AI Overviews Opt-out: Taking Your Content Out of Google’s AI Search

For two years the deal with AI Overviews was one-sided. Your content ends up in the AI answer – whether you want it or not. The only way out was to lock Google out entirely, which also dropped you from normal search. For any site that lives on search traffic, that was not a choice but a threat.

That is changing right now. Google has introduced an opt-out toggle in Search Console that removes your content from the generative AI features of search – and which, according to Google, is explicitly not a direct ranking signal for normal search. Sounds like what publishers have been waiting for since AI Overviews launched. It is. With three big asterisks.

First: the toggle is UK-only for now. Second: it bundles several functions that you would actually want to control separately. And third – this is the German special case – the Regional Court of Munich I has, in parallel and according to media reports, ruled by preliminary injunction that Google is to be held liable as a direct interferer for false statements in AI Overviews (not yet legally binding). The regulatory and the legal front are moving at the same time, and both change the question of whether you even want to stay inside AI Overviews.

Important for context, because two completely different things get mixed up here: this article is not about the browser extensions that let end users hide the AI overview in their own search. It is about operator control – that is, making sure your website is no longer used and linked as a source for the AI answers. That is an SEO decision, not a browser setting.

I cannot touch the toggle myself yet – in the DACH region it simply is not enabled, and that is exactly the catch. What I could do: go through Google’s documentation line by line, ever since the UK function was documented. And it becomes immediately clear how easily you can grab the wrong one here. Four levers that sound similar, three of them do something entirely different from the new toggle. This confusion gets expensive as soon as it cascades across an entire property.

The control is officially called Search generative AI control at Google and sits in Search Console under Settings → Search generative AI. According to the official Search Console help it manages your site’s visibility in three areas: AI Overviews, AI Mode and – according to the documentation, currently planned this way – generative AI features in Google Discover. Google itself notes that it will expand this list over time. What is actually switched on and off here I broke down in detail in the pillar on the difference between AI Mode and AI Overviews – this post is the spoke to it: there the what, here the whether and how of the exclusion.

You have three settings to choose from:

The decisive sentence is in Google’s documentation in black and white: this control is not used as a ranking or inclusion signal for other parts of search. An exclusion therefore costs you – according to Google – no position in the classic blue links. Your content still feeds into search’s general language understanding, and other Google services such as Merchant Center or Google Ads remain unaffected.

Two technical details that matter in practice: a change typically takes effect within one to two days; due to caching and propagation it can, according to Google, take longer in individual cases. And during the current test phase you can set the option, but according to Google it only actually affects search since 17 June 2026. Anyone who flipped the toggle earlier was not yet protected.

Google is rolling the toggle out first only to a subset of website operators in the United Kingdom – explicitly for “thorough testing” before it follows globally. There is no concrete date for Germany, Austria or Switzerland. The driver is the British competition authority CMA, which obligated Google to provide these controls under the Digital Markets, Competition and Consumers Act 2024; page-level controls are, according to the current CMA requirements, expected to come into force from around March 2027.

This creates a lopsided situation you need to be aware of: the AI Overviews have already been rolled out in Germany and Austria since spring 2025. You are therefore long since a source for the AI answers – you just do not have control over it here yet. You know exactly this “feature already here, control not yet” pattern from my post on the Google Discover update February 2026, which likewise started US-only. The lesson is the same: nothing to fear yet – but the signals are clear, and whoever is smart prepares now.

Preparation means one thing above all for me: making AI visibility measurable before the toggle even arrives in DACH. In parallel with the opt-out, Google has introduced its own performance data for the AI features in Search Console (labelled by Google as “insights”). They show impressions in the AI features, affected URLs, countries and devices – separate from classic performance.

That the data basis in GSC suddenly becomes a lever is not a new theme: back when Google disabled the num=100 parameter, many had to learn to read their impression figures anew. With the AI report it is the same reflex – first understand the data basis, then interpret.

This is the point where most articles and many SEOs are sloppy. Until this toggle there was no lever that only controlled the AI display. Google’s help is, for once, clear here – it names the distinction explicitly:

Concretely this means, with Google’s own words as evidence: the new toggle does not affect AI training – for “to limit training of the models” Google explicitly points to Google-Extended. And to take your content out of search completely, you still need noindex. Google-Extended, in turn, limits training but does not pull you out of the AI Overviews display. And nosnippet was, until now, the only emergency exit from the AI features – at the price that your normal snippet disappears too. This coupling is exactly what the new toggle dissolves, as the Search Engine Journal also highlights.

This is the part that Google’s detailed help added later – and it is error-prone in practice. The logic in brief: by default a property inherits its control from the nearest parent property that has actively stopped inheriting. If none of the parent properties has manually configured anything, the top-level domain’s setting applies.

An example from Google’s documentation: the child property https://example.com/business/cats inherits, by default, the setting of its domain property example.com – unless a closer parent property such as https://example.com/business/ was manually configured. Owners of a child property can at any time either follow the parent or set their own configuration for the property and its sub-pages.

Why this regularly goes wrong in technical audits: anyone managing several properties in Search Console – a domain property plus various URL-prefix properties – likes to flip the toggle at the most convenient level and overlooks that the setting cascades downwards. When in doubt, check per property individually whether it inherits or has its own configuration before you set anything to Exclude.

The data fuelling the debate is real – but no licence for panic. An Ahrefs analysis sees the organic CTR on position 1 around 58 percent lower when an AI Overview is present – as a correlation, not a proven cause. Pew Research data show an 8 percent click rate on a classic search result when an AI Overview is present, versus 15 percent without. The British Publishers Association reported a 19 percent CTR decline for academic reference services. And in a non-representative X poll by SEO industry observer Barry Schwartz (just over 350 participants), around 33 percent said they want to exclude their content as soon as Google allows it. The figures come from different surveys, each with its own methodology, and are not directly comparable – but they still work as a directional indicator.

Even so, the criticism of the toggle itself carries weight, and you should know it before you praise the toggle. Paul Bannister of Raptive calls the opt-out, in essence, a light switch while the power plant keeps running – real control looks different. Stuart Forrest (formerly of Bauer Media) puts the core problem in a nutshell: without AI-specific click data you cannot judge at all whether AI Overviews bring you traffic or only cut it off. And Chris Dicker (Candr Media) warns that Google lumps AI Overviews, AI Mode and Discover together instead of making each feature individually controllable – a problem as soon as Discover hangs in the same toggle as a traffic source.

Against the gut feeling, a sober self-test helps. These are exactly the questions I go through before I would ever advise a client to opt out – more than one or two clear yeses argue for it, otherwise you stay in:

• Business model: Does your site live on paid content or subscriptions whose value drops as soon as the AI summarises them for free?
• Reputation risk: Is there a concrete risk that a faulty AI summary of your content harms you (see the Munich ruling below)?
• Traffic dependency: How large is the share of visibility you lose in the AI answer – and can you absorb it?
• Data basis: Do you have solid figures that AI Overviews kill clicks for your topics rather than bring reach?
• Alternatives: Do you have distribution channels beyond Google search that can cushion a loss of reach?

While the industry discusses the UK opt-out, a German court has in parallel drawn a possibly more far-reaching line. According to consistent reports from several trade media, the Regional Court of Munich I (case no. 26 O 869/26) has prohibited Google by preliminary injunction from spreading untrue factual claims about two Munich publishers in its “AI overview”. The AI had falsely linked the plaintiffs to subscription traps and dubious business practices – statements that appeared in none of the linked sources.

The legal core, and this is what is genuinely new: the court classifies Google not as a merely indirect but as a direct interferer (“unmittelbarer Störer”). The AI overview summarises sources “in its own words and according to its own structure” and thereby makes “its own statements, put forward by the defendant itself”. The liability privilege of classic search engines – under which an operator only makes third-party content findable – does not apply here, not even via the Digital Services Act. According to the ruling, Google must bear around 80 percent of the proceedings costs; in case of violation a penalty of up to 250,000 euros looms. Notably, but to be read with caution: according to the reports so far it is unclear to what extent the injunction has effect beyond Germany. In principle, preliminary injunctions by German courts are territorially bound, but they can have de-facto global effects.

Context, without overdoing it: this is a preliminary injunction, not a legally binding judgment – Google can appeal, and the decision ties in with a ruling by the Regional Court of Frankfurt am Main from September 2025 (case no. 2-06 O 271/25). It is a single case, not settled case law; whether this legal view prevails is open and will probably only be clarified in higher instances. SISTRIX boss Johannes Beus classifies the ruling as a possible directional decision, and Christian Kunz of SEO Südwest rates it as significant; OnlineMarketing.de documents Google’s statement that it considers the decision not final. For practice this means: in DACH, the law – unlike, until now, only the technology – gives you a second instrument when an AI overview harms you. How robust Google’s quality mechanisms really are I covered in detail when I explained how Google rates AI texts.

My advice: treat the topic as preparation, not as a to-do. As long as the toggle has not arrived in DACH, the most sensible thing is to build a clean baseline of your AI visibility, learn to keep the four levers apart and watch how the regulatory and legal situation develops. When the opt-out is then available, you decide with numbers instead of reflexes – and know exactly which lever you are touching.

The more interesting movement in Germany is coming from two sides at once anyway: the opt-out from Brussels and London – and the liability from Munich. Both shake the same question, namely who the AI answer actually belongs to. This question will keep us busy often in 2026.