AI search visibility is evidence from a particular set of questions, systems and dates. A website can be accessible and well structured while still being absent from the answers people receive. Technical readiness and observed visibility therefore need separate interpretation.
Begin with the questions people actually ask
Use the audience's language and describe their situation, constraints and desired outcome. For an office furniture supplier, a useful question is “Which adjustable desks suit a small shared office and have a quiet motor?” Repeating the supplier's name in every question would measure recognition of a prompted brand rather than discovery in the market.
Include informational, comparison and decision questions. Keep explicit brand checks separate so that you can understand both discovery and reputation. Record the region and language: the same wording can produce different results for different audiences.
Read the evidence before the headline score
The audit report should show what happened for each query and each system. These observations answer different questions:
| Observation | What it tells you | What it does not prove |
|---|---|---|
| Brand mention | The answer named the business | That the system used its website as a source |
| Link or citation | The answer included an identifiable source URL | That the answer recommended the business |
| Relevant answer | The system returned an answer to the question | That the business was present in that answer |
| Provider error or timeout | That observation could not be completed | That the business was absent |
| Technical finding | A page has a particular accessibility or content issue | That fixing it will guarantee a citation |
Review the underlying answer text and cited URLs. A short summary helps navigation, but the saved evidence makes the finding reviewable.
Keep the denominator visible
If a measurement says that 18 of 60 successfully collected answers mention a brand, its mention rate is 30%. This is an illustrative calculation, not a White GEO rating scale. If additional requests failed, report their count separately and explain which observations were included in the calculation.
Do not silently treat unavailable providers as negative mentions or compare an incomplete run with a complete one as though coverage were identical. The report's methodology should explain its own indicators, weighting and limitations. There is no universal score threshold that guarantees stable visibility across all AI systems.
Turn findings into a practical next step
Start with the questions that matter to customers and where competitors appear with useful sources. Check whether your site has a clear answer, current product details, verifiable business information and accessible supporting material for those questions.
After changing the site, repeat a comparable measurement. Keep the query set, locale and available systems visible so that changes in the experiment are not mistaken for changes in market visibility. Recommendations are hypotheses to test; an audit cannot promise inclusion in future answers.
Further reading
-
Answer Engine Playbook: build useful answers for realistic questions.
-
Entity Footprint Model: keep business identity consistent and verifiable.
-
Google's guidance on AI features and websites: understand the search platform's published guidance.
llms.txt, llms-full.txt, Speakable, and FAQPage do not earn automatic readiness points and are not additional requirements for Google's AI search features. Google Search does not use llms.txt; appropriate structured data must match visible page content. Accountability and provenance signals are sampled diagnostic heuristics, not verified authority or a Google E-E-A-T rating. Indexing and snippet eligibility do not prove actual indexation or guarantee citations. Sources: Google's generative-search guide and Google's AI-feature requirements.