Method
How to Read a Site's AI Visibility with How it works
How it works is a map of separate observations, not a single score. Read each public page engine by engine, keep Gemini and Google AI apart, and mark unconnected engines as pending.
To read a site's AI visibility starting from How it works, treat the page as a map of separate observations, not a single score. For each public page, check what each named AI engine returns — ChatGPT, Perplexity, Gemini, Grok and Google AI — and record the result engine by engine. Take a site with three pages (home, a services page and an article): open How it works, list the pages, compare the answers per engine and fill in a table. The reading rule is simple: every cell is a distinct observation, not an overall grade.
Why How it works is a map, not a score
The How it works page lists the AI engines Pico Neural observes and how each one reads the site. The operational point is that observations cannot be added up: a page can appear in a ChatGPT answer and be missing from Perplexity, or be cited by Gemini but not by Google AI. That last distinction is explicit: Gemini and Google AI Overviews are not the same observation, so they belong in separate columns. Likewise, Grok and Claude are measured only after you connect them, so on a first read without a connection those columns stay empty and should not be read as absence.
The pages × engines table, with a filled-in example
The simplest way to keep observations apart is a two-way table: rows are the site's public pages, columns are the named AI engines. Here is one filled-in row, with values that illustrate the structure only, not coverage:
| Page | ChatGPT | Perplexity | Gemini | Google AI | Grok |
|---|---|---|---|---|---|
| /services | cited in the answer | not cited | cited | not cited | to connect |
The row shows the rule: each cell holds an observation, not a number. If a cell is empty because the engine is not connected, mark it "to connect", not "absent".
Working through How it works
- List the public pages you want to observe and enter them as table rows. Check: every row matches a reachable public URL.
- Open How it works and find the named engines: ChatGPT, Perplexity, Gemini, Grok and Google AI. Check: the table columns match the engines listed on the page, with Gemini and Google AI in separate columns.
- Fill one cell per page and per engine, noting whether the page appears in the observed answer. Check: no cell holds an aggregate judgment; each cell is a single observation.
- Mark unconnected columns (Grok and Claude) as "to connect" rather than "absent". Check: the Grok column shows connection status, not a result.
- Read the table by row and by column: a row shows how one page is observed across engines; a column shows how one engine observes the different pages. Check: no conclusion rests on a single cell.
A Pico Neural working proposal, not a universal standard: use How it works as a log of observations, not a dashboard. In practice, keep the pages × engines table as a working document and update it whenever you add a page or connect an engine. The benefit is that differences between engines stay visible: if a page appears in ChatGPT and not in Perplexity, the table shows it without folding it into an average. A second step is to separate observations made through an API from what a person would see on screen: an API answer is not a person's screen, so note the observation channel next to the cell. Finally, the table is there to ask precise questions — "is this page observed by Gemini?" — instead of asking for an overall judgment on AI visibility. That keeps the reading repeatable and comparable over time, without turning observations into a score.
A limit you can check
The limit is the scope: How it works and this method cover only the named engines — ChatGPT, Perplexity, Gemini, Grok and Google AI — with Grok and Claude measurable only after connection, and with Gemini and Google AI Overviews as distinct observations. Anything outside that list is not read by Pico Neural and should not be added to the table as if it were.
For the full picture of the page and its sections, see the in-depth page.
AI Visibility: Measuring the Gap Between Reading and Reporting