Google added a Generative AI performance report to Search Console for site owners who wanted to know how often their pages turn up inside AI Overviews and AI Mode. The impressions it counts were already sitting in your Search Console data before the report existed, folded into your Search totals with nothing to pull them apart. The report gives them their own view.
It sits with the rest of the performance reporting in Search Console, and there is a separate version of it for Discover. Everything below is about the Search version, the one that covers AI Overviews and AI Mode. Google documents it on a single help page, the Generative AI performance report article, and that page is short enough that most of the misreadings in circulation come from filling the gaps with guesses.
Table of Contents
Exactly what the report measures #
One metric: impressions. Google defines an impression here as how many times links to your site were shown to a user in a generative AI feature on Google Search. The generative AI features in question are AI Overviews and AI Mode.
That is the whole metric set: no clicks, no click-through rate, no average position. If your reporting template has columns for those three they will stay empty, and there is no derived version of them hiding elsewhere in the report.
The dimensions you can slice impressions by are:
- Pages , grouped by canonical URL rather than by the exact URL a user saw
- Countries
- Dates , as days, weeks or months, in Pacific Time
- Devices
Queries are not on that list, and that absence shapes everything you can do with this report. You can see that a page was shown inside an AI answer. You cannot see what was asked.
Pacific Time is easy to skim past. If your other reporting runs on UTC or on local time, day-level comparisons drift at the boundaries, and that is enough to make a Monday spike look like a Sunday spike.
The part people misread: these impressions are already in your Search totals #
The generative AI data is also included in the Web search type of the main Performance report. Anyone who is about to present these numbers to a client or an executive should know that before the first slide goes up.
The new report filters impressions you already had. Google did not start counting a new pool of exposure when it appeared, so you cannot add generative AI impressions to your Search impressions and call the result a total. Say Web search shows 400,000 impressions for the month and the generative AI report shows 30,000. Your total is 400,000. The 30,000 tells you how much of that total came from AI Overviews and AI Mode.
Anyone who does add the two will report growth that never happened, and the error compounds as AI features take a larger share of impressions. If you are building AI-era dashboards, the common myths in AI rank tracking are worth a read before you commit to a metric definition.
Limits and caveats #
The report inherits the standard Performance report limits, so nothing here is special-cased for AI:
- The 1,000 row limit applies. A large site will not see its full page list. Filter by country, device or a URL path before you conclude that a page has no AI impressions, because it may simply be below the cut.
- Search Labs experiments are excluded. Whatever Google is testing in Labs does not show up here, so the report covers generally available features only.
- A property needs enough impressions before data appears. Small sites and new properties can see nothing at all, which tells you nothing about the state of the site.
- The rollout has been phased. Availability has not arrived everywhere at the same moment, so a missing report is not the same as missing exposure.
One more caveat catches teams by surprise. Search Console offers a Search generative AI control that lets a site opt out of having its content used in generative AI features. If someone at your company turned that on, you will see the effect of that choice in this report. What the control does is remove your content from the surface. It does nothing to make you appear more favorably in an AI answer.
How to actually use it #
The report is thin, so its value comes from what you compare it against. Four comparisons are worth the effort.
Find pages with generative AI impressions and no matching clicks. Pull the page list from the generative AI report, then pull clicks for the same pages from the main Performance report over the same dates. Pages that rank high on the first list and low on the second are being used as source material without sending traffic. That is not automatically bad, and for a glossary entry it may be exactly what you expected, but it should be a deliberate choice.
Compare page sets over time. Group your pages by template or topic, sum the impressions, and track the groups. Individual URLs move around inside AI answers enough that a single page’s week-to-week change tells you very little, so chasing one URL will waste your afternoon.
Segment by country and device. AI feature coverage differs by market, and a global total can hide a market where your exposure collapsed. Device splits are worth a look when your content has a heavy layout on one form factor.
Watch pages that gained AI impressions after a content change. Note the date you shipped a rewrite, then check the page’s impressions a few weeks later. This is the closest thing the report gives you to a feedback loop, and it is weak evidence on its own: Google changes these features constantly and you have no control group. Treat a gain as a hint worth testing elsewhere, and never as proof.
A workable routine, weekly if you publish often and monthly otherwise:
- Export the generative AI page list and the Web search page list for the same date range.
- Join them on canonical URL and work out, page by page, what share of the impressions came from generative AI features.
- Any page whose share moved by more than a few points against the previous period gets flagged.
- Break your top five flagged pages down by country, in case a single market is doing all the moving.
- Line the flagged pages up against your content change log and your AI answer tracking, then decide whether anything needs action.
- Archive the export. Search Console will not hold this history for you forever, and your own copy is what makes year-over-year comparison possible later.
If you are deciding which of these numbers belongs on a dashboard at all, our rundown of the GEO metrics that actually matter covers how impressions fit alongside mention rate, citations and share of voice.
Five questions this report cannot answer #
- Which query triggered the answer. There is no query dimension.
- Whether your brand was named in the answer text. A link to your site anda mention of your brand are different events, and the report counts only the link.
- Whether a competitor was recommended instead. The report sees your property and nothing else.
- What the answer said about you. Impressions carry no sentiment and no context.
- How many people clicked. No clicks, as above.
Two other data sources cover most of that gap. For clicks, use GA4, with one definition to keep straight: Google’s default channel group documentation puts Google’s AI Overviews and AI Mode inside Organic Search, not in the AI Assistant channel. AI Assistant is for arrivals from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok. So a click from an AI Overview will look like an organic click in GA4, and if you want it attributed differently you have to build that yourself. LLM Pulse reads that traffic through its GA4 integration and reports the human visits arriving from AI assistants.
For the content of the answers, you need prompt-level tracking: running a fixed set of prompts on a schedule and recording what each model actually said, which brands it named, in what order, and which sources it cited. That is the layer that answers the middle three questions. LLM Pulse connects to Search Console, imports your real queries by impressions to build the prompt set from what people are already searching for, and reads Search Console performance data through the app, the REST API and MCP server, so the impressions sit next to the answer-level data instead of in a separate tab you have to remember to open.
Google’s own optimization guidance #
Google publishes an AI features optimization guide in its Search documentation, and it is the closest thing to an official answer on what to do with what the report shows you. Read it with the right expectation. It covers making content accessible and useful to Google’s AI features, and it stops well short of a ranking recipe: it will not tell you why one specific page gained impressions last week.
The practical value is a shared reference for when someone asks whether a proposed change is sanctioned or speculative. Our guide on optimizing for Google AI Overviews works through the same ground with examples.
FAQ #
Does the Generative AI report show clicks?
No. Impressions only, with no CTR and no average position. Clicks from AI Overviews and AI Mode do exist, they just are not in this report, and in GA4 they land in the Organic Search channel.
Is AI Mode included, or only AI Overviews?
Both. Google describes the report as covering generative AI features on Google Search, and names AI Overviews and AI Mode. The dimensions available are pages, countries, dates and devices, so there is no split that separates one feature from the other.
Does this double count my Search impressions?
The report does not double count anything by itself. The double counting happens in your spreadsheet, when you add generative AI impressions to Web search impressions. The generative AI data is already included in the Web search type of the main Performance report, so treat it as a share of your Search total.
How far back does the data go?
The standard Performance report limits apply, which means the same history and row limits as the rest of Search Console, with no separate rule for this report. Check Google’s current help documentation for the exact window, and export regularly if you want history beyond it.
Can I get this data through the API?
Check Google’s Search Console API documentation for the current state of it rather than trusting a blog post, including ours. Do not assume that everything visible in the interface is available programmatically, and confirm what is exposed before you scope a pipeline around it.
Why do I see no data at all?
Four common reasons, in the order worth checking them: your property has not reached enough impressions for data to show, the report has not reached your property yet because the rollout has been phased, the exposure you are thinking of came from a Search Labs experiment and is excluded, or someone enabled the Search generative AI control and opted your content out of these features.