Google’s AI Search Opt-Out Is Real Progress. The Missing Click Data Is the Tell

Google’s new AI Search controls are not a generosity story. They are a power story with a dashboard attached.

On Wednesday, Google began testing a new Search Console toggle that lets website owners decide whether their content can appear in and help ground generative AI Search features, including AI Overviews, AI Mode, and AI Overviews in Discover. The rollout starts with a subset of UK website owners, and the geography is not incidental: the UK Competition and Markets Authority has imposed a new conduct requirement on Google Search after designating Google with strategic market status in general search services.

That makes this one of the more consequential AI Search product changes of the year. Not because the toggle will instantly fix publisher economics. It will not. But because Google is finally acknowledging, in product form, that AI Search is not just “Search with a nicer answer box.” It is a distinct distribution surface with different incentives, different measurement needs, and a very different bargain for the people whose content makes the answers possible.

The opt-out matters because the old choice was fake

Google says the new Search Console control lets website owners decide whether their site can “appear in and help ground responses” in generative AI Search experiences. If a site opts out, Google says it will not receive traffic or impressions from those generative AI features, and the choice will not be used as a ranking signal for traditional Search results outside those features.

That last sentence is doing a lot of work. For publishers, software companies, documentation teams, marketplaces, and anyone else with a search-driven acquisition model, the problem was never merely that Google used content in AI answers. The problem was that opting out of the AI layer without taking collateral damage in regular Google Search was not a clean, practical choice. The Guardian describes the earlier trade-off plainly: websites could not meaningfully opt out of being used for AI Overviews without also withdrawing from traditional Google Search, which is not much of a choice when Google is still the dominant gateway to the web.

The CMA’s version is even sharper. It calls the requirement a “world first” and says publishers will have effective tools to prevent their content being used to power AI features in Search, putting news organizations in a stronger position to negotiate content deals with Google. CMA chief executive Sarah Cardell said publishers need “appropriate bargaining power over how their content is used.” That is the regulatory translation of what operators have been saying for a year: if your content is useful enough to ground the answer, you should have more leverage than “accept the default or disappear.”

Google’s framing is softer. It says AI Overviews now has over 2.5 billion monthly active users, while AI Mode has surpassed one billion monthly users, and argues that generative AI Search creates “new opportunities for brands, publishers and creators.” That is not wrong, but it is incomplete. A surface can create new opportunities and still weaken the economics of the pages it summarizes. Those two things can be true at the same time, which is why this control exists.

Impressions without clicks are a useful half-dashboard

The reporting side is where the announcement gets more interesting, and more frustrating. Google is rolling out new Search Console insights showing impressions, pages that appear in AI responses, countries, devices, and dates. The help documentation says the Search generative AI performance report covers AI Overviews and AI Mode, with a separate report for Discover. It also includes the usual Search Console machinery: canonical URL reporting, page/country/device/date dimensions, Pacific Time dates, preliminary dotted-line data, export support, and familiar row and data limitations.

For teams that have been flying blind, that is real progress. If you run a developer docs site, a product-led-growth blog, an ecommerce advice hub, or a technical publication, merely knowing which pages show up in AI responses changes the conversation. You can identify which assets Google treats as answer inventory. You can see whether AI visibility is concentrated in troubleshooting pages, explainers, comparisons, tutorials, reference material, or buying-intent content. You can start mapping “AI Search exposure” as its own channel instead of treating it as an invisible side effect of organic search.

But the first version apparently does not include click data. Search Engine Land’s Barry Schwartz asked Google about it and reported that the report includes impressions, pages, countries, devices, and dates, “but does not include click data.” Google’s response was the kind of product sentence that tells you exactly what is missing while promising nothing specific: it will “introduce additional metrics over time.”

That omission is the tell. Impressions answer “was my page used?” Clicks answer “did that usage help me?” Without clicks, publishers and builders still cannot resolve the boardroom question: is Google’s AI answer driving qualified visits, or satisfying the user before the click?

That distinction matters. A page that appears 100,000 times in AI Mode and produces 20,000 visits is a powerful discovery asset. A page that appears 100,000 times and produces 300 visits may be unpaid training wheels for Google’s answer layer. Both look successful in an impressions-only report. Only one helps the site owner build a business.

What operators should actually do

The wrong move is to treat the opt-out as a moral referendum and immediately flip it everywhere. The better move is to treat AI Search like a new acquisition channel with bad early instrumentation.

Start by segmenting your content. Some pages benefit from being cited even when click-through is lower: API references, canonical docs, high-authority explainers, standards pages, and public troubleshooting material can reinforce trust when Google’s answer points back to you. Other pages are more vulnerable: comparison articles, implementation guides, calculators, pricing explainers, migration checklists, and buyer-intent content often contain the decision logic users came for. If AI Mode extracts the useful decision tree without sending the visitor through, that page’s business value changes.

Then define an experiment plan before the global rollout arrives. If Google eventually allows property-level or more granular controls, teams should compare sections rather than making one sitewide emotional decision. Track normal Search Console clicks, server logs, referrers, branded search, assisted conversions, signup paths, and any visible AI Search impressions. The result will be messy. That is still better than debating the future of search with vibes.

For engineering and product teams, there is also a content architecture lesson hiding here. Google’s updated guidance emphasizes “unique, non-commodity content.” That phrase deserves less eye-rolling than usual because it is exactly the line between content that AI Search can safely compress and content that remains worth visiting. Commodity SEO pages are easy to summarize and easy to replace. Original benchmarks, field notes, failure reports, code samples, migration playbooks, pricing math, architecture diagrams, and opinionated comparisons are harder to flatten into a generic paragraph.

If your content can be replaced by an AI Overview without meaningful loss, Google is not the only problem. The page was already thin. AI Search just made the diagnosis harder to ignore.

The practical action list is boring and useful: inventory high-value organic pages, mark which ones are likely to appear in AI answers, identify which pages depend on click-through rather than reputation, add source-worthy assets that AI systems can cite but not fully substitute, and build reporting that separates AI exposure from traditional search performance as much as current tooling allows. Documentation teams should make canonical answers clear. Marketing teams should stop shipping interchangeable keyword pages. Publishers should watch whether attribution and links improve before assuming the new control changes the economics.

This is regulation becoming product UI

The most important part of the announcement may be that the CMA requirement is not abstract policy language anymore. It is becoming a Search Console control, a performance report, attribution obligations, an opt-out from fine-tuning uses, and compliance reporting every six months for the first year. That is what modern platform regulation often looks like when it matters: not a press conference, but a new checkbox that changes who has leverage.

Google will argue that AI Search increases query volume, surfaces more links, and gives users better answers. Sometimes it will. Publishers will argue that answer engines absorb value from original reporting and weaken the click economy. Often they will be right. The useful position is not to pretend one side is lying. The useful position is to demand measurement good enough to tell which outcome is happening for a given site, page, and query class.

This rollout is progress. A separate AI Search control is better than a bundled, all-or-nothing bargain. Impression reporting is better than black-box extraction. Regulatory pressure that produces actual operator tools is better than another panel about the future of the open web.

But until Google shows clicks, the dashboard is still missing the metric that turns visibility into accountability. AI Search is now big enough to need its own controls. It is also big enough to deserve honest accounting.

Sources: Google, UK Competition and Markets Authority, Google Search Console Help, Search Engine Land, The Guardian