Measure AI search visibility by tracking whether a brand is mentioned and whether its pages are cited across a stable set of relevant questions. Keep the exact prompts, answer context and observation dates so the results can be compared. Website visits are a separate outcome, because an answer can mention a business without sending a click.
Reviewed on 1 November 2026, this guide proposes a repeatable observation method rather than an industry-standard score. The worked percentages below use invented counts to explain the calculation. No visibility results for Anteomnio or a competitor are claimed.
What should you count?
Separate a brand mention from a link to the brand's website. Also distinguish a favourable recommendation from a neutral reference or warning. A brand appearing first in an answer is not automatically a useful outcome.
| Measure | Definition for this worksheet |
|---|---|
| Mention rate | Valid observations containing the brand ÷ valid observations |
| Direct citation rate | Valid observations linking the brand's site ÷ valid observations |
| Context | Human-reviewed description of how the brand appears |
| Referral outcome | Visits and useful actions attributable to identifiable referrals |
Count each brand once per answer for the mention rate, even if it appears repeatedly. Use the same rule for every competitor. Record unsuccessful collection attempts separately instead of quietly treating them as answers with no mention.
Separate three kinds of evidence
Record context, including negative references.
Inspect the cited page and supported claim.
Track identifiable referrals separately.
How do you build a useful prompt set?
Start with questions customers actually ask when learning about a problem or choosing a service. Include informational and commercial questions, but report those groups separately. A brand's absence from a broad educational answer means something different from its absence from a supplier shortlist.
Keep the original wording, intended audience, language and location context. Avoid adding your own brand name to every question; that would measure recognition under a brand cue rather than unprompted inclusion.
Maintain a stable core set for trend comparisons. You can add exploratory questions, but do not mix them into the historical series without noting the change. Percentages do not make different prompt pools automatically comparable.
How should observations be recorded?
Record the engine or interface, date, prompt, visible answer and cited URLs. Where available, record model or mode, search settings and whether the session had prior conversation context. Different settings can change the result.
Repeat observations across your chosen review periods. Preserve enough evidence to inspect a surprising change. If a tool provides a score without the underlying answer or collection method, ask how the score is produced before relying on it.
Keep the observation reproducible
- 01PromptPreserve wording and intent.
- 02ContextRecord engine, mode and session conditions.
- 03AnswerSave the response and cited URLs.
- 04ReviewApply the same counting rules.
How do you calculate the rates?
Suppose a fixed set produces 40 valid observations. If 10 mention the brand, the mention rate is 10 divided by 40 = 25%. If six link to the brand site, the direct citation rate is 6 divided by 40 = 15%. These hypothetical inputs explain the calculation; replace them with your recorded observations.
For repeated runs, state whether the denominator counts unique prompts or all prompt-run observations. This worksheet uses observations. A prompt asked three times contributes three observations, which should be disclosed so readers can interpret the result.
Compare competitors using the same collected answers. Multiple brands can appear in one answer, so their mention rates do not need to add to 100%. Avoid presenting this measure as an exclusive market-share allocation.
What can Google Search Console tell you?
Google's Generative AI performance report shows impressions for supported generative features on Google Search, including AI Overviews and AI Mode. Review the report by page, date, country and device to understand where links to your site appeared. Google says the report may not appear when a site has insufficient impressions or is excluded from its generative features. A missing report is therefore not a measured zero across every AI engine.
Use that report alongside ordinary search performance and your recorded answer observations. Google's generative AI guidance says no special AI schema or llms.txt file is required. Crawl access, index eligibility, useful content and the site's Search Console inclusion settings still matter. None guarantees that an engine will cite a page.
Keep referral analytics beside your observation series. Referrals can show some visits and outcomes, but they will not capture every path where someone sees a brand and returns later through another route.
A hypothetical observation set
Count each brand once per answer.
Count answers linking the brand website.
How do you connect changes with observations?
Maintain a change register with the page, the substantive edit and the actual publication or update date. Record the baseline before making the change, then review the same prompt set afterward.
An increase after an edit is an association worth investigating. It does not prove the edit caused the increase. Model changes, source changes, competitors and collection conditions may also affect answers. Even a stable competitor trend does not remove all those explanations.
Inspect cited passages and unanswered questions to choose the next content improvement. Give readers clear definitions, useful examples and evidence for claims. Keep important information in readable page text as well as visual artifacts.
What should a monthly report contain?
An AI visibility report should identify the prompt-set version, collection period, engines and valid observation count before presenting a rate. Report brand mentions and direct citations separately, show representative answers, and disclose collection failures. Add platform impression data and attributable enquiries as separate measures; neither is a substitute for the recorded answers.
Keep the change register beside those observations so the reader can see which pages changed during the period. Explain changes to the prompt set or collection method before comparing two reporting periods.
The SEO agency guide can help define reporting responsibilities. Anteomnio's search and AI visibility service connects technical access, content and repeatable observation without promising placement in generated answers.
The mention-and-citation tracking framework is informed by the GEO content-engineering material based on Eugen Ullrich's work under CC BY 4.0. The definitions, hypothetical example and caution about causal attribution above state the method used in this guide.
Sources and calculation notes
The counts and rates are hypothetical: 10 of 40 observations is 25%; six of 40 is 15%. No live tracking study was conducted. The method defines observations explicitly and does not claim causal proof from a before-and-after change.
- Google: Generative AI performance report. Defines the supported Google Search features, impression reporting, available dimensions and reasons the report may not appear.
- Google: optimizing for generative AI search. Explains crawl/index eligibility, Search Console inclusion settings and why special AI markup is not required.
- GEO Content Engineering ? Eugen Ullrich. Attribution: this guide adapts the measurement and change-register approach from Eugen Ullrich?s GEO Content Engineering framework (CC BY 4.0). Anteomnio has added observation definitions, hypothetical examples and limitations; this source does not substantiate measured Anteomnio results.
- Creative Commons Attribution 4.0. Licence terms for the adapted framework.
External references checked on 1 November 2026.