Can you see AI Overviews traffic in Google Search Console?

Search Console shows AI Overviews exposure rather than AI Overviews traffic, and the distinction is the whole answer. The generative AI performance report tells you that links to your pages appeared in Google's AI surfaces and how many times. It does not tell you how many people clicked, where your link sat in the answer, or which query triggered it.

Google's definition of the one metric it does report is worth reading literally: "Impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search" (Search Console Help, read 12 November 2026). An impression means a link to you was displayed inside an AI answer. It does not mean your company was named in the text, and it does not mean the answer relied on your page.

Two surfaces are inside the report, AI Overviews and AI Mode, and Google notes that it excludes experiments in Search Labs because those are "still in active development". So a page can be appearing in an AI experience that the report does not count.

What this gives a marketing team is a trend line for exposure on Google, by page, which is genuinely useful and routinely oversold. Treat it as the Google-only half of an exposure measure, pair it with sampling of the engines directly, and keep it out of any sentence that contains the word traffic. The difference between being named and being cited, which this report cannot tell you either, is set out in mention rate against citation rate.

What exactly does the generative AI report show?

The generative AI report shows impressions across four dimensions and nothing else, and knowing its counting rule matters before anyone compares two numbers in it. Google lists the available dimensions as pages, countries, dates and devices, with a search type filter that separates text queries from multimodal ones such as Google Lens and image uploads.

The counting rule catches people out. Google states that the chart aggregates by property, so where several of your results appear in one AI response "they count as a single impression in the chart total", while the table aggregates by page when grouped by page. Chart and table totals can therefore disagree for an honest reason, and a team reconciling them by hand will waste an afternoon.

What you want to know	Generative AI report	Main Search results report	Needs direct engine sampling
Were my links shown in AI Overviews or AI Mode	Yes, as impressions	No	No
How many people clicked from an AI answer	No	Counted inside web search clicks	No
Which query or prompt produced it	No query dimension	Yes, for web search	Yes
Which page was shown	Yes	Yes	Yes
Was my company named in the answer text	No	No	Yes
What ChatGPT, Perplexity or Claude did	No	No	Yes
Pull the data into a dashboard automatically	No API, manual export only	Yes, via the API	Depends on the tool

Export is manual. Google says the report "provides an export button to download both the chart and table data", which is the workaround for everything the next section describes.

Why can't you see clicks from AI Overviews?

Clicks from AI Overviews exist in Search Console, just not in the AI report: they are counted inside the ordinary web search numbers rather than separated out. That is why the generative AI report carries impressions alone, and why nobody can hand you a clean AI Overviews click-through rate from first-party data.

The consequence for reporting is specific. Your organic click total already contains clicks that came from an AI answer, so an organic decline alongside stable AI impressions does not establish that AI answers took your clicks, and an organic rise does not prove the opposite. Both readings are available from the same data, which is how two agencies can present contradictory stories from one Search Console account without either of them lying.

What you can do is compare exposure and clicks as two separate series by page over the same window, and treat a widening gap as a question rather than a conclusion. A page whose AI impressions climb while its organic clicks hold is probably being read in the answer by people who never needed the page. A page whose AI impressions climb while its clicks fall sharply is worth opening in an AI answer yourself to see what the summary now says.

And because the AI clicks are mixed into web search, the only way to attribute a specific enquiry to an assistant remains asking the person. Our guide on tracking leads that come from ChatGPT covers the self-reported attribution route, and whether your direct traffic spike is hidden AI traffic covers the analytics side of the same problem.

Is there an API for the Search Console AI data?

There is no API for it, and that is confirmed in the API reference rather than inferred. The Search Analytics API's type parameter accepts only discover, googleNews, news, image, video and web, with web as the default, and there is no value for AI Overviews, AI Mode or generative AI features (read 12 November 2026). The report is a screen with an export button.

One qualification, because this is the kind of claim that ages badly. The same reference documents a searchAppearance dimension whose full value list lives in the help documentation rather than in the reference, and it suggests querying grouped by searchAppearance to see which values appear in your own data. Nothing in the reference says an AI-feature value exists there, and nothing rules it out, so the honest statement is that the documented type filter has no AI value and a site owner who needs certainty should run that grouped query against their own property.

This has a direct consequence for anyone building AI visibility reporting on Search Console data, including us. Our own demand data is generated from the Search Analytics API, so every position and impression figure in it is a web search number, and none of it can distinguish a page that appeared in an AI answer from one that appeared as a blue link. Anybody presenting a Search Console export as an AI visibility dashboard is presenting web search data with a new label, unless they have manually exported the generative AI report and said so.

The practical approach is to export the report monthly on a fixed date, keep the files, and treat the series as exposure by page. Automating the rest of an AI visibility report is covered in how to automate AI visibility reporting.

Does an impression without a click mean you lost the customer?

An impression without a click is usually not a lost customer, and there is now a measurement of that from inside a search engine. A team of seven authors at Microsoft with one at Simplex AI built a human-grounded session success measure for search experiments where a language model answers directly, labelling whole sessions from interaction data and video.

Their headline number is the one to bring to the next meeting where somebody calls zero-click a crisis: 53.2% of abandoned queries were successful, meaning the user got what they needed without clicking, across 6,904 labelled out-of-sample queries (arXiv 2611.01582, submitted 1 November 2026). The same work reports 6.5% of clicked queries as unsuccessful. A search that ends without a click succeeded more often than not.

Read it with its limits. Microsoft can measure this because it owns the session, which means the finding is about searches on a surface it controls, and the authors have an interest in a metric that makes their own product look better understood. Nothing any vendor sells can reproduce that measurement from outside a search engine, and anyone claiming to has not explained how.

What follows for your own reporting is modest and useful. Impressions in the generative AI report carry more meaning than the absent clicks beside them, because being shown inside an answer that satisfied the reader is a real outcome that your analytics will never record. It also means an impression trend is worth reporting on its own terms rather than as a proxy for sessions.

What does Search Console not tell you about AI visibility?

Search Console tells you nothing about any engine except Google, and in this category that is the larger half of the question. ChatGPT, Perplexity, Claude and Copilot publish no equivalent report to the sites they cite, so everything about them has to be measured by asking them.

The engines also disagree with each other enough that Google data cannot stand in for the rest. Mahe Chen and Nick Koudas, studying 158,099 generative engine answers carrying 468,037 citations, found that ChatGPT only cites pages its own search retrieves, and that those pages overlap little with Google's or Bing's results (arXiv 2610.15033, submitted 10 October 2026). Google exposure and ChatGPT citation are different outcomes with different causes.

Our own category research shows how far apart they can sit. Across 27 buyer queries in three pharmaceutical skid categories, one fabricator was named in the top three on Google, Perplexity or Claude in twelve of the fourteen queries in its lane while scoring at or near zero on ChatGPT. A team reading Google data alone would have concluded that firm was winning the category.

Three further things the report cannot show you. Whether your brand was named in the answer text when no link to you was displayed, which happens often enough to matter. Whether the answer described you accurately. And which phrasings your competitors own, which is the question most buying decisions actually turn on. Those need the engines sampled directly, and how many prompts that takes is covered in how many prompts you need to track AI visibility.

How should you use the report in a monthly report?

Use the generative AI report as one exposure series among several, labelled as Google-only, and never as the headline number for AI visibility. The reporting mistake that follows from it is subtle: a number that is easy to export tends to become the number everyone watches, regardless of what it measures.

A monthly layout that holds up has four parts. Google AI exposure, from this report, by page, with the counting rule noted. Citation and share of voice from direct sampling across the engines you care about, with the number of runs stated. Organic clicks and impressions from the main report, kept separate and never added to the AI figures. And a short list of what changed on the site, so a movement can be lined up against a cause.

Keep input metrics and outcome metrics in different sections. Our own early results page makes that split explicit about our own engagements: an audit score moving from 17 to 68 is an input metric describing how well a site is built for citation, while share of voice is the outcome. Mixing the two produces a report where a week of engine source rotation looks identical to a week of progress.

Then set the cadence against the noise rather than against the calendar. A single reading of any engine cannot be told apart from sampling, which is why answers change every time you ask, and reporting upward on a number that moves on its own is how credibility gets spent. What an executive summary should contain instead is in how to report AI search visibility to your executive team. If you want a baseline across all four engines rather than Google alone, a free visibility assessment provides one with the raw answers and cited sources attached.

What else do people ask about Search Console and AI search?

Marketing and SEO teams ask these five questions about Search Console most often once they open the generative AI report for the first time.