Method
AI Visibility: Measuring the Gap Between Reading and Reporting
AI visibility isn't a single score. It's the gap between what an AI engine reads on a page and what it reports back, measured one page and one engine at a time.
How do you measure a site's AI visibility when the AI engines read and report pages in different ways? You measure it by comparing what an AI engine reads on a page with what it reports back in its answer. RankAI starts from the method page to make this clear: AI visibility does not come from a single number. It comes from the gap between reading and reporting, page by page and engine by engine.
To make the comparison concrete, imagine an illustrative case that was not observed: a page with three distinct sections, where one engine reads the full text and another reports only the title in its answer. This is a hypothetical example, useful only to show why measurement has to start from the comparison and not from an aggregate value.
Why the gap is the working criterion
RankAI's public pages name ChatGPT, Perplexity, Gemini, Grok and Google AI. Claude and Grok are measured only after you connect them. Gemini and Google AI Overviews are not the same observation, so they stay separate in any comparison: what an engine reads and what it reports depends on the specific observation. An API response is not a person's screen, so the report has to be read in the context where it was produced.
How to classify the types of gap
RankAI's working proposal is not to stop at recording the gap but to assign it a type. Three categories make the classification actionable:
- Omission: the engine reads a piece of content but does not report it. Verifiable criterion: the passage is present in the reading and absent from the report.
- Summary: the engine reports the content in reduced form. Verifiable criterion: the report keeps the meaning but cuts details or passages that were present in the reading.
- Substitution: the engine reports content that differs from what it read. Verifiable criterion: the report attributes information to the page that was not present in the reading.
Steps for measuring AI visibility
- Fix the page and the engine. Verifiable criterion: each comparison covers a single page and a single observation, with no mixing of engines.
- Record the reading. Verifiable criterion: you note what the engine actually read on the page, section by section.
- Record the report. Verifiable criterion: you note what the engine reports in its answer, distinguishing an API response from a person's screen.
- Classify the gap. Verifiable criterion: every difference gets one of the types (omission, summary, substitution) and does not stay generic.
- Compare over time. Verifiable criterion: the same comparison is repeated on the same page and the same engine, so the gap reads as a variation rather than an impression.
The point this piece adds is the distinction between recording and classifying. Recording the gap between reading and report is the first step, but on its own it produces a list of differences that is hard to use.
Classifying the gap by type — omission, summary, substitution — turns that list into something you can act on. A recurring omission on a section tells you that part is not being picked up. A summary tells you the content is picked up but compressed. A substitution tells you the report drifts away from the reading.
This classification is RankAI's working proposal, not a universal standard. Its value lies in making different observations comparable without reducing them to a single number. Because Gemini and Google AI Overviews are not the same observation, and because an API response is not a person's screen, the type has to be applied inside the specific observation, not on top of it. That way the comparison between what an engine reads and what it reports stays a verifiable criterion, repeatable on the same page and the same engine.
Verifiable limit: this method measures the gap between reading and report for one page and one observation at a time. It does not produce an aggregate value, and it does not cover observations outside RankAI's public pages. For the full picture of the method, see the method page.