AI Search

Can You Opt Out of AI Overviews? What the Toggle Does

Editor8 min read

Yes — since June 2026 there is a genuine control. A Search generative AI setting in Search Console removes a site from AI Overviews, AI Mode, and AI Overviews in Discover, and Google has said it is not used as a ranking signal. The catch is not the ranking risk everyone expected. It is that the control is domain-wide, and you cannot measure what you are giving up before you give it up.

Key takeaways

  • The toggle lives in Search Console → Settings → Search generative AI and covers three surfaces: AI Overviews, AI Mode, and AI Overviews in Discover.
  • Google: the control "won't be used as a ranking signal" and opting out "will not impact its appearance in regular Search."
  • It does not cover the Gemini app, and it is not a model-training opt-out — that is still a robots.txt decision.
  • It is domain-level only. Page-level controls are required by March 2027.
  • Available to a subset of UK site owners first, because it exists to satisfy a CMA order, not as a product initiative.
  • You cannot A/B test this. Search Console reports AI impressions but no clicks, and the control is all-or-nothing.

What exactly does the toggle control?

The single most common mistake is assuming this is one switch over "AI". It is not. It governs three specific Google Search surfaces and leaves everything else alone.

Surface Covered by the toggle?
AI Overviews Yes
AI Mode Yes
AI Overviews in Discover Yes
Ordinary blue-link results No — unaffected
Google Merchant Center / Ads No — unaffected
Gemini app No — content can still surface there
Model training / fine-tuning No — separate control

Two of those rows matter more than the rest. The Gemini app is excluded, so a site that opts out can still be surfaced inside Gemini — you have not left Google's AI estate, only its Search surfaces. And training is a separate question entirely, governed by robots.txt directives. If your reason for opting out is "I don't want my work training a model," this toggle does not do that job; see AI crawlers and robots.txt for the control that does.

Opting out is also not subtle in its effect. Google's wording is that sites which opt out "will not receive traffic or impressions from AI Overviews, AI Mode, or AI Overviews in Google Discover." Not reduced. None.

Why does this control exist now?

Because a regulator ordered it, which explains every one of its limitations.

On 3 June 2026 the UK's Competition and Markets Authority imposed conduct requirements on Google under the digital markets regime, having designated it as holding strategic market status in general search. The CMA called the publisher opt-out a world first. The requirements also force Google to attribute publisher content with clear links in AI-generated results, and to let publishers opt out of their content being used for fine-tuning models.

Google has nine months to implement the full set of changes, and must publish compliance reports every six months for the first year. The CMA said it expects important parts of the controls to arrive well before that deadline — which is why a UK-first, domain-level, partially built version shipped in June rather than a polished global one.

This framing is worth holding onto. A control built to satisfy an order is scoped to the order. The UK-only availability, the domain-level granularity, the March 2027 deadline for page-level controls — none of those are product decisions you can appeal or accelerate.

What does opting out actually cost you?

Here the evidence is genuinely mixed, and the honest answer depends on whether you are counting clicks or attention.

The strongest measurement remains the Pew Research Center analysis of real browsing behaviour from 900 US adults. On searches showing an AI summary, users clicked a traditional result 8% of the time, against 15% on searches without one. Clicks on the links inside the summary were rarer still — about 1% of visits.

That 1% figure is the one to sit with. The citation traffic you forfeit by opting out is, for most sites, close to a rounding error. The 8%-versus-15% gap is the real loss, and opting out does not recover it — those searches still happen, an AI Overview still appears, and it is simply assembled from someone else's page instead of yours. You do not get the pre-AI SERP back by leaving. You get the same SERP with a competitor quoted in it.

That asymmetry is the core of the decision, and it is why "opt out to protect my clicks" is usually reasoning backwards.

Can you measure the decision before making it?

No — and this is the part almost nobody states plainly.

To decide rationally you would need two numbers: how much traffic the AI surfaces currently send you, and what disappears if you leave. Neither is available.

Google launched Search Generative AI performance reports in Search Console on 3 June 2026, the same day as the CMA announcement. They report impressions across AI Overviews, AI Mode, and AI features in Discover, broken down by page, country, device and date. They do not report clicks, CTR, or queries. Data starts 18 May 2026, with no historical backfill.

So you can see that you appeared. You cannot see whether anyone came.

The analytics side does not fill the gap either. GA4's native AI Assistant channel, added on 13 May 2026, separates chatbot referrals into their own channel with medium = ai-assistant. Useful — but it covers third-party assistants, and Google named ChatGPT, Gemini and Claude at launch. Notably, Perplexity is absent from the definition and still lands in Referral. More importantly for this decision, the channel excludes Google's own AI Overviews and AI Mode, whose clicks arrive as ordinary google / organic and are indistinguishable from any other organic click.

Put those together and the position is:

  • AI impressions: visible, from May 2026, no clicks attached.
  • AI clicks: commingled with organic, in both Search Console and GA4.
  • The control: domain-wide, so no holdout section to compare against.
  • Once opted out: impressions go to zero, destroying the baseline.

There is no experiment available here. You cannot run it on a subfolder, you cannot isolate the clicks, and you cannot compare against yourself afterwards. Anyone presenting you with a confident before-and-after number for an AI Overviews opt-out is estimating, and should say so.

Is AI Mode traffic really untrackable?

You will still read that AI Mode links carry a noreferrer attribute that makes the traffic permanently invisible. That was true briefly, and it is out of date.

When AI Mode rolled out, its links did carry noreferrer, which stripped the referrer and pushed those visits into Direct. Google's John Mueller said it looked like a bug on Google's side, the attribute was removed within days, and AI Mode clicks resumed reporting as organic.

The durable problem is different and larger: a substantial share of AI referrals — commonly reported as well over a third, mostly from mobile apps — arrive with no referrer at all and land in Direct. That is not a Google policy, it is how app webviews behave, and no channel grouping fixes it. Treat any "AI traffic" number as a floor rather than a measurement.

Who should actually consider opting out?

For the large majority of sites, the answer is no, and it is not close. If search is a meaningful acquisition channel and you have no alternative distribution, leaving the surface where answers are increasingly assembled is a structural loss with no recovery mechanism.

The cases where it is defensible are narrow and mostly not about traffic:

  • Your content is the product. Subscription archives, paid research, licensed data — if grounding an answer substitutes for the purchase, extraction is the harm, and the toggle is a reasonable commercial lever.
  • You are negotiating. The CMA framed this as bargaining power for publishers. For a rights-holder with leverage, an enforceable opt-out is a position, not just a setting.
  • Attribution genuinely fails you. If you can show your material summarised without a usable link, that is the specific harm the control was built for.

What is not a good reason: opting out to punish Google, or to "protect" clicks that the Pew data says go to a competitor's citation instead of returning to you. And if your goal is the opposite — to be the page that gets quoted — the work is unchanged and sits elsewhere: what the citation research actually supports, self-contained passages, and the credibility signals that make a source worth naming. It is emphatically not a file you add to your root directory; see does llms.txt do anything.

What should you do if you don't have the toggle yet?

Most sites reading this are outside the UK subset and have no decision to make today. That is a reprieve, and it is worth using:

  1. Start the impression baseline now. The generative AI report has no backfill, so every week you wait is data you never get. If you have access, it is already accruing from 18 May 2026.
  2. Separate your two AI decisions. Grounding (this toggle) and training (robots.txt) are independent. Decide each on its own merits — most sites want to be in answers and are far more ambivalent about training.
  3. Don't pre-commit. Page-level controls arrive by March 2027. A domain-wide switch is a much worse instrument than the one you will be offered later.
  4. Check the attribution requirement instead. The order compels clear links to publisher content. If you are being summarised without one, that is now an enforceable complaint rather than a grievance.

If you are new to how these answers get assembled in the first place, what AI Overviews are and how they work covers the retrieval mechanism this control switches off.

The uncomfortable summary is that Google has offered publishers a real, legally-compelled choice while withholding the measurement that would let anyone make it well. Both of those facts are true at once, and the second one is the one to plan around.

Sources: CMA — secures fairer deal for publishers (3 June 2026) · Google Search Central — Search Generative AI performance reports · Pew Research Center — Do people click on links in Google AI summaries? · Search Engine Journal — Google Analytics adds AI Assistant as default channel group

FAQ

Frequently asked questions

Can you opt out of AI Overviews?

Yes. Since June 2026 Search Console has a Search generative AI setting that removes a site from AI Overviews, AI Mode, and AI Overviews in Discover. Google states the control is not used as a ranking signal and does not affect a site's appearance in regular Search. It is currently available to a subset of UK site owners, with a global rollout planned but not dated.

Does opting out of AI Overviews hurt my rankings?

No. Google has said the control will not be used as a ranking signal and that using it will not impact a site's appearance in regular Search. Ordinary blue-link results are unaffected. What you lose is traffic and impressions from the generative AI surfaces themselves, which is the intended effect rather than a penalty.

Does the opt-out stop my content being used to train AI models?

No. The Search Console toggle governs whether your content grounds answers in AI Overviews, AI Mode, and AI Overviews in Discover. Model training is a separate decision controlled through robots.txt directives such as Google-Extended. The two choices are independent and should be made separately.

Can I opt out of AI Overviews for only some pages?

Not yet. The control operates at domain or property level, so it applies to the whole site. Page-level grounding controls are required by March 2027 under the CMA order. Until then you cannot exclude one section and keep another, which also means you cannot run a partial test.

B

Written by

BlogsPublication Admin

Editor

BlogsPublication reporting is guided by our editorial standards.

The newsletter

Good writing, once a week.

Our best essays and reporting, delivered to your inbox. No noise, unsubscribe anytime.

Comments

Sign in to join the discussion.

Loading comments…

Keep reading