Google finished rolling out its Search generative AI control to every website worldwide on August 31, 2026, a binary switch that removes a site from AI Overviews and AI Mode on request. The control exists because a UK regulator ordered it in June, not because Google decided publishers deserved the choice, and the traffic numbers explain why almost no one is using it.
Search Console has offered site owners crawl controls, sitemap tools and indexing reports for two decades. What it never offered, until this year, was any way to keep a page fully visible in ordinary blue-link results while pulling it out of Google’s AI-generated answers. Google’s own support documentation now states plainly that as of August 31, 2026, the Search generative AI control has rolled out to all websites worldwide, following an initial test that began with a subset of UK site owners on June 3.
The mechanism is simple to describe. Under Settings in Search Console, every property carries a three-way choice: include the site in AI Overviews, AI Mode and AI Overviews in Discover; exclude it from all three; or inherit whatever setting a parent property has chosen. Google says exclusion is not a ranking signal and does not touch ordinary indexing.
The origin story is less simple, and it is the part most coverage of the toggle skips. Google did not build this because publishers asked nicely. It built it because the UK’s Competition and Markets Authority wrote a legally binding deadline into a conduct requirement, backed by a penalty that reaches into the billions. That distinction, regulatory compulsion rather than product goodwill, shapes everything about how the control behaves, what it does not cover, and why publisher take-up so far has been muted.
Three settings now decide whether a page exists inside an AI answer
The control lives at Settings > Search generative AI inside Search Console, and it resolves to one of three states for every property: Include, Exclude, or Inherit control from parent. Include is the default for every property that has no parent, and it is the setting the overwhelming majority of sites carry today, since switching requires a deliberate action nobody takes by accident.
Nine months separate the UK test from the worldwide release, and the gap looks deliberate rather than administrative. Google described the UK-only phase as thorough testing before a global rollout, language that suggests the company wanted time to watch how a binary exclusion setting behaved across property trees, subdomains and multilingual sites before exposing every verified property in Search Console to the same choice at once.
Choosing Exclude removes a site’s links and content from AI Overviews, AI Mode and the generative features inside Discover. Google’s help documentation is specific about what that means mechanically: content already crawled from the site becomes ineligible as an input for generating an AI response or preview in those three surfaces, and the site stops receiving any impressions or clicks that would otherwise have come from them. A change generally takes a few days to register, and Google says exclusion itself should apply within one to two days after that, though caching across its systems can delay full propagation further. Google is also explicit about what the control does not reach: it does not override a publisher’s separate choices to participate in Merchant Center or Google Ads, and content can still be used to help Google’s broader systems understand query language and page content, a use case Google treats as distinct from generating a visible AI answer.
Inheritance is the setting most site owners never look at, and it is where mistakes are likeliest. A URL-prefix property nested under a domain follows the closest parent that has explicitly stopped inheriting, cascading down from the top-level domain unless someone at a subdomain or subdirectory level overrides it. A publisher who excludes the parent property without checking every child property beneath it can end up blocking sections of the site nobody intended to touch, or leaving sections exposed that someone believed were already covered.
A regulator in London wrote the deadline Google is now meeting
Google announced the toggle on June 3, 2026, in a post on its Keyword blog attributed to Mrinalini Loew, general manager of the Google Search Ecosystem. The same day, the UK’s Competition and Markets Authority imposed what it called a Publisher Conduct Requirement on Google, the first binding conduct requirement issued under the Digital Markets, Competition and Consumers Act 2024. The CMA had designated Google as holding strategic market status in general search and search advertising on October 10, 2025, following a nine-month investigation that examined evidence from Google and more than eighty market participants. That designation, which runs for five years, explicitly brought AI Overviews and AI Mode inside its scope, on the reasoning that both sit directly on Google’s results page and form part of its general search infrastructure.
The requirement itself instructs Google to give publishers genuine control over whether their content trains or grounds generative AI features, to supply engagement metrics through a commonly accessible platform such as Search Console, and to ensure that content is attributed clearly wherever it is used. The CMA framed the outcome as a world first: a search engine legally compelled to let publishers withhold content from AI-generated summaries while keeping it fully eligible for conventional search results.
Google’s own account of the timeline fills in the earlier steps. The company first said in January 2026 that it was exploring an opt-out mechanism, and a Google executive speaking at an industry conference in London in February described the underlying engineering as unusually difficult, because excluding content from AI features while keeping the same pages fully indexed and normally ranked everywhere else required separating two systems that had never been built to operate independently. By March, Google had confirmed to the CMA that the feature was in active development, months before the June 3 launch made it public.
The compliance calendar extends well past the August rollout. The core obligations under the requirement come into legal force on December 3, 2026, Google owes its first formal compliance report to the CMA by January 3, 2027, and page-level grounding controls, a more granular tool than today’s property-wide switch, are due by March 3, 2027. The requirement itself is set to expire on October 10, 2030 unless the CMA extends it. Breach of a conduct requirement under the Act can draw a fine of up to ten percent of a company’s global annual turnover, a penalty large enough that no company designated with strategic market status has treated the regime as optional so far.
Google built this because the alternative was a bigger fight
Google had every commercial reason to avoid building an opt-out at all. In its own June 3 announcement, the company noted that AI Overviews had passed 2.5 billion monthly active users and AI Mode had surpassed one billion, framing generative search as a feature people actively want rather than one it needed to apologize for. Search advertising remained Google’s largest single revenue line through the first quarter of 2026, and every publisher who exits AI Overviews is, by definition, a publisher whose content Google can no longer draw on to keep users inside its own results page.
What changed the calculation was not publisher pressure alone, though that pressure had been building for two years. It was the CMA’s designation, which converted a policy request into a legal obligation with a fine attached. Google’s public framing has been careful to avoid conceding that the toggle is a defensive measure. The company’s blog post described it as building on a “long history of designing tools, like snippet controls and Google-Extended, that give websites more choice,” language that folds a regulator-mandated feature into a narrative of voluntary product evolution. The CMA’s own press material tells a different story, describing the requirement as one it will use again and signalling that opt-out controls are a floor, not a settlement.
Google has paired the toggle with features aimed at making Include the more attractive default. Preferred Sources inside AI Overviews and AI Mode, and subscription labels that flag paywalled or member-only content, both shipped in the same period, alongside a “Further Exploration” section that adds curated links at the end of AI Overview responses. Each addition increases the visible upside of staying in without changing the binary nature of the underlying choice.
The toggle sits between two older controls that do different jobs
Confusion about what the Search generative AI control actually touches is the single most common error in early coverage of the feature, and Google’s own documentation goes out of its way to draw the boundaries. The control governs one thing only: whether a site’s content can appear inside, or help ground, live responses in AI Overviews, AI Mode and Discover’s generative features. It does not affect how Googlebot crawls a site, whether pages get indexed, or where they rank in ordinary search results. Google states directly that the setting is not used as a ranking or inclusion signal anywhere else in Search.
That leaves two adjacent controls that site owners routinely confuse with the new one. Google-Extended, a user-agent token set in robots.txt, governs whether Google can use a site’s content to train the underlying models behind its AI products; it has no effect on whether an already-trained model cites or grounds itself in that content during a live AI Overview. Noindex, the long-standing meta directive, removes a page from Google Search entirely, AI features included, at the cost of losing organic visibility too. The distinction runs deeper than terminology, because a publisher who reaches for the wrong tool ends up either doing nothing useful or losing far more visibility than intended.
Three mechanisms, three different jobs
| Control | What it governs | Effect on organic ranking |
|---|---|---|
| Search generative AI control | Appearance and grounding in AI Overviews, AI Mode, Discover AI features | None, by Google’s stated design |
| Google-Extended (robots.txt) | Use of content to train Google’s AI models | None |
| Noindex | Presence anywhere in Google Search, AI included | Removes the page entirely |
A site owner who sets Google-Extended to disallow and assumes their content is safe from AI Overviews has changed nothing about live grounding, since that model was very likely trained before the block took effect and continues drawing on freshly crawled pages regardless. A site owner who wants out of AI answers specifically, without sacrificing organic rankings, has exactly one lever built for that job, and it only became available worldwide this year.
Click-through data explains why the choice carries real weight
The reason a site owner would even consider flipping the switch shows up in the traffic numbers publishers have been assembling since AI Overviews first scaled. An Ahrefs study published in February 2026, examining roughly 300,000 keywords, measured a 58 percent reduction in click-through rate on the top-ranking organic page when an AI Overview appears, nearly double the decline the same firm had documented in April 2025. Pew Research Center, tracking actual browsing behavior across tens of thousands of real Google searches, found that users clicked a traditional result only 8 percent of the time when an AI Overview was present, against 15 percent when it was absent, a decline of close to half.
The damage does not land evenly. Chartbeat’s analysis across more than 2,500 publisher sites found smaller publishers losing roughly 60 percent of search-referred traffic while the largest publishers in its sample lost closer to 22 percent, with total traffic across the group holding closer to steady as other channels partly absorbed the difference. DMG Media, which owns MailOnline and Metro, told the CMA directly that click-through rates on some of its queries had fallen by as much as 89 percent once an AI Overview appeared above them.
None of that data was collected by Google. It came from third-party crawlers, publisher-side analytics platforms and direct submissions to the regulator, which is itself informative: until this year, the only party with a complete, first-party view of how AI Overviews affected a specific site’s clicks was Google, and Google was not required to share it.
The researchers who measured the damage disagree about its size
The 58 percent figure is the most cited number in every article about the opt-out toggle, and it is also the most frequently misread. Ahrefs measured the decline by comparing December 2023 click-through rates against December 2025, not against a stable, unmoving baseline, and a later independent recheck of the same underlying data found that comparable keywords with no AI Overview attached fell by nearly as much over the same two years. AI Overviews accelerated an existing decline in click-through rates; they did not single-handedly create one from a flat starting point. That distinction changes what a publisher should expect from opting out, since a meaningful share of the traffic loss being blamed on AI summaries would likely have happened anyway as search behavior shifted toward zero-click patterns across the whole industry, AI Overviews or not.
The report Google shipped alongside the toggle compounds the uncertainty rather than resolving it. The Generative AI performance report inside Search Console shows impressions from AI features broken down by page, country, device and date, down to hourly granularity. It does not show clicks, click-through rate or query-level detail. A publisher deciding whether to exclude a section of their site can see how often it appeared inside an AI Overview and cannot see how many of those appearances would ever have converted to a click if the AI summary had not been there, which leaves the decision resting on outside studies rather than the site’s own first-party numbers.
That gap matters because of what excluding a page actually forfeits. Opting out does not remove the AI Overview from the results page; it only removes that site’s citation and link from it. Analysts at firms including ZeroClick Labs have made the point plainly: a publisher who excludes itself hands the citation, and whatever residual click it might have generated, to a competitor who chose to stay in. The trade is not AI visibility versus no AI visibility. It is AI visibility for this site versus AI visibility for the next-ranked one.
No comparable switch exists anywhere else publishers actually get cited
Google is not the only generative system drawing on the open web to answer questions, and the contrast with its rivals sharpens what the toggle is actually worth. OpenAI, Perplexity and Microsoft’s Copilot all ground responses in live web content, and none of them currently offers a publisher-facing, per-property control comparable to Google’s Search generative AI setting inside a tool as widely used as Search Console. Robots.txt directives such as GPTBot or PerplexityBot disallow rules exist for some of these systems, governing crawling in the same coarse way Google-Extended does, but a mechanism that keeps a page indexed and ranked in ordinary results while surgically removing it from AI-generated answers is, so far, specific to Google and specific to a legal order the CMA issued.
That asymmetry matters for anyone advising publishers on where to invest attention. A site that excludes itself from Google’s AI Overviews loses citation opportunities inside the single largest AI answer surface by user count, without gaining any equivalent lever to manage its exposure inside ChatGPT, Perplexity or Copilot, where the crawl-level opt-outs available are blunter and where no regulator has yet forced a comparable reporting obligation. For a content operation weighing where to put its generative-engine-optimization effort, the practical reality in September 2026 is that Google is simultaneously the platform offering the most granular control and the platform under the most binding obligation to keep improving that control, which makes it the wrong place to disengage first even for publishers seriously considering an eventual opt-out elsewhere.
Deciding whether to flip the switch depends on what kind of site this is
The decision is not the same for every property, and the property-level structure of the control makes that granularity awkward to exploit. A site built on informational queries, the category Ahrefs and Pew both found hit hardest by AI Overviews, has the clearest case for testing exclusion on specific sections, since those are the pages least likely to convert a click into revenue even when a visitor does arrive. A site whose traffic comes from branded or transactional queries, where AI Overviews have shown far smaller click-through effects, has a much weaker case, because those visits are the ones most likely to end in a purchase, a subscription or a lead, and Include keeps the door open to Preferred Sources placement and subscription-label visibility that Exclude forecloses entirely.
The safest approach for most publishers, including the long-form and evergreen content that makes up the bulk of an editorial pipeline, is to treat the toggle as a section-level experiment rather than a domain-wide decision. Because a URL-prefix property can carry its own setting independent of its parent, a publisher can exclude a specific content vertical, a discontinued product line or a paywalled section, and measure the effect on that section’s impressions in the Generative AI performance report before touching anything else. Doing this blind at the domain level, without first confirming which child properties inherit what, risks silently excluding pages nobody meant to remove from AI visibility, or leaving paywalled content grounding a free AI summary that the publisher assumed had already been walled off, an error that is easy to make and slow to notice once the property tree has more than two or three levels.
Two deadlines still ahead will decide whether the toggle was the real fix
The August 31 global rollout closes one chapter of the CMA requirement, not the whole file. The core conduct obligations do not take legal effect until December 3, 2026, and that is the date Google must start delivering the detailed engagement metrics the requirement demands, including click-throughs and click-through rate disaggregated from ordinary organic results, a category of data the current Generative AI performance report does not yet supply. A first formal compliance report is due to the CMA a month after that. The bigger structural change lands on March 3, 2027, when page-level grounding controls are required, replacing today’s property-wide switch with something closer to the granularity publishers have actually been asking for since AI Overviews first launched at scale.
The UK’s move has not stayed contained to the UK. The European Publishers Council has lodged a formal complaint with the European Commission over the same underlying practices, and Penske Media, which owns Rolling Stone, Variety and The Hollywood Reporter, filed a US antitrust suit covering comparable ground. None of those actions has yet produced a binding requirement outside the UK, which means the worldwide rollout of the toggle is, for now, a voluntary extension of a UK-only legal obligation rather than a response to a second regulator.
What would change the current calculation is data the industry does not yet have: click-level figures from Google’s own systems, due under the CMA’s December deadline, that would finally let a publisher measure the actual cost of staying included against the actual cost of opting out, rather than inferring both from third-party studies that disagree with each other by tens of percentage points. Until that report exists, the toggle gives publishers a real switch and very little reliable evidence for which way to set it.
Questions publishers keep asking about the AI opt-out toggle
Yes. Google’s own support documentation states that as of August 31, 2026, the control has rolled out to all websites worldwide, after an initial test that began with a subset of UK site owners on June 3, 2026.
Under Settings, in the section labeled Search generative AI, which offers Include, Exclude and Inherit control from parent for each verified property.
Google states directly that the control is not used as a ranking or inclusion signal for any part of Search outside AI Overviews, AI Mode and Discover’s generative features. Ordinary organic rankings are unaffected by the setting.
Google-Extended governs whether Google can use a site’s content to train its underlying AI models. It has no effect on whether an already-trained model grounds a live AI Overview in freshly crawled content from that site. The Search generative AI control governs that live grounding directly.
Noindex removes a page from Google Search entirely, including AI features, and also removes it from ordinary organic results. The Search generative AI control removes a page only from AI-generated features while leaving organic visibility intact.
Google says a control change generally takes a few days to register, and content should be excluded from applicable AI features within one to two days after that, though caching across Google’s systems can delay full propagation further.
No. The AI Overview still generates for the same query. Excluding a site only removes that site’s links and content from being cited or used to ground that specific response; a competing site that stayed included can still appear in its place.
Yes, if that section is configured as its own URL-prefix property in Search Console. A property inherits its parent’s setting by default but can be manually configured to override it, which allows section-level control rather than an all-or-nothing domain decision.
No. The control governs three specific Search-surface features: AI Overviews, AI Mode and AI Overviews in Discover. It does not affect the standalone Gemini app, Google-Extended’s training scope, or other Google products.
Detailed engagement metrics, including click-throughs and click-through rate disaggregated from organic search, are due when the core conduct obligations take legal force on December 3, 2026. More granular page-level grounding controls are due by March 3, 2027, replacing today’s property-wide toggle.
Author:
Jan Bielik
CEO & Founder of Webiano Digital & Marketing Agency
This article is an original analysis supported by the sources cited below
Search generative AI control
Google’s official Search Console documentation confirming the worldwide rollout on August 31, 2026, and detailing the three-way control, inheritance rules and exclusion timeline.
New opportunities, control and insights for website owners
Google’s official Keyword blog announcement of June 3, 2026, introducing the toggle, citing AI Overviews’ 2.5 billion monthly active users and AI Mode’s one billion, and confirming the August 31 global rollout.
Google reacts to UK order with a Search Console AI opt-out toggle
Reporting establishing the same-day link between Google’s June 3 announcement and the CMA’s Publisher Conduct Requirement, plus the compliance calendar through 2030.
Google AI Search Rules: The CMA’s First Order Won’t Be the Last
Coverage of the CMA’s conduct requirement under the Digital Markets, Competition and Consumers Act 2024, including its scope over AI Overviews, AI Mode, Gemini and Vertex AI.
Publishers will be able to opt out of AI Search, thanks to new regulation
Contemporaneous reporting on the June 3 announcement, including Google’s stated AI Overviews and AI Mode user figures and the CMA’s description of the requirement as a world first.
UK regulator tells Google to let publishers opt out of AI Overviews
Reporting detailing the CMA’s specific obligations on Google, including engagement metrics, attribution requirements and Google’s own public response to the order.
The CMA Made Google Show Its UK Search Data
Analysis of the CMA’s October 10, 2025 strategic market status designation, its five-year duration, and its explicit inclusion of AI Overviews and AI Mode within scope.
CMA to investigate Google search dominance and impact on news publishers
Reporting confirming the ten percent global turnover penalty available to the CMA for breach of a conduct requirement under the Digital Markets, Competition and Consumers Act 2024.
Google’s general search and search advertising services
The CMA’s official case page listing the exact dates of the publisher conduct requirement, the fair ranking and data portability requirements, and related consultation milestones.
AI Overviews Cut Clicks 58%. What That Number Actually Measures
Analysis of the Ahrefs February 2026 click-through rate study, including the methodological caveat that comparable non-AI-Overview keywords fell nearly as far over the same period.
Google updates AI Overviews with Further Exploration links, subscription labels as 58% publisher click decline triggers antitrust suits
Reporting on the Ahrefs and Pew Research findings on AI Overview click-through effects, and Google’s Further Exploration, Preferred Sources and subscription-label features.
AI Overviews Traffic 2026: 58% CTR Drop and Google’s Response
Analysis comparing the Ahrefs, Pew and Chartbeat click-through methodologies and their differing but directionally consistent findings on AI Overview impact.
Google AI Overviews Impact On Publishers & How To Adapt Into 2026
Aggregated publisher-reported data, including DMG Media’s submission to the CMA describing click-through rate declines of up to 89 percent on affected queries.
Google’s Search Generative AI Control Explained
Analysis of the strategic trade-off in opting out, including the argument that exclusion transfers citation opportunity to competing sites that remain included.
Search Generative AI Control Explained
Guide distinguishing the Search generative AI control from Google-Extended and noindex, and detailing the UK-first launch and its June 17, 2026 effective date.
| Citing this article? Brief excerpts are welcome. Please credit Webiano.digital, name the author where stated, and include a link to https://webiano.digital and to this original article. Full or substantial republication requires prior written permission. Read our Copyright and Content Use Policy. |
This article was prepared with the assistance of artificial intelligence tools. The content underwent expert human review, and Webiano Digital & Marketing Agency assumes editorial responsibility for its final version and publication.
