CMA orders Google to offer AI search opt-out in UK

Britain's Competition and Markets Authority has done what publishers across Europe have been demanding for two years: forced Google to separate the question of whether a site appears in search from whether its content trains and feeds AI Overviews. Search Engine Journal reports that the CMA's new conduct requirement, issued under its Strategic Market Status regime, compels Google to let UK publishers opt out of AI search features without being penalised in ordinary search rankings, and to provide clearer attribution when their content is used.

Until now the choice was binary and ugly. Block Google-Extended and lose your shot at AI Overview citations; allow it and watch your work paraphrased above the blue links, with clicks falling accordingly. The CMA has prised those two decisions apart. In theory, a UK publisher can now stay fully indexed in classical search while withholding consent for generative features. In practice, the mechanism, the granularity, and the enforcement teeth all remain to be specified.

A regulator picks a side

The decision matters less for its immediate operational detail than for the precedent. The CMA is the first major regulator to treat AI search inclusion as a distinct consent question from search indexing, and to back it with the SMS designation Google received in October. The EU's AI Act addresses training data; the CMA is addressing live retrieval and synthesis. That is a sharper instrument. Expect the European Commission and, more slowly, the US to study the wording.

Google's counter is predictable: opting out will mean reduced visibility in the surfaces where users increasingly start their journeys. The company is not wrong. AI Overviews now appear on a rising share of UK queries, and the citation slots inside them are becoming the most valuable real estate on the results page. A publisher who opts out preserves dignity and loses traffic. A publisher who opts in keeps the traffic, for now, and surrenders the margin to a summariser.

What this changes for brands

For B2B brands publishing thought leadership, the calculus is not the same as for news publishers. The Financial Times worries about lost subscriptions. A reinsurer, a development bank, or an industrial group publishing research worries about something else: being cited as the authoritative source inside the AI answer. For them, opting out of AI Overviews would be self-defeating. The whole point of the content is to shape the answer the model gives when a procurement director or a finance ministry asks about catastrophe bonds, financial inclusion, or low-carbon cement.

Which is why the CMA's intervention, welcome as it is for the FT and the Guardian, slightly misreads the market for institutional content. Multilaterals such as the UNDRR or CGAP, standards bodies such as ISO and IEEE, and the research arms of large industrials are not trying to monetise clicks. They are trying to be the citation. The opt-out is a lever they will almost never pull. What they need from regulators is the second half of the CMA's order: clearer, more consistent attribution inside AI answers, with source links that survive the user's scroll.

That attribution piece is the underrated part of the ruling. If Google is now obliged to surface provenance more visibly in UK AI Overviews, the citation becomes more valuable, not less. Brands whose content is structured for retrieval, dense with named data, clear claims, defensible methodology, will benefit. Brands whose content is a swamp of adjectives will not be cited, opt-out or no opt-out.

The fragmentation problem

The deeper consequence is jurisdictional. Google now has to operate a UK-specific consent layer for AI features, on top of whatever it builds for the EU, and whatever the US eventually demands. The model's training corpus will increasingly diverge from the content it is permitted to retrieve and cite in real time, by country. Expect AI answers to look measurably different in London than in Frankfurt or New York within eighteen months, not because the model differs but because the permissioned source set does.

For communications leaders at global institutions, that means the assumption of a single "AI visibility" strategy is already obsolete. The same white paper may be cited in a US AI Overview, paraphrased without attribution in a French one, and absent entirely from a UK one if the host publisher has opted out. Tracking AI citations will require geographic segmentation that almost no brand currently does.

The CMA has handed publishers a lever. Most institutional brands should not pull it. They should instead spend the next quarter auditing whether their content is built to win the citation slot the regulator has just made more visible, and whether they can even tell, today, in which jurisdictions they are being cited and in which they are being quietly summarised away.