Search Console's New Generative AI Report: What It Shows, and What It Won't

For about a year, the honest answer to "how much traffic am I getting from AI Overviews?" was that you couldn't know. Google folded AI Overviews and AI Mode into your ordinary Search Console numbers, under the "Web" search type, with no way to pull them apart. Anyone showing you a precise AI-Overviews-only figure was showing you something the tool did not produce.

That changed on 3 June 2026. Google introduced a dedicated report that separates out how often your site appears inside generative AI features on Search, plus a second one for Discover.

It is genuinely useful, and it is genuinely narrower than most people expect. It carries impressions. That is the whole metric. No clicks, no click-through rate, no queries, no average position. So you can now watch your visibility inside AI answers rise for three months straight and still have no idea whether it sent you a single visitor.

Here's what the report actually contains, how to read it without fooling yourself, and what to pair it with to answer the question you actually care about.

What Changed, Exactly

Before June, AI Overview and AI Mode activity was real but invisible. Google counted a link shown inside an AI Overview as an impression and a click on that link as a click, then reported both inside the "Web" search type alongside every ordinary blue link. There was no filter, no breakout, no way to say "show me only the AI part."

Now there are two new reports sitting in Search Console, one for Search and one for Discover, and they do exactly one thing: they show you your impressions inside generative AI features, on their own.

What has not changed is the rest of your data. Clicks from AI Overviews and AI Mode are still counted inside your Web totals, with no separate view. So the new report tells you where you showed up. Your existing Performance report still mixes what happened next in with everything else.

That gap between the two is the single most important thing to understand about this report, and most of the mistakes below come from missing it.

What the Report Actually Shows

The Search version covers two features, and Google names them explicitly:

The generative AI performance report includes impressions for the following generative AI capabilities on Google Search: AI Overviews

and AI Mode. Anything running as a Search Labs experiment is deliberately left out, because those are still changing.

You get one metric, impressions, which Google defines plainly:

Impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search.

And you can slice that metric four ways:

• Pages, so you can see which of your URLs are actually being surfaced.
• Countries, so you can see where that is happening.
• Devices, split across desktop, tablet and mobile.
• Dates, down to hourly granularity, and up through daily, weekly and monthly.

The Discover version works the same way with one difference: it gives you Pages, Countries and Dates, but not Devices.

The Counting Rule Most People Miss

There is one line in Google's documentation that quietly changes how you should read every number in this report:

If two results from the same site appeared in a generative AI search results feature, they count as a single impression.

Read that again, because it is not how the ordinary Performance report behaves for every surface, and it has a real consequence.

An AI Overview that cites you three times, pulling three different pages from your site into one answer, counts as one impression, not three. So the number in front of you is closer to "how many AI answers included me at all" than "how many times I was mentioned."

That is arguably the more useful measure. It just is not the measure most people assume they are looking at. If you are comparing this number against a third-party AI visibility tool that counts every citation separately, the two will never agree, and neither one is broken.

What It Doesn't Show, and Why That Matters

Four things are missing, and they happen to be the four that would tell you what your AI visibility is worth:

Clicks. You cannot see how many people clicked through to your site from an AI Overview. Those clicks exist and are counted, they are just pooled into your Web total where you cannot separate them.

Click-through rate. Follows from the above. With impressions on one report and clicks on another that does not break them out, there is no way to divide one by the other.

Queries. You cannot see what someone asked to trigger the AI answer you appeared in. This is the one that hurts most, because knowing which questions surface you is how you would decide what to write next.

Average position. Not reported here at all.

So the report answers "am I appearing, on which pages, where, and is it trending up." It cannot answer "is this worth anything," and it cannot answer "what should I do more of." Treat it as a visibility signal, not a performance metric.

This distinction matters more than it sounds, and it is the same one that separates readiness from visibility: impressions inside an AI answer tell you a machine chose to show your link. They do not tell you a person read it, trusted it, or acted on it.

Why You Can't Just Subtract It Out

The first thing most people try is arithmetic: take the AI impressions, subtract them from the Web total, and call what's left "real" search. It does not work, and it is worth seeing why, because the reason explains the whole shape of this report.

Say your Performance report shows 100,000 impressions and 4,000 clicks for the month under the Web search type, and your new generative AI report shows 12,000 impressions.

You cannot conclude that ordinary search gave you 88,000 impressions. The two reports count differently. Your Web total counts each result shown, while the generative AI report collapses multiple results from your site inside one AI answer into a single impression. The 12,000 is not a subset measured the same way, so subtracting it produces a number that corresponds to nothing.

The click side is worse. Those 4,000 clicks include an unknown number that came from AI Overviews and AI Mode, and nothing anywhere tells you how many. You cannot split them, estimate them from the impression ratio, or infer them from a CTR, because the CTR you would need is the thing that isn't reported.

What you can legitimately say from those numbers is narrow: your links appeared in about 12,000 AI answers, here are the pages it happened on, and here is whether that is going up or down. That is the honest ceiling on what this report supports, and stopping there beats inventing the rest.

Two Reports, Not One

It is easy to miss that Discover has its own separate report for generative AI features. If your traffic skews toward Discover, and for publishers and lifestyle sites it often does, the Search report alone will understate your AI footprint badly.

Both count impressions the same way, and both drop Search Labs experiments. The one structural difference is that the Discover version gives you Pages, Countries and Dates but not Devices, which makes sense given Discover is overwhelmingly a mobile surface to begin with.

The practical point is that these are two separate pictures of two separate systems. A page can be quoted constantly in AI Overviews and never surface in Discover, or the reverse. Averaging them together, or checking one and assuming the other looks similar, is how people end up confidently wrong about which of their content is working.

The Control Sitting Next to It

Alongside these reports, Google added a setting under Settings, then Search generative AI, letting you decide whether your content can appear in these features at all.

The wording is worth reading carefully, because it is more reassuring than most people assume:

This control only affects whether your content can appear in certain Search generative AI features; this control isn't used as a ranking or inclusion signal affecting other parts of Search.

So turning it off does not remove you from ordinary Search results and does not count against you in ranking. It removes you from AI Overviews and AI Mode, which also means your generative AI report goes quiet, since there is nothing left to count.

Google points at the report as the way to think about that decision: use it to get a sense of how changing your control might affect traffic. That is sensible advice with an obvious catch. The report shows impressions, not clicks, so it can tell you how much visibility you would be giving up but not how much traffic. You are being asked to price a decision with half the numbers.

Our own read: for most businesses this is not a close call, and leaving it on is the right default. Being absent from the answer does not send the customer to your site instead, it sends them to whoever is still in the answer. The case for turning it off is narrow and mostly applies to publishers whose entire business is the click.

How to Actually Read It

A workflow that gets something useful out of a report this narrow:

Start with Pages, not the total. The headline impressions number moves for reasons you cannot see or control, including how often Google decides to generate an AI answer at all for your topics. The page-level list is the part you can act on, because it tells you which of your content the system considers answer-worthy.

Look for the pages you expected and don't see. This is the most valuable read in the whole report. If your best commercial page never appears while an old blog post does, that is a real signal about which of your pages are legible and self-contained enough to be quoted, and it is worth acting on.

Compare the shape, not the size. Because there is no click data, absolute numbers cannot be turned into value. Trends and relative differences between pages are where the meaning lives.

Check Discover separately. Two reports, two different pictures.

Pair it with your server logs or analytics referrers. This is the practical workaround for the missing clicks. Referral traffic from Google's AI surfaces does show up in your analytics. It will not reconcile neatly with the impressions figure, but it is a real number about real people, and having both is better than having either.

Common Mistakes

Treating impressions as traffic. They are not the same thing and the gap between them is unusually wide here. An AI Overview is designed to answer the question in place. A high impression count with flat referrals is a completely normal and expected outcome, not a bug in your setup.

Trying to reconcile it with a third-party AI visibility tool. They count different things. Google counts one impression per AI answer regardless of how many of your pages appear in it. Most third-party trackers count citations, or track a fixed list of prompts they run themselves. Two different methods, two different numbers, no contradiction.

Subtracting it from your Web totals. You cannot. The report gives you AI impressions; your Web total contains AI impressions and clicks mixed with everything else, in a way you have no key to separate. Any spreadsheet doing this arithmetic is producing a made-up number.

Reading a flat line as a problem you caused. The report is still rolling out to a subset of website owners, and a site needs enough impressions to appear at all. An empty report can simply mean you are not in the rollout yet or are under the threshold.

Trusting the last few days. Google warns that the newest data can be preliminary, still being collected, and may change within hours. Do not draw conclusions from yesterday.

Forgetting the 1,000-row limit. Same cap as the standard Performance report. On a large site the page list you are looking at is a truncated view, not the whole picture.

Attributing a rise to whatever you changed last. If your impressions climbed the month you added structured data, that is a correlation with no controlled test behind it. We have written before about the schema claims made about AI features, and Google's own position is that no special markup is required for them. This report does not give you the isolation you would need to prove otherwise.

What to Do This Week

1. Check whether you have it yet. Open Search Console and look for the generative AI entries near your existing Performance reports. If they are not there, you are not in the rollout, and there is nothing to do but wait.

2. Export the page list from both reports. Search and Discover. Get them side by side.

3. Mark the pages you expected to see and don't. This list is the actual output of the exercise. It is where the report stops being a dashboard and starts being a to-do list.

4. For each missing page, ask whether it can be quoted at all. A page that buries its answer six paragraphs down, or that only exists after JavaScript runs, or that never states plainly what the business is and who runs it, is hard for a system to lift a self-contained answer out of. That is the underlying work behind AI Overview visibility generally, and it applies here unchanged.

5. Set a baseline and leave it alone for a month. Hourly granularity is available and is almost always a trap. Note where you are, come back in four weeks.

6. Pull your AI referral traffic from analytics in the same session. Two imperfect numbers viewed together beat one imperfect number viewed alone.

Appearing in an AI answer starts with a page a machine can actually read and quote. Run a free AI readiness check to see how your site looks to the systems doing the reading, before you go hunting for it in a report.