Define what you are measuring
A brand mention, a source citation, a referral visit, and an inquiry are different observations. Record them separately. An engine may name a business without linking to its website, or cite a page that produces no useful visit.
Use available first-party reporting
Google introduced Generative AI performance reports in Search Console in June 2026, initially for a subset of websites. Where the report is available for your property, record its AI-feature impressions, pages, countries, devices, and date range alongside the overall search performance report. If it is unavailable, record that gap rather than treating overall Web impressions as a separate AI-only measure.
Keep platform observations, referral visits, and qualified inquiries separate from those impressions. This guide does not claim access to your property or that a report is enabled for every website.
Build a sampling plan
Choose a set of customer questions and record country, language, platform, account context, date, and whether web search was used. Repeat observations consistently enough to understand variation. No single sample size, including 50 runs, is a universal industry standard.
| Field | Meaning |
|---|---|
| Citation presence | Whether the answer links to the intended source |
| Description accuracy | Whether the answer represents the business correctly |
| Share of voice | Your stated proportion within a defined comparison set |
| Referral outcome | Visits and qualified inquiries attributed to the source |
An illustrative calculation
If a source is cited in 3 of 10 observations, its observed citation frequency in that sample is 30%. That is not a 30% probability of being cited for all users. The sample may vary by time, engine, location, and personalization.
Turn observations into work
If an answer repeats an outdated fact, correct the underlying page and profile. If the source cannot access the page, inspect status and access rules. If competitors provide more useful evidence, improve the substance rather than inventing another markup field.
Report uncertainty
Keep missing results and failed requests visible. Record simultaneous site changes and platform changes. Use permitted interfaces and collection methods. Explain the cost and feasibility of repeated sampling before treating a large run count as mandatory.
Connect to business value
Compare referral visits, inquiry quality, and engagements over time. A citation count without a defined query set or outcome is too ambiguous to serve as the only performance measure.