The setting is now sitting in your Search Console account, and the reason it is hard to act on has nothing to do with nerve. Flipping it is a one-way trade you cannot value: you know exactly what you surrender, because Google states plainly that opted-out sites receive no traffic and no impressions from its generative AI features, and you do not know what that traffic is currently worth, because the reporting shipped alongside the control counts impressions and stops there. A publisher deciding whether to leave AI search is being asked to price an asset from one side of the ledger. That is the real content of the choice, and it is worth understanding before deciding whether the toggle is a win or a decoy.
What the Search generative AI control actually does
The control is a property-level setting inside Search Console. Switching it off removes your site from the pool of pages Google uses to ground its generative answers and stops your links appearing in them. It covers three surfaces: AI Overviews, AI Mode, and AI Overviews inside Google Discover.
Google's documentation is explicit about three consequences. Sites that opt out will not receive traffic or impressions from those features. The setting is not used as a ranking signal for search results outside the generative AI surfaces, so ordinary organic ranking is unaffected. And it has no bearing on AI model training, which remains governed by the separate Google-Extended token.
The rollout ran from the first week of June 2026, starting with UK properties, and finished globally at the end of August. Search Engine Land's Barry Schwartz reported on 31 August 2026 that Google had updated its blog post, its developer documentation and its Search Console help page with the same line: as of 31 August 2026, the control is available to all websites worldwide. The AI performance reports, which break out impressions in generative AI features by page, country, device and date, went global at the same time.
Three things the control is not. It is not page-level in practice for most publishers, despite the CMA requiring the opt-out to be effective at both site-wide and page level. It is not separable by surface, so you cannot leave AI Overviews while staying in Discover. And it is not an instruction to Google's crawler, so your pages continue to be fetched and indexed exactly as before.
Why the control exists at all
It exists because a regulator ordered it. On 3 June 2026 the UK Competition and Markets Authority imposed a conduct requirement on Google Search under the digital markets competition regime, following its October 2025 decision to designate Google with strategic market status in general search. The CMA called it a world first, and it was: the first time any major regulator has forced a search engine to let publishers withhold their content from AI-generated answers without losing ordinary search visibility.
Sarah Cardell, the CMA's chief executive, framed the purpose in bargaining terms rather than technical ones, saying that with features like AI Overviews rapidly reshaping online search it is crucial that content publishers have appropriate bargaining power over how their content is used.
The order goes further than the toggle. Google must not downrank or otherwise disadvantage sites that opt out. It must offer a separate opt-out from fine-tuning. It must attribute publisher sources clearly in AI summaries and explain how it does so. It must publish engagement metrics for content appearing in AI features, separately from general search. And it may not acquire publisher content through third-party scrapers to route around an opt-out. Google has nine months from the order to implement, with compliance reports every six months in the first year, and non-compliance carries penalties of up to ten per cent of global turnover.
Google published its own post the same day. Mrinalini Loew, general manager of Google Search Ecosystem, said the company would test a control letting website owners manage how their links and content appear in AI search features, starting with a subset of UK site owners before going global, and confirmed the control would not be used as a ranking signal outside the generative AI features. That last clause is what makes the decision thinkable at all. Before it, as Foxglove co-executive director Rosa Curling put it, the only way to stop Google using your work was to opt out of being visible in Google search entirely.
The measurement problem that makes the toggle unusable
Here is the asymmetry. Google's new AI performance report gives you impressions in generative AI features. It does not give you clicks from them.
Stuart Forrest, former global audience development director at Bauer Media Group, put the gap precisely to Digiday: "They've broken out search impressions about AI features, but they haven't given us search clicks for AI features - that doesn't exist." Without that number, a publisher cannot tell whether AI Overviews are sending readers or absorbing them, which is the only question that decides whether opting out costs anything.
Forrest also identified why the default matters more than the switch. "If this were opt-in, Google would have to prove to us why we should let them use our content," he said. "Instead, they keep us in by default and don't give us the click data we need to know if AI search is actually hurting our business."
A head of SEO at a large lifestyle publisher, speaking anonymously to the same reporters, offered the commercial reading of the omission: Google's reluctance to release AI clickthrough data is likely because the numbers would undercut its own position that AI Overviews are good for publishers. Impressions are the flattering metric. Clicks are the contested one.
Paul Bannister, chief strategy officer at Raptive, called the opt-out nonsensical and doubted anyone would use it. "Giving publishers a front-end toggle for AI Overviews or AI Mode is not real control; it is Google offering publishers a light switch while keeping the power plant running," he said. His objection is that publishers asked for two things, a guaranteed way to stop their content training Google's models and a way to stop Google's AI systems drawing on their sites when generating answers, and were given a control over where answers are displayed instead.
Thomas Lue Lytzen, director of ad sales and tech at the Danish publisher Ekstra Bladet, described the practical result: judging the consequences of opting out against staying in is impossible, so many publishers will stay because they fear the unknown. Phil Andraos, general manager of Reuters Digital, said the same thing in one line: "There's not enough data. If there was, we could make that decision very quickly."
What the evidence says AI Overviews actually cost
The best available answer does not come from Google. It comes from a randomised field experiment by Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon University's Heinz College, posted to SSRN on 3 April 2026 and last revised on 17 June 2026. It is the first causal evidence on the question rather than the correlational before-and-after comparisons that Google has spent two years disputing.
The design matters, because it removes the objection Google usually makes. The researchers built a browser extension and randomly assigned real users either to a standard results page or to a version with AI Overviews hidden. The analytic sample was 1,065 United States desktop Chrome users recruited through Prolific, generating 68,089 observed searches, with AI Overviews triggering on roughly 41 per cent of queries.
The headline findings: hiding AI Overviews raised outbound organic clicks by 39.8 per cent and cut zero-click searches by 34.5 per cent. On queries where an overview would have appeared, outbound clicks rose from 0.37 to 0.62 per search, and the probability of a search ending without any click fell from 0.73 to 0.54. The effect was concentrated where the overview sat at the top of the page, which was more than 87 per cent of the time, producing an 88 per cent relative increase in outbound clicks there and no measurable effect when the overview appeared lower. Informational queries, 71 per cent of the sample, drove the whole result.
Two secondary findings deserve more attention than they have had. First, clicks to Google's own properties did not move, and neither did sponsored clicks, which means AI Overviews substitute specifically for visits to publisher websites rather than redistributing attention around the page. Second, the study directly tested the "higher-quality clicks" defence: back-button navigation, bounce rate and total time on page showed no meaningful difference between the groups. Nor did user satisfaction, information quality or ease of finding information, all of which the authors report as precisely estimated nulls. On this evidence AI Overviews cost publishers clicks and buy users nothing measurable in return.
Set that against the scale. Google has said AI Overviews reaches more than two billion monthly users, and AI Mode passed a billion monthly users during 2026. And set it against what publishers are already reporting: People Inc. disclosed in earnings that Google sessions from the first quarter of 2024 to the first quarter of 2026 represented a total loss of 800 million visits. A business that size can absorb it. Most cannot.
So who should actually consider opting out?
The honest answer is that the decision splits by revenue model, not by principle.
Subscription-heavy, niche and brand-led publishers have the most room. If your business runs on paid relationships rather than impression volume, the marginal search referral is worth less to you than the perceived value of your archive, and an opt-out is a defensible experiment rather than a gamble. Some of these titles were always the least well served by AI answers, because a summary of a specialist analysis is a poor substitute for the analysis.
Advertising-funded publishers have almost none. Every referral that disappears removes an ad request, and the supply loss compounds through a programmatic auction that was built for abundance. For these businesses a meaningful hit to Google traffic is not a strategic setback, it is a hole in the quarter.
Anyone in a Google commercial programme should read their contract first. Press Gazette reported claims, originating with the New York Post in June 2026, that Google had indicated publishers opting out of AI Overviews could be excluded from its pilot AI partnership deals. Treat that as reported rather than confirmed, but treat it as a real term to check, because a control you have contractually agreed not to use is not a control.
And everyone should consider the bundling risk. Chris Dicker, chief executive of Candr Media Group, warned that Google is already grouping AI Mode and AI Overviews together rather than letting publishers govern each separately, and that Discover could be folded into the same framework. Many publishers now depend on Discover for a large share of traffic. If it ends up inside the same switch, the opt-out becomes unusable for a second reason.
There is a collective argument on the other side, and it is not weak. Eugene Huang, a senior strategy adviser at Prorata and a former Google employee, told Press Gazette's AI Summit on 30 June 2026 that it would be a travesty if no publisher decided opting out was an action worth taking. Forrest made the same point in reverse: a remedy that almost nobody dares to use is a gift to Google, which can then cite low adoption as evidence that publishers are content with AI search. He also fears the UK settlement becomes the template regulators elsewhere copy, importing an opt-out-and-partial-data standard into Brussels before anything stronger is written.
What the toggle does not fix
It does not restore the click. Jason Kint, chief executive of Digital Content Next, has argued that the remedy touches only future use of publisher content in AI search, not the material already taken to train models. Sajeeda Merali, chief executive of the Professional Publishers Association, called the proposal a major step forward while noting that AI Overviews still replace clicks in many contexts and that without a clear licensing model the commercial imbalance remains.
It does not fix the crawler. What publishers actually asked for was structural: separate crawlers for search and for AI, so that consent could be given to one and withheld from the other. Owen Meredith, chief executive of the News Media Association, told Digiday the association was sceptical of a remedy that relies on Google to separate data after it has been scraped, describing that as behavioural where the clean solution is structural. Bannister argued Google could do it quickly if it wanted to and does not because keeping the two joined is a competitive advantage over OpenAI and others.
It does not settle payment. The CMA postponed any formal requirement that Google negotiate fair payment for at least twelve months, which Meredith called disappointing. Will Hayter, the CMA's executive director for digital markets, has said the opt-in versus opt-out argument is being overblown and that it is for Google to propose how it complies. The European Commission is separately surveying publishers on whether the opt-out is of any practical use to them.
And it does not follow the reader. Whether you are in the answer or out of it, the reader who gets what they needed from a generated summary does not arrive on your page, and no page of yours renders, and no ad call is made. That is the underlying condition the toggle sits on top of rather than changes.
The decision this really is
Framed as a traffic question, opting out is close to unanswerable on the data Google supplies. Framed as a control question, it is clearer. The toggle is the first thing publishers own in this market: a binary, enforceable statement about whether their work grounds someone else's answer. Its value is not in switching it off today. Its value is that it exists, and that it converts a taking into a negotiation.
That is why the more useful reading of the CMA remedy is as leverage rather than as a remedy. A right nobody exercises has no price. A right a few credible publishers exercise, publicly and with their reasoning shown, sets one.
Meanwhile the commercial question moves elsewhere. Opting out removes you from the answer; staying in leaves you in an answer you do not monetise. Neither option puts revenue against a retrieval. That is the layer blankspace works at: detecting agent and assistant retrieval at the CDN edge and attaching commercial value to it directly, rather than waiting for a click that increasingly does not come. It does not replace the referral economy and it is not a substitute for the licensing question. It is simply the part of the problem that a Search Console setting was never going to reach.
Whatever you decide, decide it deliberately. Read the setting, read your Google contracts, model the downside against the Agarwal and Sen figures rather than against Google's framing, and record why. The publishers who will negotiate well in 2027 are the ones who can show they understood the switch in 2026.
Frequently asked questions
What exactly does Google's AI opt-out control do?
It is a property-level setting in Search Console that removes your site from AI Overviews, AI Mode and AI Overviews in Google Discover. Google states that opted-out sites receive no traffic and no impressions from those features, that the setting is not used as a ranking signal in ordinary search results, and that it has no effect on AI model training, which is governed separately by Google-Extended. It finished rolling out to all websites worldwide on 31 August 2026.
Will opting out of AI Overviews hurt my Google rankings?
Google says no, and the CMA's conduct requirement of 3 June 2026 obliges it to say so and to mean it: publishers that opt out cannot be downranked or displayed differently in general search results. The control applies only to the generative AI surfaces. The practical risk is not ranking but reach, because you lose whatever referrals and impressions those surfaces were producing, and Google does not report the click side of that figure.
How much traffic do AI Overviews actually cost publishers?
The strongest evidence is a randomised field experiment by Saharsh Agarwal and Ananya Sen, posted to SSRN in April 2026, covering 1,065 users and 68,089 searches. Hiding AI Overviews increased outbound organic clicks by 39.8 per cent and reduced zero-click searches by 34.5 per cent, with the effect concentrated on informational queries and on overviews placed at the top of the page. The study found no offsetting improvement in click quality or user satisfaction.
Why do so few publishers seem willing to use the opt-out?
Because they cannot price it. Google's AI performance report supplies impressions from generative AI features but no click data, so publishers have no way to calculate what leaving would cost them. Combined with an opt-out rather than opt-in default, competitive pressure not to move first, and reported suggestions that opting out could affect eligibility for Google's pilot AI deals, the rational individual choice is to stay in even where the collective case for moving is strong.
Is opting out of AI Overviews the same as blocking Google's AI crawler?
No, and this is the most common confusion. The Search Console control governs whether your content appears in and grounds generative answers on Google's surfaces. It does not stop Googlebot fetching your pages, because AI Overviews and AI Mode are built from the same index that powers ordinary search. Google-Extended, a separate token, covers model training and Gemini grounding. Publishers who want to withhold content from Google's AI entirely currently have to combine controls, and the structural separation of search and AI crawling that publisher bodies have asked for does not exist.