About
Citation Checks measures whether AI assistants can read a business's website, and what they tell customers about it. It asks real customer-style questions across ChatGPT, Claude, Perplexity, Gemini and Grok, requests the site the way each assistant's crawler does, and reports what came back — with the error rates of the instrument published beside the results.
It is built and run by one person, and it is independent of every platform it measures.
It exists because this category runs on unverifiable numbers. The figure usually quoted for AI-crawler blocking is 30%; we drew 300 US small businesses at random, measured, got 8%, and published the confidence range and every correction that moved the number — including the ones that moved it against us. That is the whole editorial position, and it is cheaper to keep than to fake.
The standard the numbers are held to
- Every figure on the site traces to a measurement we ran, or to a cited study with its conflict of interest stated on the page — for instance that the author of the largest dataset we cite competes with us.
- Retracted claims are scanned for mechanically: when a number is corrected, the old wording goes into a test that fails the build if it ever reappears anywhere on the site.
- The methodology page states what the method cannot tell you, and the classifier's own measured error rates, before any number is offered.
- The paid report carries an unconditional 30-day refund, because “did this show you something you can act on” is a question only the buyer can grade.
Contact
hello@citationchecks.com — or reply to any email we've sent you; replies are read by the person who built this, because there is nobody else here.
Tell us what to build
If something here is wrong, unclear, or missing, say so. The useful ones get built.