What Search Console now shows of Google's AI Overviews
In June 2026, Google opened a report in Search Console covering the generative AI features of Search. The announcement was published on Google's Search Central blog. The report covers two surfaces named by Google: AI Overviews and AI Mode.
That report counts impressions. An impression means a link to your site was displayed inside one of those two features.
It stops there. No clicks, no click-through rate, no position. The groupings offered are pages, countries, devices and dates. The data cannot be filtered by query.
Two limits are worth knowing before you open the screen. The rollout is gradual, to a subset of site owners. A site receiving few impressions does not see the report at all.
| What you get | What stays missing | What it means for you |
|---|---|---|
| Impressions in AI Overviews and AI Mode. | No click, no click-through rate, no position. | You know you were displayed. You do not know whether it produced a visit. |
| A grouping by page, country, device and date. | No filter by query. | You see which page appears, never which question it was answering. |
| The two generative AI surfaces of Google Search. | ChatGPT, Perplexity and the other assistants. | The report says nothing about most of the answer engines your prospects use. |
| Gradual access, opened site by site. | Availability guaranteed for everyone. | If the screen is absent, that is not a defect of the site. The manual survey takes over. |
A counting rule explains why the usual report was not enough. When the same address appears both in an AI Overview and in the blue links of the same results page, Search Console records one impression, not two. Those impressions are included in the "Web Search" type of the standard performance report, without being distinguishable there.
Put another way: outside that dedicated report, nothing in your data separates an appearance in an AI Overview from a classic result. Connecting your Google accounts is described on the page your Google data.
Nothing equivalent exists for ChatGPT or Perplexity
Google publishes this report because Google indexes the web and knows your pages. You verify ownership of your domain, and you get your data back.
Conversational assistants do not work that way. ChatGPT and Perplexity open no console to site owners. No ownership to verify, no impressions to consult, no export to download.
So you cannot know how many people asked a question your business could have answered. You will know no better how many times it was named.
One thing remains: ask the question yourself, and read the answer. That is exactly what a survey is.
Are ChatGPT's citations real?
The question comes up often, and it is a fair one. Three cases occur, and they are not worth the same.
An answer can show a clickable source. Open it. If the page exists and does say what the assistant attributes to it, the citation is verified.
An answer can name your business with no link at all. That is a brand citation, not a source. It builds recognition, it brings no visit.
An answer can also attribute a statement to a page that does not contain it. It happens. That is why a serious survey notes the domain cited, then opens the link, instead of copying the name displayed.
Why this survey is done by hand, and why it stays free
The fact that this survey is free is not a commercial position. It follows from what precedes: the information exists in no data source you own, and nobody can sell it to you at a better price than direct observation.
The market offers paid tools that track brand visibility in generative AI. They are useful, and it is worth understanding what they sell.
They have no privileged access. None of them receives from OpenAI or Perplexity a data feed that would be denied to you. They ask the questions, as you do, and read the answers.
What they bring is volume and memory. Asking hundreds of questions every week across several engines, keeping the history, flagging a shift: that is real work, and it is legitimately billed.
What they do not bring is data you could not obtain. On ten questions and two engines, once a month, the gap is in the time spent, not in the quality of the observation.
What a survey measures, and what it does not prove
Discipline matters more here than the tool. Three limits keep a survey honest.
The first: an absence of measurement is not a zero. A question you did not ask this month leaves an empty cell. A zero is an observed result, an empty cell is not.
The second: a share of voice is computed, not announced. It only means something on a basket of questions identical from one month to the next. Change one wording and the series starts over.
The third: an AI Overview impression is not a visit. Google's report gives no clicks. Tying a rise in revenue to a citation stays out of reach of this instrument.
Those three limits say the same thing. Being cited by ChatGPT and Perplexity cannot be ordered: it is observed first, then worked on. Visibility in the answers of AI answer engines is measured, it is not promised. A monthly AI citation survey is an instrument, not a target to hit.
Where the two articles that carry the method pick up
This text describes the instrument and its current state. It does not replace the how-to, which is published separately and still holds.
The basket of ten questions to ask, with the page meant to answer each one, is in the AI citation observatory. You will also find there the five technical checks that decide whether your site is readable by an assistant, and the difference in nature between ChatGPT and Perplexity.
The monthly cadence, the grid to keep over time and the rule that separates a signal from a simple variation are in the tracking protocol. That is where the number of runs per question is handled, the one thing that turns a yes or no into a comparable report.
- Check whether the generative AI report is open to you
In Search Console, on the Search performance side. If it is absent, that is not a negative signal: the rollout is partial. Move to the next step without waiting.
- Reuse a basket of questions already published
Do not invent a new one every month. The observatory's ten questions are made to be copied as they stand, brand queries included.
- Ask the questions in a private window
In a session with no account, on each engine, with the exact wording. A signed-in account personalises the answer from your history and makes two surveys incomparable.
- Log the domain cited, then date it
Note the source displayed, open it to check, and write down the date along with the engine version. An undated survey is of no use the following month.
How the audit connects this survey to the rest of your site data
A survey on its own points at a gap without locating it. You know a question is not cited. You do not yet know why.
The audit runs more than ninety checks, several of which bear on how readable the site is to assistants. They verify AI robot access, business identification and the structure of answers on the page.
Crossed with Search Console, those checks separate three situations. No page covers the topic, and one has to be written. A page covers it but a check blocks, and the fix comes before the writing. The page answers and stays readable, but the business is not identifiable, and it is the site's identity that has to be reworked.
Order matters. Strengthening a page while AI robots are blocked amounts to writing for a reader with no access to the text. The detail of those checks is described in the AI citation observatory, and the audit itself is presented on the SEO audit page.
What happens once the survey is in place
A list of uncited questions is not a work plan. It becomes one when every line receives a rank and a reason.
Iris, the workshop's assistant, works on the same data as the audit and Search Console. It first surfaces the questions whose answer already exists on the site, because those are the only ones where a correction produces an effect in the short term. Questions calling for a new page come after, and they cost more.
The demonstration on a real case shows the same question asked without the site's data, then with it. The gap between the two answers is the best way to understand what connecting your Google accounts changes.
What follows depends on who executes. The plan is read and applied in your own publishing tool, with no dependency on the platform. It can also be applied by the team, and the sequence is described in our working method.
Does Search Console show my citations in ChatGPT?
No. The report on performance in generative AI covers only two Google surfaces: AI Overviews and AI Mode. ChatGPT, Perplexity and the other assistants are not in it, and they expose no equivalent report to site owners.
Does the Search Console generative AI report give clicks?
No. It exposes impressions, grouped by page, country, device and date. There is no click, no click-through rate and no position, and the data cannot be filtered by query. Google has indicated that other metrics could follow.
Why does this report not appear in my Search Console?
The rollout is gradual, to a subset of site owners, and a site receiving few impressions does not see it. Its absence says nothing about the quality of your site. The manual survey stays the fallback method.
How do you test your brand on ChatGPT without a paid tool?
Ask the question in a session with no account, note whether your business is named, record the domain cited in its place, then date it. The AI citation observatory publishes the basket of questions to copy as it stands, along with the logging grid.
Trust & E-E-A-T
