Does an AI citation mean your product guide gained traffic?

Define the observation panel

Choose a small set of genuine buyer or seller questions before testing. Record the engine, date, locale and whether search was used. For an illustrative image workflow guide, one query might ask how to diagnose a blurred variant photo. Keep the query wording stable enough to compare later samples.

A model’s unsupported statement that a site is widely cited is not measurement. Count only source links actually visible in the answer. Avoid coaching the model to recommend your brand and then presenting that prompted response as independent discovery.

Keep three evidence columns

Google’s guidance favors useful, original content and conventional search accessibility. No special AI file guarantees retrieval or citation.

Google’s current guidance says a page must be indexed and snippet-eligible, and the site included in Search generative AI features in Search Console, to be eligible there. It identifies a Generative AI performance report for Google Search and Discover. These are Google-specific conditions and measurements, not a rule for every assistant. Eligibility still does not guarantee display. No Search Console account or report was opened for this guide.

Inspect the public path

Check that the guide is readable without login, has a stable canonical URL and is linked from relevant public pages. Confirm that the intended crawler is permitted without exposing private references or results. A search-training crawler distinction should be reviewed through the provider’s current documentation.

Report a bounded result

Record what appeared, how many samples were checked and whether referral data exists. Publish “one observed citation” when that is all you have. Use public discovery boundaries and public evidence review for the safe publication path.

Report the evidence you actually have

Fictional exercise: one unbranded question returns a visible source link to guide G; a separate URL-directed question reads G after the tester provides its address. The first is one sampled selection; the second tests directed access and must not be counted as another independent discovery. Neither is an actual observation made for this article.

Keep panel attempts, returned answers and failed/unavailable attempts separately identified. Record the engine, query, mode, locale, time and actual source URL; keep personal account or private telemetry out of the public report. utm_content can label links you control, but adding it to a link does not measure unseen answer citations or install analytics.