Method
How AI visibility is measured: gaps between Google and AI engines
AI visibility is not a single score. It is the measurable gap between how Google reads a public page and what ChatGPT, Perplexity, Gemini, Grok and Google AI report.
Suppose a services page was updated in March and Perplexity, asked about that page, still returns the earlier wording. That is a hypothetical scenario, but it shows how AI visibility is measured in Pico Neural. We do not start from a single score. We compare Google's reading of the same public pages with what ChatGPT, Perplexity, Gemini, Grok and Google AI report. For us, AI visibility is the observable distance between those readings, not an average.
What a gap is not
A gap is not an engine error, and it is not proof that a page is penalised. It is not the difference between two answers from the same engine at different times when the page has not changed. And it is not a comparison between Gemini and Google AI Overviews: those are two separate observations and should never be merged or swapped. Keeping these cases out stops us from blaming a page for something that belongs to the way we observe it.
Two gaps, opposite outcomes
Take the illustrative example again: the page updated in March, Google reads the new version, Perplexity returns the earlier one. The gap is limited to one engine and one specific piece of content. The second example, also hypothetical, runs the other way: a page that is present and readable for Google but missing from the answers of an AI engine asked about the same topic. Here the gap is not about wording but about presence. Telling the two apart changes what you fix: in the first case you work on the content, in the second on whether that engine can reach the page.
A filled-in matrix, not just a description
To make the comparison reusable, Pico Neural proposes a working matrix: one row per page, one column per observed source. It is our reading proposal, not a universal standard. A filled-in example, with illustrative values:
| Page | Perplexity | Gemini | Google AI Overviews | |
|---|---|---|---|---|
| Services page updated in March | new version | earlier wording | new version | not observed |
| Topic guide | present | absent | present | present |
The table does not produce a score. It shows where the reading diverges and on which source.
How we proceed, step by step
- We select the public pages to observe and fix their reference version, so every later comparison has a checkable criterion: the date and the content of the page at the moment of observation.
- We compare Google's reading with what ChatGPT, Perplexity, Gemini, Grok and Google AI report, keeping the Gemini and Google AI Overviews columns separate: the rule is that no observation is merged with another.
- We classify each divergence as a wording gap or a presence gap, naming the source and the page: the rule is that every row of the matrix has one label per cell.
- We flag the engines that cannot yet be observed: Claude and Grok are measured only after you connect them, so they stay out of the matrix until that step is done.
The point of this piece is to treat the gap as a unit of measurement, not a symptom. In Pico Neural, when a page diverges between sources, the first question is not “how visible are we” but “which kind of gap are we looking at”. A wording gap, as in the hypothetical example of the page updated in March, concerns the text an engine reports: you address it by working on the content and its clarity. A presence gap, such as a page missing on one engine, concerns access to the page: you address it by checking that the page is actually reachable and readable for that engine. Keeping the two types apart avoids fixes in the wrong place. The filled-in matrix serves exactly this purpose: to show in one view where the reading diverges and on which source, without turning the view into a ranking. It is a working proposal from Pico Neural, built to be reused across pages, not a universal method and not a ranking index.
Verifiable limit: the matrix only observes the public pages included in the observation and only the engines named; an API answer is not the same as a person's screen, and Claude and Grok stay out until they are connected.
For the full picture of how Pico Neural sets up observation, see the in-depth page.
How to Read a Site's AI Visibility with How it works
AI Visibility: Measuring the Gap Between Reading and Reporting