Guide · Measurement
How to measure AI visibility: Question sets, citation share and data sources
AI answers can change from person to person and day to day, so a single screenshot proves nothing. For measurement to mean anything, the same questions must be asked on the same engines with the same method every month. This guide builds that method step by step.
AI visibility is measured by asking a fixed question set on the same engines in the same way every month. For each answer you record whether the brand is mentioned, whether the site is cited and whether the information is correct. This is read together with Search Console’s generative AI report, the AI Performance report in Bing Webmaster Tools and analytics data.
What do we measure?
| Measure | Definition | Calculation |
|---|---|---|
| Mention rate | Share of answers that name the brand | Answers with a mention ÷ all answers |
| Citation share | Share of answers that list the site as a source | Answers with a citation ÷ all answers |
| Accuracy | Share of mentions with correct information | Correct mentions ÷ all mentions |
“Citation Share” in Bing Webmaster Tools measures something different: the percentage of all citations shown for the same grounding query that belong to your siteSource 3: New AI Visibility Insights in Bing Webmaster Tools (16 Jun 2026). State your definition in the report so the two are not confused.
How is a question set built?
Collect sources
Questions your sales and support teams hear, Search Console queries, customer emails.
Group them
Brand, discovery, learning, decision and comparison. In regulated industries, leave out “best”, price and comparison questions.
Fix the format
Record each question’s wording, language and the engines it is asked on.
Freeze it with a date
If you change the set, report old and new results separately.
For a first draft, use the question set generator.
Measurement protocol
- The same question is asked with the same wording, in the same language, on the same engine surface; whether web search was on is noted.
- Each question is asked twice; answers can change even for the same question, so no single result decides.
- Each record holds: date, engine and surface, question code, answer text, mention, source list and an accuracy note.
- Raw data is kept as a table; the report is built from the raw data, not the other way round.
Kaynaq’s own measurement follows this protocol and is published in the Open Ledger. The full method is on the Kaynaq Method page.
Where do you get platform data?
| Source | What it shows |
|---|---|
| Search Console Performance report | Traffic from AI features is counted within the “Web” search type with overall dataSource 2: AI features and your website. |
| Search Console generative AI report | Your pages’ impressions in Google’s AI featuresSource 1: Generative AI performance report (Search). |
| Bing Webmaster Tools AI Performance | Citations, grounding queries and citation share in Copilot, Bing and partner experiencesSource 3: New AI Visibility Insights in Bing Webmaster Tools (16 Jun 2026). |
| Analytics | Visits from ChatGPT carry utm_source=chatgpt.comSource 4: Publishers and Developers – FAQ; referrals from other engines’ domains can be defined as a separate channel. |
How should results be interpreted?
- Look at the trend over at least three months, not a single month.
- Report engines separately; a total score hides different behaviour.
- If mentions rise but accuracy falls, find the source of the wrong information first.
- Log changes with dates: a new page, a corrected directory entry, a robots.txt change.
Frequently asked questions
01How many engines should be measured?
The ones your buyers use. ChatGPT, Gemini and AI Overviews, Perplexity, Copilot and Claude make a good starting set; Kaynaq measures eight engines.
02Why do results fluctuate month to month?
Answers can change even for the same question, and models and search infrastructure get updated. That is why each question is asked twice and the trend is read over several months.
03Can a tool do this measurement automatically?
Some steps can be automated, but judging the accuracy of answers needs human review. As long as the protocol stays fixed, the choice of tool is secondary.
Sources
- [1]Generative AI performance report (Search)Search Console Help · accessed: 26 September 2026
- [2]AI features and your websiteGoogle Search Central · accessed: 26 September 2026
- [3]New AI Visibility Insights in Bing Webmaster Tools (16 Jun 2026)Microsoft Bing · accessed: 26 September 2026
- [4]Publishers and Developers – FAQOpenAI Help Center · accessed: 26 September 2026
How to cite this page
Kaynaq. (26 September 2026). How to measure AI visibility: Question sets, citation share and data sources. https://kaynaq.pages.dev/en/guides/measuring-ai-visibility
Updates and corrections
No corrections on this page yet. If you spot an error, write to us; we publish every fix here with its date.