ChatGPT is the default general-purpose assistant for most of the world, and the breadth is the point: writing, summarizing, coding, brainstorming, data analysis, image generation, and web-grounded research all live behind one text box. Where specialist tools beat it on any single axis, ChatGPT wins on not making you decide which tool to open.
The features that change how you use it
- Projects — a folder with its own files, custom instructions, and memory scope. The single biggest upgrade for repeat work: stop re-explaining your context in every new chat.
- Memory — persistent facts carried between conversations. Worth curating deliberately; worth clearing when you switch domains.
- Deep research — a long-running agent that browses, reads, and returns a cited report. Minutes, not seconds.
- Agent mode — it operates a virtual computer and browser to actually do multi-step tasks, not just describe them.
- Canvas — a side-by-side document/code editor with targeted edits, so revision stops meaning "regenerate the whole thing".
- Advanced data analysis — uploads a CSV or spreadsheet, writes and runs Python against it, and returns charts plus the code it used.
- Connectors — read from Google Drive, SharePoint, GitHub and similar so answers can span your own documents.
- Custom GPTs — a packaged assistant with fixed instructions, files, and actions, shareable with a link or a team.
- Codex — the coding agent, available from ChatGPT, the CLI, and the IDE extension, sharing the same account and task history.
Pricing shape
A free tier; Go as a low-cost entry plan (about $8/month in the US); Plus at the familiar ~$20/month with expanded access to the reasoning models, deep research, agent mode, and Codex; Pro an order of magnitude above that for the largest model, maximum memory and context, and early feature access; and per-seat Business and Enterprise plans that add admin controls, SSO, and a no-training-on-your-data default. OpenAI has said ads will be tested on the free and Go tiers while Plus, Pro, Business, and Enterprise stay ad-free. Current numbers: openai.com/chatgpt/pricing.
Best for
Generalists and small teams who want one subscription that covers most knowledge work — and anyone who benefits from the enormous surrounding ecosystem of prompts, tutorials, and integrations that comes with being the most-used assistant.
Where it struggles
- Model sprawl. Picking between the fast model, the thinking model, and the pro model is a real tax on new users, and the wrong pick silently costs you quality or minutes.
- Long-context fidelity. In a very long thread it will start contradicting things it said earlier. Start a fresh chat with a summary rather than pushing on.
- Codebase depth. It sees what you paste or connect; a repo-indexed tool sees the repository. For serious code work, reach for Codex or an editor-native agent.
- Sycophancy. Ask it to critique, and it grades gently. "Argue the strongest case against this" gets you an actual review.
Getting more out of it
- Put the standing context in a Project, not in each prompt.
- State the output contract up front — audience, length, format, what to omit.
- For anything factual and load-bearing, ask for sources and then open them.
- Use Canvas for iteration; use a new chat when a thread has gone stale.
- Turn a prompt you've reused three times into a Custom GPT — or publish it here.
Worth knowing
Consumer plans may use your conversations for training unless you turn that off; Business and Enterprise do not by default. If you're pasting anything covered by a customer contract, settle that before you paste, not after.
No comments yet.