Google finished rolling out the generative AI performance report in Search Console to every property worldwide on August 31, 2026, closing a phased release that began on June 3 with a subset of UK site owners. The report tells you how often links to your site were shown in AI Mode and AI Overviews, broken down by pages, countries, devices and dates. It does not tell you how often anyone clicked. Shipping in the same release is a control that withholds your content from those surfaces entirely, which means Google has handed publishers a switch and a dataset with the most decision-relevant number removed from it. This piece covers what the report contains, what the control does and does not do, why it exists at all, the measurement problem the pairing creates, what independent research says about the clicks you cannot see, what falls outside the control’s scope, and how to approach the decision.
The short version: do not flip the switch yet. The impressions data is worth collecting for a few months precisely because it is the only Google-supplied signal you will have, and the control is reversible in days, so there is no first-mover advantage in opting out early and a real cost to doing it blind.
What the generative AI performance report actually shows
Impressions, and dimensions to slice them by.
Search Console’s help documentation defines the metric as how many times links to your site were shown to a user in a generative AI feature, with pages, countries, devices and dates available as breakdowns. That is the complete list.
The absences are more instructive. Search Engine Journal cataloged them in August: no queries, no clicks, no click-through rate, no average position, no citation placement, no indication of which passage supported the answer, and no conversion or revenue data. Google has said it is working with site owners to determine which additional metrics would be useful. We could not find any published explanation of why clicks were excluded in the first place, from Google’s documentation or from anyone reporting on it.
So the report answers "were we shown" and refuses "did it do anything." For a surface whose entire controversy is whether appearing in it produces traffic, that is a conspicuous shape for a dataset to have.
The control beside it, and what it actually does
Several things about this control are commonly reported wrongly, so they are worth stating precisely.
It is a Search Console setting, found under Settings and labeled Search generative AI. It is not a robots.txt directive and not a new meta tag token, which distinguishes it from Google-Extended, the older and separate mechanism governing model training.
It withholds your content from AI Overviews, AI Mode and generative AI features in Discover. It does not stop crawling or indexing, and Google’s documentation is explicit that your content may still be used to power Search generally.
It takes effect quickly. Google’s help text says content will be excluded within one to two days after the control goes live, with some content taking longer because of caching. Reversal works the same way.
And it is not a ranking signal. Google’s own documentation states that the control "isn’t used as a ranking or inclusion signal affecting other parts of Search." That sentence is doing more work than it appears to, for reasons in the next section.
Granularity is currently property and directory level. Page-level control is coming but is not due until roughly March 2027.
Why this control exists
The proximate cause is a regulatory one, and knowing it changes how much weight to put on Google’s assurances.
The UK Competition and Markets Authority issued a Publisher Conduct Requirement on June 3, 2026, under Google’s Strategic Market Status designation. Paragraph 3 requires that "Google shall provide publishers with effective controls to withhold their Search Content from being used in generative AI services and features," and paragraph 4 requires those controls to work at both directory level and page level.
Paragraph 9 is the one publishers should read. It requires that "Google shall not seek to prevent publishers from, or penalise them for, using the controls described in paragraph 3, for example by downranking their Search Content, or otherwise disadvantaging them, relative to opted-in Search Content, in general search outside of search generative AI features."
That is the same promise Google’s help documentation makes about ranking, except it is enforceable by a regulator rather than being a statement on a support page. For anyone weighing whether opting out will quietly cost them ordinary search visibility, the distinction is the whole answer.
Two boundaries worth holding onto. The conduct requirement is UK-scoped, and Google chose to ship one global control rather than a UK-only feature, so the worldwide rollout is Google’s decision rather than something the order compelled. And the August 31 completion actually preceded Google’s own compliance deadline, which falls around December 2026 for most provisions.
The measurement problem the pairing creates
Here is the decision a publisher is now being asked to make.
You can see that you appeared in AI surfaces 40,000 times last month. You cannot see how many of those produced a visit, what those visitors did, or what they were worth. You also cannot separate AI-surface traffic from ordinary organic traffic in your own analytics, because referrals from both arrive looking the same.
So the question "is appearing in AI Overviews earning or costing us" has no answer inside Google’s tooling. What you have is one number on one side of a ledger.
We made a related argument in AI citations and ranking decoupled, where the thing being measured and the thing being optimized had come apart. This is the same problem with an action attached: a control that does something consequential, sitting beside a report that cannot tell you whether to use it.
The saving grace is reversibility. One to two days out, one to two days back. That makes this a testable change rather than a commitment, which is not how most coverage has framed it.
What independent research says about the clicks you cannot see
Since Google will not supply the click data, three outside measurements are worth knowing, and they are not of equal weight.
Pew Research Center, July 2025, is the strongest methodologically. Working with 900 US adults on its KnowledgePanel and analyzing 68,879 unique Google searches, Pew found users clicked a traditional result in 8% of visits where an AI summary appeared, against 15% where one did not, and clicked a link inside the AI summary itself in 1% of visits. Session abandonment ran 26% with an AI summary against 16% without.
Agarwal and Sen, a randomized field experiment run through a custom browser extension with 1,065 US desktop participants between January and February 2026, found outbound organic clicks fell 39.8% when an AI Overview was shown, with zero-click searches rising 34.5%.
Ahrefs published a figure of a 58% CTR reduction for top-ranking pages in February 2026. We could not recover the sample size or methodology from secondary coverage, and Ahrefs sells SEO tooling, so treat that number as vendor research until you can read the original.
Two independent studies, one academic and one nonprofit, both find substantial click suppression. Neither tells you what happens to your site specifically, which is the number Google is not giving you.
What the control does not cover
Gemini. The control governs Search surfaces, and the app is a different product under a different mechanism.
Google-Extended, the robots token, governs whether your content trains and grounds Gemini. It does not determine what appears in Search. So a publisher who opts out of AI Overviews can still find their content surfacing in Gemini responses. If your concern is model training rather than search appearance, this control is not the one you want, and if it is both, you need both.
We should be honest about the sourcing here: Google’s own documentation is silent on Gemini rather than affirmatively excluding it, so this reading comes from trade coverage tracing the CMA order’s scope to Search products.
What to actually do
Turn on the report and leave the control alone for now. You need a baseline, and you currently have none.
Log impressions monthly by page. The report has no historical backfill, so the record starts when you start keeping it, and in six months the trend will be the only evidence you have about whether AI surfaces are growing or shrinking as a source of exposure.
Stop trying to isolate AI referral traffic in analytics. You cannot, and the measurement approaches that do work are manual rather than reported.
If you are seriously considering opting out, do it as an experiment on a directory rather than a property, once page-level and directory-level control is available to you, and watch total organic sessions rather than AI impressions for the effect. And note that the ranking protection you are relying on is a regulatory requirement, not only a Google statement.
Finally, separate the two decisions. Whether to appear in AI Overviews and whether to allow model training are governed by different mechanisms and deserve different answers. Conflating them is the most common error we see in crawler access decisions.
Frequently Asked Questions
What is the generative AI performance report?
A Search Console report showing how often links to your site were displayed in Google’s generative AI features, specifically AI Mode and AI Overviews. It provides impressions broken down by pages, countries, devices and dates. It completed a global rollout on August 31, 2026 after a phased release that began June 3.
Does it show clicks?
No. Impressions are the only metric. Clicks, click-through rate, queries, average position, citation placement and conversion data are all absent. Google has said it is working with site owners on which additional metrics would be useful, but has published no explanation for why clicks were excluded.
Where is the opt-out control?
In Search Console under Settings, labeled Search generative AI. It is a setting in the interface rather than a robots.txt directive or meta tag, which makes it different from Google-Extended.
Does opting out stop Google crawling my site?
No. It withholds your content from appearing in AI Overviews, AI Mode and generative AI features in Discover. Crawling and indexing continue, and Google states your content may still be used to power Search generally.
Will opting out hurt my rankings?
Google’s documentation states the control is not used as a ranking or inclusion signal affecting other parts of Search. Separately, the UK CMA’s Publisher Conduct Requirement of June 3, 2026 prohibits Google from penalizing publishers for using the control, including by downranking them relative to opted-in content. The assurance exists in both a support page and an enforceable requirement.
How quickly does the control take effect, and can I undo it?
Google’s help text says content will be excluded within one to two days after the control goes live, with some content taking longer because of caching. It is reversible on a similar timescale, which makes opting out a test rather than a commitment.
Can I opt out a single page?
Not yet in full. The control currently operates at property and directory level. Page-level granularity is required under the CMA timeline but is not due until roughly March 2027.
Does this stop my content appearing in Gemini?
No. The control governs Search surfaces. Gemini is addressed by Google-Extended, a separate robots token that governs training and grounding but does not determine what appears in Search. Content withheld from AI Overviews can still surface in Gemini responses.
Why did Google build this?
The UK Competition and Markets Authority required it. Its Publisher Conduct Requirement of June 3, 2026, issued under Google’s Strategic Market Status designation, requires effective controls for publishers to withhold content from generative AI features. The requirement is UK-scoped; shipping the control worldwide was Google’s own choice.
What do we actually know about AI Overviews and clicks?
Pew Research Center found clicks on traditional results at 8% of visits with an AI summary present against 15% without, across 68,879 searches by 900 US adults. A randomized experiment by Agarwal and Sen with 1,065 participants found outbound organic clicks fell 39.8% when an AI Overview appeared. An Ahrefs figure of 58% circulates widely but its methodology is not recoverable from secondary coverage and it is vendor research.
Compliance note. The report’s metrics and the control’s behavior are taken from Search Console help documentation. The August 31 completion date and the June 3 phased-rollout start come from Search Engine Land, Search Engine Journal and Search Engine Roundtable quoting Google’s Search Central post, which we could not fetch directly, so those two dates rest on trade press quotation rather than a verified reading of the primary post. The conduct requirement text is quoted from the CMA’s Publisher CR Notice of June 3, 2026. Pew and the Agarwal and Sen experiment are independent research with published methodology; the Ahrefs figure is vendor research whose methodology we could not verify. The reading that Gemini falls outside this control comes from trade coverage, not from a Google statement naming Gemini.
