Method

How we check sources

AI is a field where rules, products and claims change every few weeks. This page explains what we do on every page so you can judge our work, and check it yourself.

Last updated October 24, 2026

A library reading table with an open printed report, a pencil and a brass lamp
AI-generated photograph. See the image credits.

The four steps

  1. Find the primary source. For a law, that is the official text. For a framework, the body that issued it. For a product, the company's own policy or help page. For a case, the court order or the regulator's decision. We go to these before we read anyone's summary of them.
  2. Link it. A claim that matters carries a link to the document it comes from, and every page ends with a Sources list that says what each source is and what we used it for. If we cannot point to a source for a claim, we leave the claim out.
  3. Date it. Every page shows the day it was published and the day we last checked its sources. We open each source again when we update a page. If nothing had changed, we move the checked date and leave the content alone.
  4. Say what is still unknown. Each page ends with a short list of what is disputed, unsettled or likely to change. A guide that sounds certain about a moving subject is a guide that is hiding something.

What "last checked" means

It means that on that date we opened every source listed on the page and compared what it said with what our page says. It does not mean the page is a guarantee. Laws are amended, companies change their settings, and a page can be out of date the day after we check it. Open the source if the decision matters.

How we write

  • In our own words. We explain documents; we do not copy them. A short quote is marked and credited.
  • Plain English first. A technical term is explained the first time it appears.
  • Non-partisan. We describe a rule, who proposed it and what it does. We do not endorse or attack a party, a politician, a bill or a company.
  • Defence only. We teach people to recognise and report scams, fakes and misuse. We do not publish instructions for causing harm with AI.
  • No invented numbers. If a figure is not in a source we can link, it is not on the page.

What we do not do

  • We do not test products and we do not rank them. We report what a company says it does, with the date, and say that settings change.
  • We do not give legal, medical, financial or investment advice, and we do not certify or approve anything.
  • We do not claim members, a board, awards or partners. We are not a regulator or a standards body.
  • We do not name private individuals.

Found a mistake?

Tell us at [email protected], with the address of the page and the source that shows the problem. Our editorial policy and corrections page explains what happens next.