AI visibility guide
How to measure your brand's AI visibility
The short answer
You measure it with a fixed set of real customer questions, asked of each assistant separately and repeated over time. You count three things: how many answers name you, how many cite your site, and whether what they say is right.
One question asked once isn't a measurement: answers change from one run to the next.
The method in five rules
- Questions without your brand. The ones someone who doesn't know you yet would ask: by category, by problem, by area.
- Many different questions beat many repeats. Sistrix reached that conclusion in October 2026, and also found that question lists a brand picks for itself tend to flatter it against competitors.
- Each assistant on its own. ChatGPT, Perplexity, Claude and Google cite different sources; one overall figure hides the differences.
- In your customer's language and country. The same question in English from London and in Spanish from Madrid names different brands.
- Always the same questions. Change the list every month and you can't compare.
What to count
| Metric | What it is | Example |
|---|---|---|
| Named | Answers that mention your brand, out of the total | 3 of 20 |
| Cited | Answers that link to your site as a source | 1 of 20 |
| Accuracy | Facts about you the AI gets right | Right town, old opening hours |
| Competitors | The brands named instead of you, and how often | Brand X in 11 of 20 |
An example with real numbers
On 31 October 2026 we measured pablodpavon.com for the Argentine market (es-AR). Technical foundation: 94/100. Mentions: 0 of 13 AI answers about its field. That's why the main number in a SearchedBy result is how often AI names you, with the technical foundation underneath: a high technical score can hide the fact that AI doesn't name you.
Tools you already have
- Google Search Console: clicks and impressions from AI Overviews and AI Mode count in the Performance report (‘Web’), and since 31 August 2026 there's a separate generative AI report with impressions only.
- Bing Webmaster Tools: since February 2026 it has an AI performance report (public preview) for Copilot and Bing's answers.
- Your own test: a spreadsheet with the questions, the date, the assistant, the brands named and the sites cited.
- The SearchedBy free scan: it asks several assistants five questions and shows you who comes up. It's a first snapshot; following it month by month needs a different measurement.
Common questions
How often should I measure?
The useful minimum is once a month with the same list, reading ranges. Firm conclusions need more questions and more runs, and it's worth measuring more often right after a big change.
Are API answers the same as in the app?
Not always. Tools, ours included, ask through each AI's official API (Google's AI Overviews through a data provider), and the app can answer differently. Read ranges and trends.
Sources
- Sistrix, ‘Breite schlägt Wiederholung’ (in German) (2026-10-27)
- Schulte, Bleeker and Kaufmann, ‘Don't Measure Once’ (preprint) (2026-04-08)
- Google Search Central, ‘AI features and your website’ (2025-12-10)
- Bing Webmaster Tools, AI Performance report (2026-02-10)