Define the signal before measuring AI search visibility
AI search visibility is not one metric. Eligibility describes whether content can be considered. An impression describes a reported appearance. A mention names the business; a citation uses a source; a recommendation suggests a choice.
A referral is a visit, and an enquiry is a further business event. Measure them separately.
Start with a small set of real buyer questions and the services relevant to your market. Combine the account reports available to the owner with saved answer observations and suitable analytics or request records.
State the coverage of each source so a missing report or changed interface does not become a false trend.
What each signal can and cannot establish
Choose definitions the team can repeat. Decide whether a recommendation must suggest your business as a provider or whether a neutral mention counts for a separate measure.
Keep the cited URL and the wording together: a source may support a general explanation while the answer recommends someone else.
| Signal | Evidence | Conclusion to avoid |
|---|---|---|
| Eligibility | Relevant access, indexing and inclusion checks. | The brand must appear because the page is eligible. |
| Impression | A defined account report with period and filters. | Every impression is a visit or recommendation. |
| Citation or mention | A suitable report or saved answer and cited URL. | One answer represents every user or service. |
| Referral | Recorded visit source and landing page. | A visit count measures all answers that named the business. |
| Enquiry | Received request with an appropriate recorded connection. | Every attributed request proves which answer persuaded the buyer. |
Use the Google reports available for the property
Google’s current help says generative AI report insights rolled out worldwide on August 31, 2026. The Search report measures impressions in AI Overviews and AI Mode. Check your property: insufficient impressions can make the report unavailable.
Google’s report documentation describes availability and data limitations, including unavailable values that can export as zero. Preserve filters and period with the evidence.
Do not label a missing report as measured absence or reinterpret an impression metric as a click-through or enquiry rate.
Use Bing citation data within its stated coverage
Bing’s official AI Performance announcement describes citation counts, cited pages, sampled grounding queries and trends across supported Microsoft and partner experiences.
These are source-reference measures; the documentation explicitly separates them from ranking or placement in an individual answer.
Keep this coverage distinct from your ChatGPT or Google observations. A grounding query is a retrieval phrase reported by the tool, not automatically the exact customer question. Use it to investigate relevant content, while preserving the report definition and sampling limit.
Investigate recorded ChatGPT referrals
Inspect the visit source, landing URL and campaign information that your analytics actually received. If a visit is recorded from ChatGPT, keep that evidence alongside the landing page and time context.
Test the current path when attribution is important rather than assuming that every browser or answer passes the same information.
Referral records describe recorded visits, not every answer that cited or mentioned the business. They may not identify the original conversation. A page can be cited without receiving a click, and a visitor can make contact later through another route.
Keep unknown attribution open; do not reconstruct it from a plausible story.
Worked example: three changes that are not the same result
A fictional supplier records an increase in a Google impression view, citations to a guide in Bing’s report and visits identified as ChatGPT referrals. The team asks whether AI search has produced more qualified quotes.
Its received-request records do not yet connect those visits to enquiries.
Make the observation set repeatable
Keep the exact question, language, market context, date, service and available mode label. Save the relevant answer text or screenshot, brand wording and cited destinations. Record whether the attempt succeeded; a collection failure is not a zero mention.
Compare like periods and the same definitions. If questions, providers or report coverage change, identify the break before discussing a trend. A useful first plan is deliberately limited enough that someone can repeat it and inspect the original evidence.
The complete Alytixx website audit reviews the underlying pages and readiness within agreed coverage. Use this guide to define any answer observations, account reports or follow-up measurement your question also requires.
Questions about AI search measurement
How can I compare results when the reports or questions change?
Keep a record of the questions, services, filters and report coverage used in each period. Mark changes in the comparison and compare a consistent subset where possible. A broader sample or newly available report can change the count independently of your website.
What can GSC and Bing reports show?
Use each report’s actual definition and account availability. Google provides generative AI impression views; Bing describes citation activity across supported experiences. They are not interchangeable counts.
Do ChatGPT referrals prove a recommendation?
No. A recorded referral establishes a visit source under the available attribution. To assess a recommendation, inspect the answer wording and context.
Does missing data mean zero visibility?
No. Missing access, unavailable reports and failed collection are different from a measured zero. Preserve the data state and compare periods only with coverage context.
Sources
Official guidance behind the search checks on this page.
Connect your visibility question with website work
Request a complete website audit for the search, offer and enquiry issues behind your measurement question. View the sample report to see how page tasks and missing information are handled.
