Research · Methodology
How Cloute measures AI visibility
By Cloute · Published October 1, 2026
The short answer
Cloute asks ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews the questions real customers ask about local businesses, keeps every answer in full with the sources the engine used, reads each answer for whether and where a business is named, and counts. It has collected and read 120,000+ answers in 2026. Comparisons over time are only made on identical questions and identical engines, and businesses that are still behind stay in every total.
AI answers collected and read
sources the engines used, logged
searches the engines ran, recorded
What is collected
| Data | What it is | Scale, 2026 |
|---|---|---|
| AI answers | The full text of each answer an engine gave to a customer-style question | 120,000+ |
| Sources | The web pages each engine reported opening or citing while it answered | 260,000+ |
| Searches | The search queries the engines ran on their own while answering | 150,000+ |
| Websites | Distinct websites the engines cited | 12,800+ |
| Competitors | Businesses named by AI in the same answers, tracked by name | 400+ |
| Crawler visits | Server-log visits by self-identified AI crawlers to pages Cloute publishes | 80,000+ |
How questions are asked
- Questions are written the way customers ask them, about the kind of business and the town, not about the business by name.
- They are asked from the places customers are, because local answers change with location.
- Each engine is used the way a consumer uses it, with live web search on.
- Each call is independent: no conversation history, no memory, no account personalisation.
- Questions are repeated, because the same question can produce a different answer on the next run.
How answers are read
- Each answer is read for whether the business is named, whether it is named first, which other businesses are named, and whether what is said is positive, neutral or negative.
- Only answers read this way count toward any percentage Cloute publishes.
- "First pick" means the business is the first one the answer recommends.
- An answer counts as having "read" a business when the engine's own record of sources includes a page on that business's website.
Rules for every published figure
- Like for like, or not at all. A before-and-after only counts when the question wording and the engine are the same on both dates. Otherwise a figure is reported as a standing, not a change.
- Broken instruments are excluded. When an engine returned nothing for every business for a period, its recovery is left out of comparisons rather than counted as growth.
- Run-to-run variation is respected. Movement inside the normal range of variation is reported as steady.
- The misses stay in. Businesses still behind their strongest rival are included in every total.
- Headline counts are rounded down and shown with a plus sign, so they stay true as the dataset grows.
- No client is named without permission, and no internal score is published.
How AI crawler visits are counted
Every request to the pages Cloute publishes is logged on the server. A visit counts when the user agent matches a crawler name its operator publishes, such as GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot or PerplexityBot. Image, script and style files are excluded. Crawlers that do not identify themselves are not counted, so crawler figures are a floor.
Known limitations
- The engines change models and search behaviour often. Figures describe the period stated.
- Engines report only some of the pages they read, so source-based figures undercount reading.
- The businesses measured are in the United States, across several local industries.
- Comparisons of read and not-read answers are observational. They show a consistent association, not a controlled experiment.
Questions people ask
Which AI engines does Cloute measure?
ChatGPT, Google Gemini, Google AI Overviews, Claude and Perplexity.
Why repeat the same question?
AI answers vary from run to run. One answer cannot tell a real change from chance; repeated answers can.
Does Cloute publish how it improves a client's results?
No. The measurement is published in full. The work that changes the result is what clients pay for.
How to cite this research
Cloute (2026). How Cloute measures AI visibility. https://cloute.ai/research/methodology
The underlying data is open under CC BY 4.0: github.com/luked9/cloute-ai-visibility-research.
About Cloute. Cloute is an AI visibility company for local businesses. It measures how often ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews recommend a business against its local competitors, and works every month to move it to the top.