AI visibility
Gemini vs Google AI Overviews: two separate observations
Gemini and Google AI Overviews are not the same observation. RankAI keeps them on separate rows, never adds them together, and shows a missing figure as N/A, never as zero.
No: Gemini and Google AI Overviews are not the same observation. If you are looking at a Gemini answer and, next to it, an AI Overviews box inside a search page, you are looking at two distinct surfaces, not two views of the same object. RankAI keeps them on separate rows and does not add them together. Readers often arrive here after a quick check: they open Gemini, open Google, see a generated box, and assume they have measured the same thing twice. They have not. The non-equivalence is a fact of the register, not a preference: Gemini and Google AI Overviews are not the same observation. Only once that distinction is settled does it make sense to talk about brand citations on ChatGPT, which are yet another row.
Why they are not the same observation
RankAI's public pages name ChatGPT, Perplexity, Gemini, Grok and Google AI. Claude and Grok are measured only after you connect them. Within that list, Gemini and Google AI Overviews appear as separate entries: the register treats them as different observations. Nothing converts one into the other, and one does not stand in for the other.
A second limit concerns the nature of the data. What comes back from an engine through an API is the API response, not the screen a person sees. That holds even when the surface name looks the same: the API response and the human screen are two different things. When a visibility figure is unavailable, RankAI shows it as N/A, never as zero.
The RankAI three-step procedure
What follows is RankAI's own working method, not a universal standard. It exists to prevent the most common mistake: adding two rows that should not be added.
- Read row by row. Open the Gemini view and the Google AI Overviews view as two separate rows. Check: each row carries its own surface label, and neither row carries the other's name.
- Record the discordant result. If one row shows a citation and the other does not, log the case as discordant, not as an error. Check: the discordant row shows the status of the first surface and the status of the second, with no averaging.
- Do not add them up. Do not build a single total that combines Gemini and AI Overviews. Check: the table has no "total" column summing the two rows; if a figure is missing, the cell reads N/A.
A filled-in example of a discordant result
To make the method reusable, here is a two-row table filled in as an example. It is not real data from any brand: it only shows the shape a discordant result should take.
| Surface | Observed status | Notes |
|---|---|---|
| Gemini | Cited | API response recorded as an API response |
| Google AI Overviews | Not cited | No overlap with the Gemini row |
In this example the two rows do not cancel each other out: they do not become "half cited". They stay one citation on Gemini and one non-citation on AI Overviews. If a figure were unavailable on either row, the cell would read N/A, not zero.
The point RankAI wants to make reusable is not the definition of the two surfaces but the discipline of reading them. When a brand asks "are they the same observation?", there is only one working answer: no, and they stay separate even when the engine name appears in both. The frequent temptation is to build a single indicator that adds Gemini and AI Overviews to get a bigger number. That is a methodological error, not a measurement one: adding two distinct observations produces a value that matches neither. Our approach is therefore a table of rows, with a status column per surface and a note flagging any discordance. In this form, a discordant result is not a defect to correct but information to keep. The same goes for brand citations on ChatGPT: they are another row, not a benchmark that absorbs the previous ones. Keeping rows separate lets you see where the brand is cited and where it is not, without one surface covering the other. This is a reading choice, not a universal fact: others may organise the data differently, but this is how we read it at RankAI.
A verifiable limit
The limit lies in the scope of the data: what RankAI shows for a surface is what comes back from that surface through an API, not the screen a person sees. If a visibility figure is unavailable, you will find it as N/A, never as zero. This applies to Gemini and to Google AI Overviews as well, which remain two separate rows.
For the full picture on surfaces and separate rows, see the in-depth page.