Gemini Prompt

Grounded competitive analysis with Gemini

A research prompt that forces citations and a "what I couldn't verify" section — no confident guessing.

Ada · 0 points · 19 217

#writing #research

## The prompt You are a research analyst. I'll name a company or product. Produce a competitive brief. **Rules:** - Every factual claim must carry a source link. If you can't source it, it goes under **Unverified** — never stated as fact in the body. - Prefer primary sources: the company's own site, docs, changelog, pricing page, filings, job listings. Blogs and news are secondary; listicles and comparison-SEO pages are not evidence. - Put a **date** on everything, and flag anything that may be outdated with the date you found it. - If three sources trace back to the same announcement, that's **one** source. Say so rather than presenting it as corroboration. - Distinguish what a source *states* from what you *infer*. Label inferences. - Where you're reading marketing copy, say so. A vendor's claim about a vendor is a data point about positioning, not about capability. **Structure:** 1. **What it is** — one paragraph. What the product actually does, in the words a user would use, not the words the homepage uses. 2. **Who it's for** — segments, and the specific job it gets hired for. Name the job, not the demographic. 3. **Positioning** — how it differs from the top 2 alternatives (name them). Be concrete: what can it do that they can't, and what can they do that it can't? "Better UX" is not a difference. 4. **Pricing** — tiers, what gates each one, and where the pricing model creates friction (per-seat vs usage, annual lock-in, credit systems). With source and date. 5. **Signals** — recent moves with dates and sources: launches, pricing changes, hires, funding, deprecations, notable outages, docs restructures. Include what they've *stopped* doing; a quietly removed feature is a signal. 6. **Weaknesses** — the recurring complaints, sourced from real user reports rather than competitor marketing. Say how widespread each appears to be. 7. **What would change my read** — the single fact that, if different, would flip the conclusion. 8. **Unverified** — claims I should double-check myself, with what you'd need to confirm each. Target: `{company or product}` ## Why it's shaped this way **"Unverified" as a required section** is the whole design. Given a target, a model will produce a complete-looking brief regardless of whether it found anything, because a brief with gaps *reads* like a failure. Making the gaps a mandatory output section changes the incentive: admitting you couldn't source a pricing number becomes filling in a field rather than falling short. **Independence over count.** The failure mode in competitive research isn't fabrication — it's laundering. One press release, echoed by twelve outlets, presented as a well-supported consensus. **"What they've stopped doing."** Deprecations, quietly removed pricing tiers, and docs pages that now 404 are among the most informative signals available, and no summary surfaces them unless asked. **Weaknesses sourced from users, not competitors.** Otherwise you get the rival's marketing back, laundered through your own brief. ## How to use it well - **Bound the time window.** "Signals from the last 12 months, and flag anything older you rely on." Competitive briefs rot fastest in section 5. - **Name your decision.** "I'm evaluating this to replace {X} for a 40-person team" produces a materially different and more useful brief than a generic profile — it tells the model which differences matter. - **Ask for the pricing page verbatim.** Summarized pricing is wrong surprisingly often. Ask it to quote the tiers and link the page. - **Then run the adversary pass:** *"What's the strongest case that this product is the wrong choice for us? Now the strongest case that it's the right one."* Both are more informative than the balanced version. ## Follow-ups - *"What does their careers page suggest they're building next?"* — hiring is a leading indicator and it's public. - *"What do their docs say that their marketing doesn't?"* — the gap between the two is where the real product lives. - *"Who has publicly migrated away from this, and what did they say?"* - *"What would I need to build myself that this doesn't cover?"* ## Failure modes to watch - **Pricing invented from the shape of the market.** Very common, very confident, and the number will be plausible. Always check the link. - **Stale feature claims.** "Doesn't support X" is often true of a version from a year ago. Check the changelog date. - **Competitors chosen for name recognition** rather than for actually competing for the same job. If the named alternatives aren't ones you'd genuinely evaluate, say so and re-ask. - **A hallucinated URL.** A 404 is a strong tell that the whole claim was generated rather than retrieved — treat everything near it with suspicion. - **Balance as a substitute for judgement.** A brief that carefully weighs both sides and concludes nothing has failed. Section 7 exists to force a position. ## Ethics note This is public-information research. Everything above is sourced from things the company published: its site, docs, pricing, filings, job posts, and what users said in public. If a technique requires misrepresenting who you are or accessing something not offered publicly, it isn't in scope for this prompt.

The prompt

You are a research analyst. I'll name a company or product. Produce a competitive brief.

Rules:

Every factual claim must carry a source link. If you can't source it, it goes under Unverified — never stated as fact in the body.
Prefer primary sources: the company's own site, docs, changelog, pricing page, filings, job listings. Blogs and news are secondary; listicles and comparison-SEO pages are not evidence.
Put a date on everything, and flag anything that may be outdated with the date you found it.
If three sources trace back to the same announcement, that's one source. Say so rather than presenting it as corroboration.
Distinguish what a source states from what you infer. Label inferences.
Where you're reading marketing copy, say so. A vendor's claim about a vendor is a data point about positioning, not about capability.

Structure:

What it is — one paragraph. What the product actually does, in the words a user would use, not the words the homepage uses.
Who it's for — segments, and the specific job it gets hired for. Name the job, not the demographic.
Positioning — how it differs from the top 2 alternatives (name them). Be concrete: what can it do that they can't, and what can they do that it can't? "Better UX" is not a difference.
Pricing — tiers, what gates each one, and where the pricing model creates friction (per-seat vs usage, annual lock-in, credit systems). With source and date.
Signals — recent moves with dates and sources: launches, pricing changes, hires, funding, deprecations, notable outages, docs restructures. Include what they've stopped doing; a quietly removed feature is a signal.
Weaknesses — the recurring complaints, sourced from real user reports rather than competitor marketing. Say how widespread each appears to be.
What would change my read — the single fact that, if different, would flip the conclusion.
Unverified — claims I should double-check myself, with what you'd need to confirm each.

Target: {company or product}

Why it's shaped this way

"Unverified" as a required section is the whole design. Given a target, a model will produce a complete-looking brief regardless of whether it found anything, because a brief with gaps reads like a failure. Making the gaps a mandatory output section changes the incentive: admitting you couldn't source a pricing number becomes filling in a field rather than falling short.

Independence over count. The failure mode in competitive research isn't fabrication — it's laundering. One press release, echoed by twelve outlets, presented as a well-supported consensus.

"What they've stopped doing." Deprecations, quietly removed pricing tiers, and docs pages that now 404 are among the most informative signals available, and no summary surfaces them unless asked.

Weaknesses sourced from users, not competitors. Otherwise you get the rival's marketing back, laundered through your own brief.

How to use it well

Bound the time window. "Signals from the last 12 months, and flag anything older you rely on." Competitive briefs rot fastest in section 5.
Name your decision. "I'm evaluating this to replace for a 40-person team" produces a materially different and more useful brief than a generic profile — it tells the model which differences matter.
Ask for the pricing page verbatim. Summarized pricing is wrong surprisingly often. Ask it to quote the tiers and link the page.
Then run the adversary pass: "What's the strongest case that this product is the wrong choice for us? Now the strongest case that it's the right one." Both are more informative than the balanced version.

Follow-ups

"What does their careers page suggest they're building next?" — hiring is a leading indicator and it's public.
"What do their docs say that their marketing doesn't?" — the gap between the two is where the real product lives.
"Who has publicly migrated away from this, and what did they say?"
"What would I need to build myself that this doesn't cover?"

Failure modes to watch

Pricing invented from the shape of the market. Very common, very confident, and the number will be plausible. Always check the link.
Stale feature claims. "Doesn't support X" is often true of a version from a year ago. Check the changelog date.
Competitors chosen for name recognition rather than for actually competing for the same job. If the named alternatives aren't ones you'd genuinely evaluate, say so and re-ask.
A hallucinated URL. A 404 is a strong tell that the whole claim was generated rather than retrieved — treat everything near it with suspicion.
Balance as a substitute for judgement. A brief that carefully weighs both sides and concludes nothing has failed. Section 7 exists to force a position.

Ethics note

This is public-information research. Everything above is sourced from things the company published: its site, docs, pricing, filings, job posts, and what users said in public. If a technique requires misrepresenting who you are or accessing something not offered publicly, it isn't in scope for this prompt.