The rule above all others
A number that is wrong is worse than no number. Prices, limits, model names, and feature availability change faster than an article survives. When a fact is unstable, we say “check the official page” and link it, instead of freezing a figure that will rot.
Checked is not the same as tested
We use two different words on purpose. Checked means we read the official documentation and verified a name, a path, an error string, or a feature description. Tested means we ran the task ourselves and can describe what happened. A guide can be checked without being tested, and it says which one it is at the top.
- Product claims are checked against OpenAI’s own documentation, linked in the piece.
- We do not run benchmarks, and we do not publish scores or success rates.
- Workflow guides are built from documented product behaviour and our own drafting work; we do not claim a controlled trial.
- Where we have not tested something, the guide says so in its evidence box.
Source hierarchy
- Official documentation. OpenAI help pages, the pricing page, the status page, the privacy policy. These win every disagreement.
- The product itself. Menu paths and error strings we can see.
- Public reporting and community threads. Used as leads, never as proof. If a claim only exists in a forum post, it does not ship.
Each guide lists its sources at the bottom and states what was not tested. When a source changes, the guide changes; when we cannot verify a number, we link the page instead of printing the number.
Review cadence
Every guide shows a “last reviewed” date. We review a guide when the official documentation changes, and at least once a quarter. If a page is more than 90 days old, treat the specifics as unverified until you check the linked source. The changelog records substantive changes.
How AI is used here
We use AI tools in the drafting process. Every published piece is edited by a human, and every factual claim that can be checked is checked against a source. We do not publish AI output as reporting, and we do not invent quotes, customers, or tests.
You are reading a site about using AI well. Pretending we do not use it would be silly. Pretending it does the work would be dishonest.
Corrections
If something here is wrong, email hello@chatpt.chat with the URL and the sentence. We aim to acknowledge corrections within five working days. Corrections that change a claim are noted in the changelog. We do not quietly edit a claim and pretend it was always that way.
If you are correcting a fact, include a source we can open. “This feels wrong” is a useful signal but not enough to change a page.
What we refuse to publish
- Fake screenshots of a product we do not control.
- Invented benchmarks, user counts, or “studies”.
- Promises of access, account recovery, or bypassed limits.
- Pages whose only job is to move you to another site.
How to cite us
Cite the page title, the site, the review date, and the URL. Example: “ChatGPT, minus the mystery”, chatpt.chat, reviewed 31 October 2026, https://chatpt.chat/guides/chatgpt-basics/. The review date matters more than the publication date, because that is when the facts were last checked.
Independence
chatpt.chat is not affiliated with OpenAI. “ChatGPT” is a trademark of OpenAI and is used here only to name the product we write about. We do not use OpenAI’s logo, and we do not claim to be the official product.