How do you check whether an answer is accurate?

How a claim gets flagged

The platform runs each service's question set daily against ChatGPT, Gemini, Perplexity, Copilot and Google AI Mode, and keeps the full answer with every source it cited. A reviewer on our team then reads those answers with your service catalogue open. The unit of review is the claim, not the answer: the fee, the deadline, the documents a resident has to bring, the office and its hours, the rule about who qualifies. Each one is marked correct, outdated or wrong.

Outdated and wrong are kept apart on purpose. An outdated fee was true once, and it usually traces back to a page or a PDF of yours that is still online. A wrong claim was never true, and it usually traces somewhere else.

Why the source matters more than the flag

Every flag carries the URL the answer cited, because the fix depends on it. A stale PDF on your own domain is a retirement or a redirect. A third-party guide that paraphrased you is a case for making your own page the easier one to quote. A page the engine never reached is a crawler access problem, and the audit will already have found it.

Why there is no score

We do not publish an accuracy percentage, and we do not think a computed one would be honest. Whether an answer about eligibility is right depends on the wording of your own rule, and reading that is the work. What you get instead is the list, claim by claim, engine by engine, with sources. After your team ships a correction the question keeps running daily, so the list also tells you when a claim stopped appearing and which engine is still repeating it.