Google's AI opt-out and the click data due December 3

As of November 2026, the decision on whether to pull your content out of Google's AI features is still open, and the number that would close it has not arrived. Google has given publishers an opt-out control in Search Console, opened it to sites worldwide on August 31, and promised impression figures, but the click data that shows whether those AI appearances send anyone to a site reaches UK publishers only by December 3. Until then you are choosing between a visibility you can see and a traffic effect you cannot. The practical move for anyone running aeo work is to measure the visibility side now, in your own answer set, so the opt-out call is not made blind.

What the control actually does

The opt-out is a Search Console setting that keeps a site's content out of Google's AI features. It comes from the UK Competition and Markets Authority's first conduct requirement on Google, and Google extended it beyond the UK: sites worldwide can use it from August 31. Finer, page-level grounding controls are scheduled for March 3, 2027.

Two limits shape how usable it is today. The controls are bundled, so a publisher cannot switch off one AI feature and keep another. And the unit of choice is the site or, later, the page, not the query, the topic or the surface.

Google has shared impression data with regulators and publishers: how often publisher content shows up inside AI features. It has not released clicks. Chris Dicker of Candr Media, the company behind Trusted Reviews, told Digiday in its 2026-11-09 report that clicks are the metric that decides whether publishers turn the feature off, and that everyone is holding back until the data appears. That is a rational wait, not indecision: a bundled switch cannot be tested halfway, and the visibility an aeo programme would protect is still unmeasured on the publisher's own side.

Why the trade-off got harder this autumn

Google moved in the other direction at the same time. AI Mode and AI Overviews now carry links more often and more prominently, and add timely article carousels together with links to related stories. Preferred Sources, the feature that lets a signed-in user pin the sites they trust, has grown sevenfold this year.

The Daily Mail's Carly Steven described the collision at Digiday's Media Strategies London event in October: just as publishers were about to gain removal at URL level, Google announced a new surface worth being on. The Economist Group's Josh Muncke framed the eventual call as two numbers: how much traffic Google sends, and how often the publication's name appears in AI answers even when nobody clicks. The second one is a visibility measure, and it survives the absence of click data.

Google's own AI data has boundaries

Search Console now has a dedicated generative AI performance report, launched on June 3, 2026 and rolled out to all websites worldwide as of August 31, 2026. It breaks impressions, pages, countries, devices and dates for content shown inside AI Overviews, AI Mode and generative AI surfaces in Discover.

Read the scope carefully before you build a decision on it: a generative engine optimization singapore read of that panel alone will miss every answer where the brand was cited but never shown to a reader. The report shows impressions inside AI features. It does not hand you the prompt list an engine worked from, the internal query fan-out behind an answer, or a conversion attributed to an AI answer. A high impression count in that report tells you a URL appeared; it says nothing about whether the appearance produced a visit, and average position in the classic report is not the same measure as a fixed-condition ranking.

The four stages a publisher must separate

The opt-out question is usually argued as one variable, traffic. It is really four, and only the first is fully outside your control once you leave a surface.

Opting out removes the page from retrieval, so the cited, clicked and converted stages all go with it. Staying in keeps the cited stage and leaves the clicked stage unproven until the December 3 data lands. Measuring only clicks treats the cited stage as worthless; measuring only citations treats traffic as irrelevant. Both errors are easy to make when only one of the two has data.

Is AI Overviews the same thing as ChatGPT?

No, and the two get confused more often than they should. AI Overviews and AI Mode are Google surfaces inside Search, and the opt-out control governs those. ChatGPT is a separate product from a different company with its own retrieval and its own terms, and a Google Search Console switch has no effect on it.

The practical difference for a publisher is where the bargaining power sits. With Google you have a formal control, a regulator, a published set of AI surfaces and, soon, click data. With other engines you have robots directives and the site's own content, and no reporting panel that tells you how often you were cited. The gap is being narrowed at the crawl-consent layer rather than the search layer: Google has announced support for a shared AI training control in robots.txt, built on the IETF's AIPREF work and extending Google-Extended, the training control it has offered since 2023. The company aims to support the new control with other AI providers in 2027.

So the question "is AI Overviews ChatGPT" has a second half: the AI visibility you protect is not one number. A brand can be cited in ChatGPT answers and missing from AI Overviews, or the reverse. Any governance plan that treats them as one channel will over- or under-estimate what an opt-out costs.

A measuring routine that runs before December 3

You do not need Google's click data to build the visibility baseline you would lose. Start from the answers themselves.

1. Fix the prompt set. Record the questions a buyer would ask in your category, and freeze their wording for the whole period. A prompt list that changes between months measures the instrument, not the market.
2. Fix the conditions. Same engine, same mode, same region, same language, same sampling cadence. A citation rate sampled in Singapore is not comparable with one sampled in the UK.
3. Count mentions and own-site citations separately. Mention rate is answers containing a brand mention divided by valid answers under identical conditions; source-appearance rate is answers where the brand's own site appears divided by the same set. The two move independently, and a merged score hides which one changed.
4. Label the answers that cannot be used. A no-citation answer and a failed sample are not zeros; they are missing data, and folding them into the denominator distorts every later comparison.
5. Score sentiment separately from presence. Being mentioned is not the same as being described accurately, and neither is the same as being described negatively.

Run that routine each month against the same questions and you hold a record of the visibility you would give up, measured before the opt-out decision becomes urgent, the same baseline an aeo programme needs on the brand side. When the click data arrives, the two records sit side by side: citations per question on one axis, visits per citation on the other. The decision turns into a comparison instead of a guess.

What Topkee's GEO work says about a visibility decision made without click data

This is where the Topkee GEO Service viewpoint is useful, with one boundary stated first: the service is sold to companies that want to appear in AI answers, mainly SMEs in Hong Kong that compare and shortlist suppliers, and Topkee Media, founded in Hong Kong in 2010, has no media-licensing or opt-out product for publishers. It is a reference for the same measurement mechanism, not a solution to the CMA question.

What the service does is turn AI visibility into a defined and bounded metric. A brand's AI visibility is the degree to which it is mentioned, accurately described and cited as a source in AI answers. The service tracks prompts, reads the query fan-out an engine actually used, compares brand descriptions with competitors', and separates retrieval from citation: its AI browsing analysis distinguishes sources retrieved from citations visible in the answer, and the citation summary includes retrieval sources marked inline=false, which cannot be called a visible-citation rate. That distinction is the exact one a publisher needs on December 3. Leaving a surface changes retrieval; what it costs you in visible citations is a second question, and what it costs in clicks a third.

The monitoring tool, GEO Insight, keeps the same discipline. Engines are sampled through APIs, so a report describes the sample and not what every consumer sees, and new projects start with engines disabled until someone chooses them. Publicly priced monthly plans anchor what such monitoring costs, with every tier stating no guarantee of AI recommendation, fixed exposure or query volume. For a publisher, the usable part is not the price. It is the reporting split: search and AI visibility reported apart from visits and engagement, and both reported apart from business results, with the line "must have usable data" attached to the traffic layer. That is how a four-stage measurement survives a period when one of the four stages has no data at all.

What to do this month

Do three things before the December 3 data lands, whether you run a publisher title or a brand site buying generative engine optimization singapore support. Build the frozen prompt set and the baseline citations while the control is still unused. Write down the surfaces you would lose, which means AI Overviews, AI Mode, the article carousels inside them and the Top Stories unit that some publishers expect to fall outside the carousel, because the bundled switch removes them together. Then decide who owns the call, so the click data arrives to a prepared comparison and not to a meeting.

If the AI referral question is the immediate problem rather than the opt-out, start with how paid placements behave inside those answers: AI answer ad counts: what apparel and electronics show covers where the panels sit. For the data layer that makes AI referral visits identifiable at all, see how leading brands achieve growth with Google Ads and first-party data.

FAQ

Is Google's AI opt-out available outside the UK?

Yes. The control follows the UK competition authority's requirement, but Google opened it to sites worldwide on August 31.

When will the click data arrive?

Google is required to provide click data to UK publishers by December 3. Publishers have asked for it globally, and Google has not confirmed a global release.

Does opting out remove my site from ChatGPT as well?

No. The Search Console control covers Google's AI surfaces. ChatGPT is a separate product with its own retrieval rules, so a Google switch leaves that channel unchanged.

Can I opt out of AI Mode but stay in AI Overviews?

Not with the current control. It is bundled at site level, with page-level controls due by March 3, 2027, and UK publisher bodies have asked the regulator to unbundle it.

Does a high AI Overviews impression count mean traffic?

No. Search Console's generative AI report counts impressions inside AI features, not clicks or conversions, and it does not expose the prompts behind them.

Sources

1. Jessica Davies and Seb Joseph, "As publishers await click data, Google's AI visibility push complicates the opt-out decision", Digiday, 2026-11-09. https://digiday.com/media/as-publishers-await-click-data-googles-ai-visibility-push-complicates-the-opt-out-decision/
2. Hillel Maoz and Moshe Samet, "Introducing Search Generative AI performance reports in Search Console", Google Search Central Blog, 2026-06-03. https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports
3. John Mueller, "Supporting a shared AI training control for websites", Google Search Central Blog, 2026-10-30. https://developers.google.com/search/blog/2026/10/ai-training-control