Google’s AI opt-out toggle is a governance question in disguise

Google's new Search Console setting lets you pull your site out of AI Overviews, AI Mode, and Discover's AI features without leaving Search. It's the control the industry has been asking for since AI Overviews launched. It's already rolling out beyond the UK. And most of the people who'll click it will regret it inside a quarter.

Not because opting out is wrong. Because the decision is being framed as a preference — a toggle you flip based on how you feel about AI scraping your content — when it's actually a distribution question with second-order consequences the toggle itself doesn't surface.

The interesting part isn't the setting. The interesting part is what the setting reveals about how tangled Google's own product boundaries have become, and how badly most marketing teams are equipped to make the call.

Let me explain.

What the toggle actually does

The Search generative AI control lives under Settings in Search Console. Two options: include your site (default), or exclude it. Excluding removes your content from AI Overviews, AI Mode, and Discover's generative features — links included. Google began respecting the setting on 17 June for UK properties and started rolling it out elsewhere in July.

It doesn't affect regular Search rankings. It doesn't override Merchant Center or Google Ads. It doesn't touch AI training (that's still Google-Extended in robots.txt). It's a scalpel for one specific surface.

That's the pitch. And on paper, it's a clean win for publisher control. The reality is messier because Google's surfaces aren't as clean as the toggle pretends.

The Top Stories problem

NewzDash — which sells news-visibility tracking to publishers — flagged this week that Top Stories carousels are now rendering *inside* AI Overviews on U.S. trending news results. Read that again. The Top Stories unit, which has been a traditional organic feature for over a decade, is showing up nested within an AI Overview.

The toggle assumes surfaces are separable. The product has been busy making them inseparable.

So what happens when you opt out? Google's documentation says the setting removes your site from AI features. But if a Top Stories carousel is architecturally *inside* an AI Overview, does opting out of the Overview also opt you out of the carousel embedded in it?

Google hasn't answered this cleanly. The honest answer is probably "sometimes, in ways we're still figuring out." Which means a news publisher opting out to protect click-through rates could quietly lose Top Stories placement — the single most valuable news surface Google offers — in the process.

The toggle assumes surfaces are separable. The product has been busy making them inseparable.

This isn't a Google conspiracy. It's what happens when a company ships AI features by wrapping them around existing organic components rather than rebuilding from scratch. The seams show up as governance problems.

The setting nobody knows who owns

Here's the question I'd love someone to answer honestly at their next agency review: who at your company makes the call on this?

The head of SEO knows what the toggle does technically. They probably don't know the commercial impact of leaving AI Overview citations on the table, because nobody knows that yet — the measurement layer isn't mature enough. The head of content might have strong opinions about AI training but is often two abstractions away from the Search Console admin. Legal cares about opt-out for reasons that have nothing to do with distribution economics. Comms wants to be seen doing the right thing on AI.

And the CMO, who technically owns the outcome, has never opened Search Console.

So what you get, in practice, is a decision made by whoever notices the setting first. In most enterprises that'll be an SEO manager two levels down who defaults to "leave it on because traffic." In news organisations it'll be an editor who's furious about scraping and defaults to "turn it off." Neither decision is grounded in the actual second-order tradeoff, because nobody has built the model for it yet.

I've spent 18 years watching companies make distribution decisions like this. The pattern is always the same. When a platform ships a control that touches multiple stakeholder concerns simultaneously — SEO, legal, editorial, brand — and nobody has clear ownership, the decision gets made by proximity to the admin panel. That's fine for changing a title tag. It's not fine for opting out of a discovery surface you don't yet understand.

What "opting out" is really opting out of

The framing of the toggle is preference-based: do you want your content used in AI features or not? That framing hides the actual question, which is a distribution economics question.

Here's the real question. If AI Overviews are cannibalising 15–30% of your organic clicks on the queries where they appear (the current ballpark from multiple studies), but AI Overviews are also generating a growing share of brand impressions — impressions that don't show up in Search Console because attribution is broken — is the net negative or positive?

Nobody has that number for their own business right now. Not the toggle-clickers, not Google, not the agencies advising the toggle-clickers. The measurement layer isn't there yet.

Opting out based on click loss you *can* see, while ignoring brand impression value you *can't* see, is exactly the kind of decision that looks rational in a quarterly review and turns out to have been catastrophic 18 months later. This is a recurring pattern. The businesses that opted out of Google Shopping's paid comparison model in 2013 because clicks got expensive missed a decade of e-commerce growth. The businesses that blocked Reddit from crawling in 2019 to "protect" their content missed the entire Reddit-as-answer-engine wave.

I'm not saying opting out is wrong. I'm saying opting out based on incomplete information, because a toggle appeared and someone clicked it, is a governance failure the toggle disguises as user empowerment.

The regulatory subtext

The CMA required this control. Google shipped it, then extended it globally. That sequence matters.

Regulators care about publisher agency. They should. The AI-scraping debate has legitimate grievances at its core — publishers whose content trained models that now compete with them for attention deserve some form of consent architecture. The Search Console toggle is a start.

But regulatory-driven controls tend to have a specific shape. They optimise for the ability to opt out, not the ability to make an informed decision. The CMA's mandate was "give publishers control." It was not "give publishers a decision-support system for understanding the distribution consequences of exercising that control."

So Google shipped what was asked for. A binary switch. Include or exclude. And now every website owner is expected to make a decision on the basis of information Google has no incentive to provide clearly and no third party has the data to provide accurately.

This is where the interesting second-order effect kicks in. The toggle creates a *governance vacuum*. Someone has to decide. But the information required to decide well doesn't exist yet. Which means the toggle rewards whoever has the strongest institutional opinion, not whoever has the best analysis. That's how you get news organisations opting out in a wave over the next six months on principle, only to discover in 2027 that AI Mode became a primary news discovery surface and they've been invisible on it for a year.

What the toggle is teaching us about product structure

Step back from the specific decision and there's a broader observation worth sitting with.

The reason the toggle is hard to reason about is that Google has spent three years building AI features by *embedding them into existing surfaces* rather than shipping them as standalone products. AI Overviews live inside the SERP. AI Mode lives inside Search. Discover's AI features live inside Discover. Now Top Stories carousels are appearing inside AI Overviews.

This is a deliberate product strategy. It maximises adoption because users don't have to change behaviour. But it makes surface-level control almost impossible to implement cleanly, because the surfaces bleed into each other by design.

Compare this to how ChatGPT handles the same problem. OpenAI's crawling controls are cleaner because ChatGPT is a separate product. You can allow or block GPTBot. There's no ambiguity because there's no embedded structure to be ambiguous about. The trade-off is simpler because the architecture is simpler.

Google's toggle exists because regulators demanded it, but the product architecture wasn't built with clean opt-out in mind. So the toggle is a patch over a structural mismatch. It'll work for now. It'll get messier as more organic features get absorbed into AI surfaces — which they will, because that's the product direction.

What I'd actually recommend

For most sites: leave it on. The click loss from AI Overviews is real but not existential, and the brand impression value is undermeasured. Don't opt out until you have a measurement framework good enough to know what you're opting out of.

For news organisations: this is genuinely hard. The click loss is more severe, the Top Stories entanglement is a real risk, and the brand impression argument is weaker because news queries have lower brand-repeat behaviour. I'd still lean toward staying in for now, because opting out on principle without the ability to measure re-opting-in is a one-way door.

For anyone about to click the toggle: at minimum, document the decision. Who made it, on what basis, when it should be revisited. The governance failure I described earlier gets fixed by writing it down. Sites that treat this as a routine SEO setting change will find themselves unable to explain the decision six months later when someone asks why AI traffic is zero. Sites that treat it as a documented commercial decision, with an owner and a review date, will be able to course-correct.

Where I'd concede

The strongest counterargument is that opting out sends a signal — to Google, to regulators, to the industry — that publishers won't accept a model where their content grounds AI answers without consent or compensation. If enough sites opt out, the economics of AI search change. That's real. Collective action has value the individual toggle-clicker's spreadsheet can't capture.

If your decision is fundamentally about political leverage rather than distribution optimisation, opting out is defensible. Just don't confuse the two decisions. A political opt-out and a commercial opt-out are different arguments requiring different evidence, and dressing one up as the other is how organisations make bad decisions with clean-looking justifications.

The real story

The toggle isn't the story. The story is that Google shipped a control for a set of surfaces it's simultaneously making more entangled, on a regulator's timeline, without giving anyone the measurement infrastructure to use the control intelligently. And the industry response has been to treat it as a preference question when it's a governance question dressed as a checkbox.

The businesses that handle this well will treat the toggle as a decision that needs an owner, a rationale, and a review date. The ones that handle it badly will click it once and forget about it. In 12 months we'll be able to tell them apart by their AI Mode visibility. The gap will be significant.

That's the loop. And this time, at least, we can see it forming.