An AI visibility report is useful when you can see what was tested and which decision follows from it. An impressive score on its own explains neither your visibility nor your revenue. This guide shows what to ask for in someone else's report – and how to run a check yourself. It contains no published measurements, customer results or internal Pakun data.
The question and the test method must be visible
Ask for the exact wording of each question, the date and the system used. The access route matters too: an API response is not automatically the answer a signed-in user gets in the app. Region, language, web access and conversation context can all make a difference.
Keep open customer questions separate from brand questions. “Who repairs roofs near me?” tests a different situation than “What do you know about [business name]?” Mixing the two produces a rate that looks clear but is hard to interpret.
Fix the questions before measuring
- Example without a brand: “What should I look for when choosing an electrician in [city]?”
- Example with a brand: “Which services does [business name] offer according to available sources?”
- Record question text, language and region in advance; don't tailor them afterwards to get the result you want.
Keep a reliable log
For every query actually run, record:
- The full question and the date with time zone
- System, model name where shown, and access route (app, browser or API)
- Language, region and any visible search or grounding setting
- The unedited original answer and source links, if provided
- Separate labels: named, linked as a source, described correctly, or not named
A system you couldn't reach gets the status “not tested” – never a visibility score of zero.
A missing mention is an observation, not proof of lost revenue
If a business isn't in an answer, that only documents this answer under these conditions. It doesn't prove lost jobs or a permanent rank. A competitor being named doesn't prove that one specific measure caused it.
Repeating the questions can show whether answers fluctuate – as long as questions, conditions and scoring stay comparable. Even a repeated sample doesn't represent all customers, systems and times. Web research is also not the same as an AI query that was actually run; label each correctly.
Three categories that belong apart in any report
- Observation: what does the original answer or the checked page actually say?
- Interpretation: which explanation seems plausible, and what are the alternatives?
- Action: what can be improved concretely and checked afterwards?
Fictional example: a system shows outdated opening hours. The next step is to compare the business sources you control and correct the wrong details. That's different from claiming that a particular schema markup will guarantee correct answers in future.
Turn findings into useful actions
- Contradictory hours: check the official details and correct them at the specific source.
- Unclear service: add an accurate explanation to the matching service page.
- Unverifiable claim: ask for evidence or remove it from your public description.
Prioritize by usefulness to real visitors and realistic effort, not by an invented AI score.
What the official tools can and can't measure
Google includes traffic from AI Overviews and AI Mode in Search Console's “Web” search type without breaking it out, and says there's no special markup that forces inclusion. [1] Bing Webmaster Tools reports citations in Copilot and Bing AI answers in its AI Performance report, which doesn't indicate placement within an answer. [4] OpenAI and Perplexity document which crawlers must be allowed for search results [2][3], but we are not aware of an official citation report from either for site owners.
Your checklist before you buy a report
- Number and choice of questions, systems, language and repetitions stated
- Original answers as evidence, not just a score
- Limits and open points named
- No ranking or revenue guarantee
- A checkable deliverable agreed in advance
Pakun doesn't query AI assistants – neither automatically nor by hand. Our checks and analyses look at the groundwork AI systems draw on: your website, structured data, business details and positions for typical searches.
Frequently asked questions
Can I call web research an AI measurement?
No. Web research and an AI query that was actually run are different observations. Name the source and access route correctly.
Does fixing the issues guarantee a recommendation?
No. Good content and consistent details help people find their way. A specific AI mention, search position or number of jobs can't be guaranteed.
Is a high AI visibility score meaningful?
Only if you can see the questions, systems, dates and original answers behind it. Without that, a score can't be checked.
Your next step
Want to start with the groundwork instead of a score? Request the free website audit. The Visibility Analysis checks up to 25 pages, structured data, crawler rules, business details, three competitors and ten customer questions and gives you a 30/60/90-day plan – with measured findings only. To do it yourself, the AI Search Starter Kit sticks to documented facts.
General information, not legal advice. Google, Microsoft, OpenAI and Perplexity change their products and help pages; the linked sources in their current version take precedence.
Sources
- Google Search Central, “AI features and your website.” https://developers.google.com/search/docs/appearance/ai-features — Accessed November 3, 2026.
- OpenAI, “Overview of OpenAI Crawlers.” https://developers.openai.com/api/docs/bots — Accessed November 3, 2026.
- Perplexity, “Perplexity Crawlers.” https://docs.perplexity.ai/docs/resources/perplexity-crawlers — Accessed November 3, 2026.
- Microsoft Bing Webmaster Blog, “Introducing AI Performance in Bing Webmaster Tools Public Preview,” February 10, 2026. https://blogs.bing.com/webmaster/2026/2/Introducing-AI-Performance-in-Bing-Webmaster-Tools-Public-Preview/ — Accessed November 3, 2026.