Reading an AI visibility report: evidence before impressive scores

Make the question and test channel visible

Ask for the exact question, date and system used. The channel matters too: an API response is not automatically the same as a signed-in user's answer in a consumer app. Region, language, web access and conversation context may also matter.

Separate open customer questions from questions that name the business. 'Which business repairs roofs in this area?' examines a different situation from 'What do you know about Business X?' Mixing them can produce a seemingly precise percentage that is difficult to interpret.

A missing mention is an observation, not proof of lost revenue

If a business is absent from an answer, that initially establishes only what happened in that answer under those conditions. It does not establish lost orders or a permanent ranking. A competitor's mention does not prove which action caused it either.

Repeated tests can help reveal variation. Questions, conditions and evaluation need to remain comparable. Even repeated samples do not automatically represent every customer, system or moment in time.

Separate three things in the report

Observation: what actually appears in the original answer or inspected page? Interpretation: which explanation seems plausible, and what alternatives exist? Action: what can be improved and checked afterwards?

Fictional example: a system gives outdated opening hours. The practical next step is to compare the accessible business sources and correct inaccurate information. That is different from claiming that particular schema markup will guarantee correct answers in future.

• Original answer is accessible
• Source and date recorded
• An inference is labelled as an inference
• Action is tied to an observed finding

Search fundamentals matter more than a magic format

Google's guidance for its AI search features continues to emphasise helpful, reliable content and established search fundamentals. Inclusion is not guaranteed, and no special markup forces a recommendation. Start with understandable service pages, accessible content and accurate business facts.

Do not buy a blanket visibility guarantee based on a single screenshot. Ask about the deliverable, method, limitations and a practical next step instead.

Your checklist before booking

Confirm the number and selection of questions, systems, language, repetitions, output format and included discussion. A small snapshot can be a useful starting point; it does not replace a comprehensive website or SEO audit.

Pakun distinguishes small samples from broader checks. Extra systems, repetitions and implementation must fit the explicitly agreed scope.

• Question count and systems defined
• Original evidence rather than a score alone
• Limitations and open questions visible
• No ranking or revenue guarantee
• An inspectable deliverable agreed

Sources & scope

The examples are illustrative, not customer case studies. This checklist supports your own review; it does not promise a business outcome.

• Google: AI features and your website