Home Editorial standards

Editorial standards

How we research, source, date and correct what we publish — and how we keep advertising separate from editorial.

ADVERTISEMENT

Last updated: July 2026

These are the rules we hold ourselves to. They exist because AI coverage is unusually easy to get wrong — the field moves fast, the marketing is loud, and a confident-sounding article can be badly out of date within weeks of publication.

1. We favour durable explanation over rankings

Where possible we explain the dimensions a decision turns on, rather than naming a winner. A guide that teaches you to evaluate an image generator will still be useful when the current leader has been replaced; a ranked list will not. When we do describe what a category of tool tends to be good at, we say so in general terms and tell you how to check it yourself.

2. Every article is bylined and dated

You can see who wrote a piece and when it was published. We think anonymous, undated AI content is a red flag, and we are not going to publish any.

3. Facts that change get a date and a source

The most volatile thing we publish is model pricing. It lives in one maintained table with a verification date and a direct link to each provider's own pricing page, and every tool on the site reads from that table rather than hard-coding its own numbers. See the model pricing page. If we cannot source a number, we do not publish it.

4. We describe our tools' limits on the tools themselves

Our token estimator uses a characters-per-token heuristic, and it says so, in plain language, on the page — including that real tokenizers differ and that code and non-English text will come out higher. Our cost calculators state which discounts they exclude. A calculator that hides its assumptions is worse than no calculator.

5. Advertising is separated from editorial

The site is funded by display advertising served through Google Ad Manager. Advertisers do not see articles before publication, cannot commission or veto coverage, and cannot buy a recommendation. We do not publish sponsored posts dressed as editorial. The tool picker deliberately recommends categories rather than named products, partly for accuracy and partly to remove the temptation to sell placement in it.

6. We are explicit about what we don't know

Where evidence is thin, contested, or based on our own limited testing rather than something systematic, we say so rather than rounding up to confidence. "It depends" is an acceptable conclusion when it is the true one.

7. Corrections are made openly and quickly

If we get something wrong, we fix it and — where the error was material — note the change and its date on the page. We do not silently edit away mistakes. Reports of errors go to hello@urdupoetry.top and we treat pricing corrections as urgent, because people budget from them.

8. AI is not our author

Our articles are written and edited by people. We may use AI tools the way we describe in our guides — for research support, or as an editor on a human-written draft — but a model does not get a byline here, and nothing is published without a human having read every line and taken responsibility for it.

9. Your data stays yours

Every tool on this site computes in your browser. We do not receive, log or store the text you paste into the token estimator or the numbers you enter into the calculators, and the tool picker keeps your answers in your own browser's local storage. See our privacy policy for what our advertising partner does collect.

Contact

Questions about these standards, or a correction to raise: hello@urdupoetry.top.