Google Shipped the AI Opt-Out Without the Data to Use It

What Google actually shipped

Two separate things arrived, and they get conflated constantly.

The first is the reporting. On 3 June 2026 Google announced Search Generative AI performance reports in Search Console, authored by Hillel Maoz, Search Ecosystem Engineering Manager, and Moshe Samet, Product Manager Lead for Search Console. The post lists exactly five things the reports show:

Field	What Google says it tells you
Impressions	How often URLs from your site appeared in generative AI features in Search and Discover
Pages	Which URLs appeared within AI features
Countries	Visibility on a country basis
Devices	Which devices people used (Search results only)
Dates	Hourly, daily, weekly and monthly granularity

Clicks are not in that list. Neither is click-through rate, position, or query. We read the announcement in full: those four words do not appear anywhere in it. Google does say it is “continuing to work with website owners to understand what insights and data would be most helpful,” including “adding additional metrics over time,” and that the reports are rolling out “to a subset of websites” first. Both caveats are Google’s own and both are fair. Neither changes what the report contains today.

The second thing is the opt-out control itself, documented in Google’s Search Console help centre. Per SEJ’s reporting, Google said it would begin respecting changes on 17 June, with content dropping out one to two days after processing. We could not find that date on Google’s own help page, so treat it as SEJ’s characterisation rather than a Google-stated commitment.

Why the control exists at all

This was not a spontaneous act of publisher goodwill. On 3 June 2026 — the same day as the Search Console announcement — the UK’s Competition and Markets Authority imposed a publisher conduct requirement on Google. The requirement obliges Google to, among other things:

• “provide publishers with effective controls over the use of their search content in generative AI”
• “publish clear, comprehensible and user-friendly information explaining how publishers’ search content is used by Google in its generative AI”
• “provide publishers with clear and detailed metrics on user engagement with their search content in search generative AI features”
• “take reasonable steps to ensure that search content is attributed clearly and accurately in general search”

The CMA’s accompanying announcement called the opt-out a world first and set a nine-month implementation deadline with compliance reporting every six months in the first year.

Read those two documents side by side and the shape of the compliance becomes visible. The controls requirement has been met. The metrics requirement — “clear and detailed metrics on user engagement” — has been met with an impressions counter.

One caution on how this is being reported elsewhere. Several write-ups state that the CMA specifically demanded impressions, click-throughs and CTR. The conduct requirement page we fetched does not name those three metrics; it points to separate interpretative-notes documents. SEJ attributes the specific metric list and a compliance timeline — click data in December, page-level controls by March 2027 — to the requirement. We are repeating that as SEJ’s reading, because we could not confirm it on the CMA summary page itself.

An impression is not a visit, and everyone knows it

The reason this matters is arithmetic, not ideology.

If you know your content appeared in an AI Overview 100,000 times, you know nothing about whether that was good or bad for you. It could mean 100,000 people saw your brand name attached to an authoritative answer. It could mean 100,000 people got what they needed and never came. Those are opposite business outcomes and the report renders them identically.

Now put the opt-out next to it. The decision Google is offering is: stay in and accept whatever that mix is, or leave and lose the impressions entirely. Making that decision rationally requires knowing the click yield of an AI impression. That is precisely the number that is absent.

There is a second-order cost too. A publisher who opts out loses the impressions data along with the placement, which means they also lose the ability to measure what leaving cost them. The experiment is not reversible in any measurable sense.

The claim nobody outside Google can check

On 22 July 2026, in Alphabet’s Q2 earnings message, Google told investors that AI Mode had “surpassed 1 billion monthly active users” since its global expansion last October, and added: “just like AI Overviews, AI Mode is driving an incremental increase in Search queries overall, and we’re now sending billions of clicks to websites every week through AI features in Search.”

We are not calling that claim false. We have no basis to. We are pointing out something narrower and, we think, harder to argue with: it is unfalsifiable at the level of any individual site, because the tool Google built for site-level AI visibility reporting does not report clicks.

A publisher cannot open Search Console, look at their share of those billions, and confirm or contradict the statement. Aggregate claims that cannot be decomposed by the parties they describe are not evidence in the ordinary sense. They are assertions with a large number in them.

What this does not tell you

We want to be exact about the limits of our own case here.

The absence of clicks in a June launch post is not proof that clicks will never arrive; Google explicitly flagged additional metrics as a possibility. The reports are also still limited to a subset of sites, so a given publisher may not have the report at all yet. And we have not tested the opt-out ourselves or measured any site’s before-and-after traffic, so nothing here quantifies the actual cost of opting out.

There is also a widely circulated figure we deliberately left out of the argument. NewzDash reported on 28 July that roughly 15.5% of tracked US trending-news results and 17.46% in the UK now place Top Stories inside the AI Overview, which would mean opting out costs a news publisher a second placement as well. It is a directionally interesting number from a vendor that sells news-visibility tracking, and per SEJ’s own reporting NewzDash has not published sample sizes or collection dates alongside it. We are not building on it.

What would settle this

One thing: clicks and click-through rate in the generative AI performance report, at page level, on the same dates as the impressions. That is a product decision, not a technical impossibility, and Google already ships exactly those fields in the ordinary Search performance report.

Until then, the honest description of the current state is that publishers have been handed a switch and no gauge. Any advice you read this month about whether to flip it — including advice arguing confidently in either direction — is being given by someone who cannot see the number that would decide it either.