AI Search Visibility: Measuring Presence Without Inventing a Score

Being mentioned is not the same as being chosen

AI search visibility asks whether a business appears in the answers people encounter through AI search and assistant experiences. It also asks what those answers say, whether they cite a useful source and whether a person takes a worthwhile next step. A screenshot of one favourable answer is a small observation, not a market share study. This guide separates platform reporting, repeatable prompt research and referral outcomes. It explains how to measure presence without inventing a universal score or claiming that a model’s answer is stable across people, places and time. Explore our marketing frameworks and measurement guide for the wider planning context.

Keep platform evidence and your own tests separate

Where a platform provides reporting, use the documented scope and current interface. Google announced dedicated Search Console reporting for generative AI features in Search and Discover in June 2026, with a later update stating worldwide availability from 31 August. The reporting includes impressions and page, country, date and relevant device views. ¹ This is Google reporting, not a report of every AI assistant. Other platforms have different tools and coverage. Keep any platform counts separate from a private prompt test, and check current documentation before repeating an older claim that no dedicated reporting exists.

Define a prompt panel before testing

For your own research, select questions that represent plausible buyer tasks: understanding a category, comparing approaches, checking capabilities or finding a local supplier. Record the exact prompt, platform, model where identifiable, date, language, location setting and whether search is enabled. Use a planned schedule and repeated observations because answers can vary. Separate branded prompts from unbranded discovery prompts. Do not write dozens of questions designed to coax the business name into the answer. That tests the prompt writer’s influence, rather than how an ordinary buyer might encounter the business.

Name the measures accurately

A panel mention rate can be defined as completed eligible test responses mentioning the business divided by completed eligible test responses in the stated panel. A citation rate uses responses citing the business’s site under the declared rule. Define whether a mention must be relevant and correctly describe the company. These are panel-based observations, not population exposure or universal share of voice. Keep answer position or recommendation categories transparent if you code them. Store the observed response for review, with appropriate privacy controls. A composite score needs explicit weights and limitations; it should never disguise incomplete coverage as a platform-wide visibility fact. If that is your next question, read our guide to SEO measurement.

Measure referrals and qualified outcomes separately

Analytics can identify some visits referred by AI services, depending on the platform, link behaviour and tracking coverage. Other visits may lose referral information or arrive later through direct or branded search. Report what can be observed rather than assigning every unexplained visit to AI. Examine useful landing-page actions and CRM qualification, with the same discipline used for other acquisition sources. An answer impression, a site visit and a new customer are different events. Where the business needs a causal claim about content changes, panel trends and referral attribution alone are not sufficient evidence. AI search visibility should therefore be reported beside qualified visits and enquiries. Share of search is explained in a separate guide.

Improve content people can verify

Prioritise accurate company identity, clear service boundaries, evidence-based explanations and maintained contact details. Write pages that answer a real question with enough context to be useful. Show dates and authors where they help readers assess claims. Use structured data accurately where appropriate; do not add facts the visible page cannot support. Precise diagrams and sources should remain readable in HTML. Avoid mass-producing near-identical answers to every prompt variation. A website that gives an assistant more text but gives a buyer less clarity has not solved the underlying communication problem. For the wider picture, see our guide to attribution vs incrementality.

Preserve a reproducible research record

Store the prompt panel and responses with a version number so later runs can be compared fairly. Record failures and incomplete outputs rather than excluding them silently when a rate improves. Decide how to handle a response that mentions the business repeatedly, cites a third-party source or gives an incorrect URL. A response-level rate should count one eligible response once under the specified rule. If multiple reviewers code answer quality, give them examples and check agreement on a sample. Report any changes in model, search mode or test conditions next to the trend. These practical details matter because a changing instrument can produce a changing score. When the platform updates, rerun a baseline and explain the discontinuity instead of interpreting every movement as the effect of your website content.

Build a report with explicit coverage

A useful review has separate sections for platform-reported visibility, prompt-panel observations, answer accuracy, observable referrals and qualified outcomes. Mark unavailable information clearly. Quiamo can help define the buyer questions, review the pages that answer them and build a measurement brief that survives changes in tools. The work should not begin with a promise to place the company in every answer. Founder participation can be agreed for positioning and service clarity. Review the research panel periodically so it reflects actual buyer tasks rather than preserving an old test because its score looks favourable. See how our SEO service can support the communication and reporting work described here.

Questions people ask

Make visibility credible

AI search visibility is worth measuring when the business can explain what was observed and what remains unknown. Use current platform documentation, a transparent prompt panel and separate referral outcomes. Track incorrect descriptions as well as positive mentions. If your team has screenshots but no repeatable method, bring Quiamo the buyer questions and the pages behind them. We can scope a content and measurement review that makes the business easier to verify and gives the report a defensible meaning. Contact Quiamo to discuss the measurement and communication scope.

Sources

1. Google Generative AI performance reporting announcementDedicated reporting and worldwide update from 31 August 2026
2. Google AI features and your websiteSEO fundamentals remain relevant and inclusion not guaranteed

Worked examples on this page are fictional and labelled. Search phrases and figures from external sources are cited; nothing here describes a Quiamo client result.