3. Fabricated or unverifiable claims (flag for removal)
4. All "experts say" have named experts
5. URLs and references are real and current
6. Claims that need human verification (first-party data)

**Output:** Fact-checked draft + verification report with confidence levels per claim.

### Phase 6: SEO Optimization

**The SEO OPTIMIZER agent handles:**

**Keyword checks:**
- Primary keyword in meta title, H1, first paragraph, 2-3 H2s
- ~2% body density (naturally distributed, not forced)
- Keyword variations in H3s and image alt text

**Internal link validation:**
- Every internal link must point to a real, existing page
- Validate against internal link data
- Use descriptive anchor text (not "click here")
- 3-5 contextual links per 1,000 words

**Meta generation:**
- Meta title: primary keyword + benefit/hook, under 60 characters
- Meta description: value proposition + CTA, under 155 characters
- 5 title variations, 3 description variations

**Schema markup:**
- Generate JSON-LD based on content type (HowTo, FAQ, Product, etc.)
- Include FAQPage schema for all content with FAQ sections
- Validate schema structure

**Featured snippet targeting:**
- Verify the target snippet section has a 40-60 word direct answer
- Format matches expected snippet type (paragraph/list/table)

### Phase 7: Quality Scoring

**The QUALITY REVIEWER agent performs a holistic review and scores on a 100-point composite:**

| Category | Points | Components |
|----------|--------|------------|
| Content Quality | 30 | Topical completeness (10), Depth/examples per section (10), Originality (10) |
| SEO Optimization | 25 | Keyword placement (8), Internal links verified (5), Meta quality (4), Schema (4), Snippet targeting (4) |
| E-E-A-T Signals | 15 | First-person experience (5), Specific data/examples (5), Source attribution (5) |
| Anti-Slop Score | 15 | Zero Tier 1 phrases (5), Burstiness (3), Voice consistency (3), Horoscope Test (4) |
| AI Citation Ready | 15 | Answer-first formatting (5), FAQ schema (3), Claim-evidence pairs (4), Entity richness (3) |

**The reviewer also checks:**
- Flow and logical progression across the full article
- Voice consistency throughout (not just per-section)
- Information gain elements are present and effective
- EEAT signals are woven in (not bolted on)

**Quality gates:**
- Score < 60: Full rewrite (re-enter Phase 3)
- Score 60-69: Auto-revise weak sections (re-enter Phase 4 for flagged sections)
- Score 70-84: Flag issues for human review, output with annotations
- Score 85+: Ready for publication

**Auto-revision:** Target the weakest scoring category first. If Anti-Slop scored low, run additional editing passes. If E-E-A-T scored low, inject experience signals. Maximum 2 revision cycles (research shows quality degrades beyond this).

## Key Design Decisions

### Sequential, Not Parallel
Each agent needs the previous agent's output. The Writer can't write without the Outline. The Humanizer can't audit without the draft. Parallel execution produces worse results because agents lack context. CrewAI and LangGraph implementations have confirmed that sequential pipelines produce higher quality.

### Context Passing Is Full, Not Summarized
Each agent receives the FULL output of the previous agent, not a summary. Summaries lose critical nuances (specific data points, exact quotes, structural decisions). The exception is section-to-section context within the Writer phase, where compression prevents context distraction.

### Anti-Detection Is Built In, Not Bolted On
The Writer already follows anti-detection guidelines. The Humanizer is a QUALITY GATE, not the primary fix. If the Humanizer has to rewrite more than 20% of the content, the Writer prompt needs improvement.

### Human Input Is Required, Not Optional
The pipeline HALTS at the Writer stage to request human input (case studies, data, opinions). Without this input, the pipeline produces content with zero information gain -- which defeats the entire purpose.

## Output Format

```markdown
---
title: [Generated meta title]
description: [Generated meta description]
keywords: [Primary keyword, secondary keywords]
schema_type: [HowTo | FAQ | Article | Product | etc.]
word_count: [Actual word count]
quality_score: [0-100]
tarsnippet data: [paragraph | list | table | none]
---

# [H1 with primary keyword]

[Article content in Markdown]

---
<!-- Internal links used: [list of URLs] -->
<!-- Schema JSON-LD: [embedded or separate file] -->
<!-- Quality breakdown: Content X/30, SEO X/25, E-E-A-T X/15, Anti-Slop X/15, GEO X/15 -->
```

## Systems Studied

| System | Key Innovation | Adopted |
|--------|---------------|---------|
| SEO Machine | Editor agent targeting AI patterns | Yes -- Phase 4 |
| DonAldente anti-slop | Naive drafter + ruthless critic | Yes -- Phase 3 vs 4 separation |
| claude-blog | Dual Google + AI citation scoring | Yes -- Phase 7 scoring |
| Agentic-SEO-Skill | Confidence labeling on claims | Adapted -- source attribution |
| AutoGen Reflection | JSON-summarized multi-reviewer | Adapted -- structured quality scoring |
| Self-Refine research | 1-2 revision cycles optimal | Yes -- max 2 revision cycles |
| Anthropic context engineering | Just-in-time context loading | Yes -- section compression |

## Common Mistakes

1. **Running all agents with the same LLM temperature**: Researcher needs low temperature (factual), Writer needs medium (creative), Humanizer needs low (precise editing). See the agent table above for recommended values.
2. **Skipping the Fact-Checker**: AI confidently fabricates statistics. One fake stat destroys the entire article's credibility.
3. **Making the pipeline fully autonomous**: Without human input, the pipeline produces high-quality AI slop. Human input at the Writer stage is the critical differentiator.
4. **Post-processing instead of pre-processing**: Anti-detection works 5x better when built into the writing prompt vs applied after.
5. **Monolithic context loading**: Loading all context at once causes performance degradation past 32K tokens. Use section compression and just-in-time context instead.
6. **Over-revising**: More than 2 revision cycles produces diminishing returns and can actually degrade quality. Fix the upstream prompt instead.

## Key Sources

- [TheCraigHewitt/seomachine](https://github.com/TheCraigHewitt/seomachine)
- [DonAldente-AI/anti-slop-system](https://github.com/DonAldente-AI/anti-slop-system)
- [AgriciDaniel/claude-blog](https://github.com/AgriciDaniel/claude-blog)
- [Self-Refine: Iterative Refinement (Learn Prompting)](https://learnprompting.org/docs/advanced/self_criticism/self_refine)
- [Effective Context Engineering (Anthropic)](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents)
- [Building Effective Agents (Anthropic)](https://www.anthropic.com/research/building-effective-agents)