What a citation tells you
Record the answer and its source link together. Check the destination and the specific claim being supported. A linked page may cover only part of the answer; a brand mention may have no link at all. Neither should be reported as a visit or an enquiry unless you have separate evidence.
When comparing competitors, use the same question and record the context. One successful sample is not enough to say that a competitor dominates the market or that your site has been penalised.
Separate published guidance from hypotheses
Platform documentation can explain access requirements and product behaviour. An independent study may show a correlation in a defined sample. Your own review may reveal a useful pattern. These are different forms of evidence; state which one supports a recommendation.
A list of “the seven AI ranking factors” is too definite without an authoritative source. Treat suggestions about writing, evidence and usability as recommendations to evaluate, not as a leaked model formula.
Compare actual sources
- Save the exact question, product and date.
- Record whether a generated answer appeared.
- List each cited URL and the claim it appears to support.
- Open the sources and review their scope, evidence and currency.
- Compare them with your intended page.
- Choose a useful change only where the question fits your audience and expertise.
For example, an answer about project cost may cite a page with explicit inclusions and recurring fees. That comparison suggests checking your own pricing clarity. It does not prove that adding a price table guarantees selection.
Use schema accurately
Structured data should describe the organisation and content that visitors can verify. Check names, URLs, service details and dates. Valid JSON is a technical result; it is not evidence that a claim is true or that a generated answer will use it.
Google retired FAQ rich results in May 2026. Keep useful questions and answers for readers, while correcting older promises about that feature. See Google’s documentation update log.
Write for the decision
Put the main point where a reader can find it. Use descriptive headings and enough explanation to answer the question. If the answer needs a comparison, use a table; if it needs instructions, use a clear sequence. There is no reason to force every section into a question or every answer into an identical length.
Read the page aloud and test the next step. Awkward headings, repeated definitions and broken forms are problems you can fix directly.
Add evidence you can stand behind
Original project details, documented methods and qualified reviews can make a page more useful. For a case study, separate delivery facts from measured outcomes. Include the baseline, time period and source when publishing a result; get the necessary permission before using client information.
For a regulated topic, involve the responsible specialist. A marketing author should not imply that they clinically reviewed treatment advice.
Verify access independently
Check the canonical page, response, indexing directives and intended search crawler rules. Review webmaster reports and hosting evidence where available. Our AI search access guide distinguishes search crawling from training controls.
Review patterns without overstating them
Keep both positive and negative observations. Use a consistent question set, report the number of samples and explain changes in context. Do not turn a small sample into a market-share percentage or attribute revenue to it without supporting records.
Use the review to prioritise a clear service explanation, an evidence gap or a usability problem. Then record the release date and inspect equivalent periods of search and enquiry data. Read our measurement guide.