How we measure AI visibility

The report answers a narrow, practical question: when people ask AI assistants for a business like yours, does the answer recommend you, and who does it recommend instead? Here is exactly how we get there.

1. Customer questions

We create a set of buying-intent questions from your category and location — for example, “Who should I hire for roofing in Austin?”. The same questions are asked on every platform so results are comparable.

2. AI platforms

Questions are sent through the APIs of ChatGPT (OpenAI), Gemini (Google), Claude (Anthropic), Grok (xAI) and Muse (Meta AI), with web search enabled. Each answer is stored with the platform, model, question, timestamp, full text and the sources it cited.

API answers are not guaranteed to match what a specific person sees in a consumer app, where results vary by account, location, conversation history and time. The report treats results as a directional snapshot.

3. Mentions and competitors

We detect your business by name and website domain. Competitors are the businesses the AI assistants named in their answers — we never add businesses that weren't mentioned, apart from competitors you list yourself, which are shown even when they received zero mentions.

4. Score

Share of AI answers that recommended your business, weighted by prominence: a mention in the top 3 counts fully, a later mention counts half. Calculated only from questions that were successfully answered. We only show a score when at least 6 answers were collected; otherwise we show raw counts.

5. Website readiness

Separately, we fetch your public homepage, robots.txt and sitemap to check whether AI systems can read and understand your site. This does not measure whether AI recommends you. Your website's content is only analyzed by deterministic checks; it is never fed into an AI model.

6. What we don't claim

• We don't estimate market share or revenue lost from a small sample of answers.
• Any financial figure in the report comes from numbers you enter yourself and is labeled as a scenario.
• When a platform fails to answer, we say so rather than filling the gap.