Answers
How to Tell Google, ChatGPT, Perplexity, Gemini and Grok Apart
One page, one test question, one row per system: a practical way to keep Google, ChatGPT, Perplexity, Gemini, Grok and Google AI readings separate instead of blending them into a single number.
You tell them apart by separating what Google indexes from what each system returns, one observation per system. Take a public page on your site, keep the same URL, and look at what Google, ChatGPT, Perplexity, Gemini, Grok and Google AI each do with it: every system gets its own note, with the context in which you observed it. RankAI's public pages name ChatGPT, Perplexity, Gemini, Grok and Google AI; Claude and Grok are measured only after you connect them, and Gemini and Google AI Overviews are not the same observation.
A three-field card for every system
RankAI's working proposal is a simple card, filled in for each page and each system. It is not a universal standard: it is our way of keeping the readings apart. The three fields are: what the system sees (the text or citation it returns), in which context (search, conversational answer, active connection), and connection status (absent, present, to be verified).
A filled-in example
Page X, a public product page.
- Google: we note the page as indexed, with the test query and the date of the observation. Premise: Google indexes public pages; here we record only what we see.
- ChatGPT: we note the answer we got and whether page X is cited or only paraphrased, with the context of the question asked.
- Gemini: we note the conversational answer separately; if a Google AI box also appears, we write it on a distinct line, because Gemini and Google AI Overviews are not the same observation.
- Perplexity, Grok, Google AI: one line each, with context and connection status; for Grok and Claude the field stays “to be measured after connection” until the connection is active.
Three concrete steps, each with a checkable criterion
- Fix the page and the question. Choose one public URL and one test question, always the same for every system. Checkable criterion: the card shows URL, question and date, and two observations made on different days can be compared.
- Write one system per row. For each one, record what is returned and in which context. Checkable criterion: no row contains more than one system, and the Gemini row is kept separate from the Google AI Overviews row.
- Record the connection status. For Claude and Grok write “to be measured after connection” until the connection is active; for the others, state whether the page is cited, paraphrased or absent. Checkable criterion: every row carries an explicit status, not an overall judgement.
Why the readings stay separate
Adding the observations into a single “how they read us” figure produces a number that matches no system. Keeping one row per system lets you see, for example, that a page is cited in a conversational answer but does not appear in a Google AI box, or that Grok cannot yet be measured because the connection is missing. An API response is not a person's screen: when we observe through a connection, we say so in the context field, so the card keeps the two levels apart.
What this piece adds is the discipline of the card: one page, one test question, one row per system, three fields always filled in. In our practice the most useful part is the context field, because it forces you to say whether you were looking at a search, a conversational answer or an active connection. Without that field, two observations made under different conditions end up in the same box and look like a comparison when they are not. The second useful part is the connection status: for Claude and Grok the measurement arrives only after the connection, and writing it down straight away stops an absence from being treated as a result. The third is the separation between Gemini and Google AI Overviews: they are two distinct observations and belong on two rows, even when they appear at the same moment. The card does not say whether a page is “well positioned”: it says what we saw, where and when. It is a working proposal from RankAI, not a universal standard, and it should be adapted to the pages and questions that actually matter to you.
A checkable limit
The limit is this: the card records only what we observed, with the date and the context, and says nothing about how the systems behave under other conditions. If a row lacks context or connection status, the observation cannot be used in the comparison.
For the full method and the FAQ on ChatGPT and Google, see the deep-dive page.