Method
How to Read RankAI's howItWorks Page
The howItWorks page is a map of what RankAI actually observes. Here is how to read it before you interpret any result, and why Gemini and Google AI Overviews are not the same measurement.
To use RankAI's howItWorks page and understand how Google and AI engines such as ChatGPT, Perplexity, Gemini, Grok and Google AI read the same page in different ways, start with a real page on your site and compare the observations RankAI makes available for each engine it names. The howItWorks page is where that reading is explained: it does not promise a single score, but it shows that the same URL can be treated differently by a search engine and by an AI engine. The starting point is always the same: one page, several readings, and the need to separate what you can verify from what stays outside observation.
What to look for on the howItWorks page
RankAI's public pages name ChatGPT, Perplexity, Gemini, Grok and Google AI. That list is the declared scope: it is not an exhaustive list of every engine out there, but it is what RankAI builds its observation on. Read the howItWorks page as a map of that scope, not as a ranking.
First point to hold onto: Gemini and Google AI Overviews are not the same observation. Treat them as two separate readings, even when they concern the same page. Blurring them gives you a false idea of what RankAI is measuring.
Second point: Claude and Grok are measured only after you connect them. Until the connection is active, there is no observation to interpret. That is an operational limit, not a detail.
A three-step method from RankAI
What follows is RankAI's operational proposal, not a universal standard. Its purpose is to make the comparison repeatable.
- Pick a page and write it down. Take a public URL on your site and record it as a fixed reference. Check: the same URL must stay identical across all later observations, or the comparison does not hold.
- Read the howItWorks page before you look at results. Open the explainer page and check which engines are named and which conditions apply (for example, the connection required for Claude and Grok). Check: you can list, without looking at the data, which engines are in scope and which are not.
- Compare the readings engine by engine, one at a time. For the same URL, look separately at Google, ChatGPT, Perplexity, Gemini, Grok and Google AI. Check: for each engine, write one line describing what was observed and one line saying what could not be observed.
- Keep Gemini and Google AI Overviews apart. Use two separate columns. Check: no row in the Gemini column contains observations attributed to Google AI Overviews, and vice versa.
What stays verifiable and what stays out
The howItWorks page is built to help with exactly this distinction. What RankAI observes and makes available for the named engines, under the stated conditions, is verifiable. Everything outside that scope stays outside observation: engines that are not named, observations that are not yet possible because the connection is missing, and above all the experience of a person looking at a screen.
That last point matters most: an API response is not a person's screen. When you read an observation, ask yourself whether you are looking at system data or at the interface a user would see. They are different things and should not be conflated.
The practical takeaway here is a reading habit: treat the howItWorks page as a scope contract, not a shop window. Before interpreting any result, RankAI suggests writing three lines on a sheet of paper: the engines named, the connection conditions, and the distinction between Gemini and Google AI Overviews. Only then do you look at the observations for your URL. This order prevents the most common mistake, which is attributing to one engine what belongs to another or to a condition that has not yet been met. The search phrase "how is AI visibility measured" often leads people to look for a single number; the howItWorks page instead teaches you to read several separate observations and to state your own limits. It is a working method, not a guarantee of results: RankAI does not promise rankings, and the value lies in the discipline of the comparison.
One limit to keep in mind
The verifiable limit is this: RankAI observes the named engines under the stated conditions, and anything outside the scope — or not yet connected — produces no observation. If an engine is not named on the public pages, you will not find a reading for it on the howItWorks page. That is why you should always go back to the explainer page before drawing conclusions about how a page is read.