An AI-answer review checks individual claims against approved source information and preserves uncertainty. A mention, a source citation, a website visit and an enquiry are different observations. Keep those measures separate.
What should the record contain?
| Field | What to record |
|---|---|
| Question and context | The exact prompt, platform, product mode and date. |
| Observed claim | The relevant factual statement, with enough context to understand it. |
| Approved evidence | The public source URL, current wording and date reviewed. |
| Conclusion | Supported, contradicted, unverified or unavailable. |
| Next action | Correct owned content, clarify the question or obtain the missing evidence. |
How does the worksheet handle an uncertain claim?
As an illustrative exercise, suppose an answer describes a course as having a guaranteed placement. If the approved course page makes no such promise, the claim is unsupported by that page. Record the evidence and uncertainty; do not invent a live response, assume every platform says the same thing or claim that editing a page will force a corrected answer.
What can you verify before sampling answers?
Check business names, service scope, course prerequisites and approved public contact information. Our Mayexa website case study describes owned-site readiness work; its diagnostic observations do not establish AI recommendations or traffic growth.
How should the result be reported?
Report the sample size, platform availability, dates and supported findings. A missing engine response is unavailable evidence rather than zero visibility. Search and referral reports require their own data source.
Explore AEO and GEO services or the brand-check guide for related context.