AI Visibility

AI visibility is a measure of how often AI systems such as ChatGPT, Perplexity, or Google AI cite, mention, or recommend a brand as a source in response to a fixed set of questions. It is measured separately for each level and each system, and the measurement is repeated because answers vary.

What does this mean for your business?

When customers ask ChatGPT or Perplexity about a provider, the answer determines whether your company is mentioned at all. Whether it is mentioned can only be determined through a measurement, not by a single query on your own computer. Require every provider offering GEO and AI-SEO to provide the list of questions, the number of repetitions, the systems tested, and the date of the measurement. A number without these four details is not a valid metric.

What metrics should a GEO provider provide?

An initial measurement using a fixed questionnaire—separated by AI system—including the date and the number of times each question was asked. In addition, a subsequent comparative measurement using the same questionnaire. Only by comparing two measurements can you determine whether anything has changed. This includes how many visits from AI systems are recorded in Google Analytics 4 and how many inquiries result from them, because a citation is not yet a visit, and a visit is not yet an inquiry. An increase in the citation rate alone does not constitute success.

How to Measure AI Visibility

• Freeze the question catalog: Formulate questions exactly as customers ask them, such as “Who provides TYPO3 support in the Lake Constance region?” The catalog remains unchanged for months; otherwise, the measurements won’t be comparable.
• Repeat: Ask each question multiple times, because the same system may give different answers to the same question. Only the average across all runs is reliable.
• Separately by system: ChatGPT, Perplexity, Google AI Overviews, and AI Mode provide different answers. A mean value across all systems masks where the brand is missing. Google also describes a “query fan-out” for its AI functions: The model generates multiple related search queries simultaneously for a single question. The interface does not fully display which ones these are; therefore, a measurement only reflects what is visible.
• Counting six stages separately: page retrieval by the bot, source citation (the website is linked), brand mention (the name appears in the text), recommendation (the brand is named as an option), visit, and qualified inquiry. Each stage is a separate value; a dashboard that aggregates them into a single number hides where the problem lies.
• Citation rate with denominator: The citation rate is the proportion of responses containing at least one relevant source citation out of a clearly defined set of evaluated responses. The denominator must be transparent, as must the method for counting responses that do not involve a web search. A rate without a denominator is not comparable.
• Google via Search Console: According to Google, the Generative AI Performance Report is available for Google Search’s AI features. For other systems, there is no comparable report from the provider.

From Our Experience

On October 12, 2026, we conducted a variance analysis for our own website: 3 questions, each asked 10 times via Perplexity using the sonar and sonar-pro models, for a total of 60 responses. We counted the “brand mention” level. When asked about a TYPO3 service provider, Sonar-Pro ranked us third in 10 out of 10 runs, while Sonar ranked us third in 0 out of 10. The Sonar model listed 35 different companies across its 10 runs, none of which appeared in every run. The measurement cost $0.29. A single query would have yielded either “always included” or “never included,” depending on the model.

A page view is not the same as a visit. In the server logs from October 5 to 11, 2026 (7 days, 67,533 lines), PerplexityBot viewed 804 pages, ClaudeBot 432, OAI-SearchBot 225, and ChatGPT users 162. Google Analytics 4 showed 0 sessions from chatgpt.com during the same period; in September, there were 2. The AI systems are therefore constantly scanning the website, but this rarely results in visitors. Anyone who measures AI visibility solely based on bot traffic is measuring only the “retrieval” stage and nothing about citations, mentions, or visits.

Since October 13, 2026, we’ve been running our own measurement using a fixed catalog of 150 questions as a baseline; 29 of these have also been tracked in a ChatGPT tracking tool in Germany since October 16, 2026. And here’s a counterexample: In a test on September 1, 2026, a tool showed 0 mentions for our domain in Google AI Overviews. This value was not reliable, so we did not include it in our free check.

Definition

• Generative Engine Optimization (GEO) refers to work done on the website; AI visibility measures whether this work is having an effect.
• The Visibility Index measures rankings in classic Google Search using a fixed set of keywords; AI Visibility measures mentions in generated responses.
• AI Overviews are a single Google interface; AI Visibility also includes ChatGPT, Perplexity, and other systems.
• AI bots crawl pages. A crawl indicates that a system is aware of the page, not that it mentions it in responses.

Red Flag in the Offering

A dashboard with a single number (“AI Visibility: 42”)—without a questionnaire, without specifying the systems, without a denominator, and without the number of repetitions—is not a measurement. Equally dubious is the promise that the brand will become visible in ChatGPT or Google AI, since the answers vary from run to run, and no provider controls them. A report that infers an increase in citations from a higher number of queries—without having measured visits or actual queries—promises something it has not substantiated. Google itself advises caution regarding tools that promise ranking success.

Source

• Google Search Central: Optimizing your website for generative AI features on Google Search, as of July 10, 2026, accessed on October 22, 2026
• Own variation measurement from October 12, 2026: 3 questions, 10 runs each, 60 Perplexity responses
• In-house log file and GA4 analysis from October 5 to 11, 2026
• Our own question set with 150 prompts, frozen on October 13, 2026
• In-house tool test on AI overviews from September 1, 2026