Method
How to Measure a Site's AI Visibility
Measuring AI visibility means watching how AI assistants answer real questions and checking whether your site shows up. Here is what RankAI observes, and what it does not.
If you are asking how to measure AI visibility for a site, start with a concrete scenario. A potential customer types a question about your industry into an AI assistant and gets an answer that cites some sites and ignores others. Measuring AI visibility means observing those answers systematically, working out whether and how your site appears, and reading the context around any mention. RankAI's howItWorks page walks through exactly this process: which AI engines are observed, how answers are collected, and how they should be read.
What RankAI observes
RankAI's public pages name ChatGPT, Perplexity, Gemini, Grok and Google AI. They do not all sit inside the same observation perimeter, though. Claude and Grok are measured only after you connect them, so whether they appear in your results depends on how your account is set up. Another point to keep in mind: Gemini and Google AI Overviews are not the same observation. They are two distinct surfaces and should be read separately, never added together or treated as one.
Reading the process on the howItWorks page
The howItWorks page lays out the observation process as a sequence of checkable steps. Here is how to read it in practice, with a test for each stage:
- Define the questions to observe. The test: every question should match a real intent in your market and be phrased the way a person would write it, not as a standalone keyword.
- Check which engines are active in your account. The test: confirm that ChatGPT, Perplexity, Gemini, Grok and Google AI are configured, and remember that Claude and Grok appear only after you connect them.
- Collect the answers and keep the surfaces apart. The test: treat Gemini and Google AI Overviews as separate observations with separate findings, and do not merge them into a single indicator.
- Read each mention in context. The test: for every answer, note whether the site is cited, in what role, and alongside which other sites, rather than settling for a yes or no.
RankAI's contribution: an operating proposal
This is RankAI's operating proposal, not a universal standard: read AI visibility as a set of separate observations, not as a single score. The reason is simple. An API response is not a person's screen. When we observe an engine through its API, we see the text returned to a request, not the interface a user sees on screen, with its visual elements, highlighted links and browsing context. That means the data should be read as signals about the content of answers, not as a snapshot of the user experience.
From this follows a three-step way of reading the results. First, keep the surfaces separate, because Gemini and Google AI Overviews are not the same observation and adding them together would produce an ambiguous indicator. Second, distinguish the engines that are always observed from those that enter the perimeter only after you connect them, such as Claude and Grok, so you know which results depend on your configuration. Third, read every answer as a context, checking who is cited alongside your site and in what wording. Only by keeping these distinctions does the observation stay checkable and useful for deciding where to act.
One limit to keep in mind
The checkable limit is this: the AI visibility RankAI observes covers the engines named on its public pages and, for Claude and Grok, only after you connect them. It is not a universal measure of every AI engine out there, and answers collected through APIs do not match what a person sees on screen. That is why results should be read as separate, contextualized observations rather than a definitive ranking.
To follow the full process and see how each engine is handled, see the in-depth page.