On 23 September 2026 the UK Competition and Markets Authority published strengthened proposals for a user choice conduct requirement on Google. Google would have to show a search choice screen when a user first sets up an Android device or first opens Chrome, and re-prompt them every year. AI assistants such as ChatGPT and Perplexity could appear on those screens if they meet technical and security criteria. Every provider on the screens would have to clearly and accurately attribute publisher content. Comments close on 9 October, and the CMA expects a final decision by the end of the year.
The strongest case for intervention
The case for acting is serious. The CMA has said that Google search accounts for more than 90% of general search queries in the UK. Defaults are sticky, and a choice that is offered once, or buried in settings, does little to change behaviour. The CMA designated Google with strategic market status in search in October 2025 under the Digital Markets, Competition and Consumers Act 2024. That designation lets it impose targeted conduct requirements without proving a specific abuse after the fact. Critics of light-touch approaches point out that traditional competition cases take years, while a search market that is changing this fast cannot wait.
That argument is strongest for the AI part of the proposal. CMA chief executive Sarah Cardell said the regulator is responding to the fact that people are increasingly using AI assistants as their search service, yet those assistants are absent from Google's choice screens. A remedy that ignored how people now search would be out of date before it took effect.
Why the revision is the right call
The January proposal was the last outstanding piece of the package. The CMA has already imposed publisher controls over AI features on 3 June 2026, and fair ranking and data portability requirements on 17 June, according to The Next Web's account. Including AI assistants in the choice screen is the pro-innovation part of that package. It lets a new entrant reach users without buying its way onto a device, and it does so without dictating which product wins. A screen listing several options preserves consumer agency, which is the free-market outcome. It is preferable to a ban, a break-up, or a mandate to license content on fixed terms.
The requirement that every listed provider attribute publisher content also carries a sound principle. It applies the same rule to Google and to its challengers, rather than treating incumbents as the only parties with duties. A citation link is a cheap obligation, and it keeps the open web's incentive to publish intact.
Where the design could go wrong
The main risk is eligibility. City AM reports that inclusion is not guaranteed: the criteria AI providers must meet are still under consultation. If the technical and security tests are drawn narrowly, they could favour large, well-resourced assistants and leave out smaller and open-source entrants. Security criteria are legitimate, since a default search service on a phone touches sensitive data. But Google, as the gatekeeper, should not be the one deciding whether a rival passes. The final decision should set objective, published, testable thresholds, with a route to appeal a rejection.
A second risk is the annual re-prompt. Google objected that annual prompts would frustrate users and argued for a permanent settings switch instead, according to The Next Web. Google has a point about friction. Repeated interruptions can produce fatigue, where users tap the first option to dismiss the screen, which would defeat the purpose. The CMA should commit to measuring outcomes, meaning how many users actually change their default and how many simply dismiss the prompt. If the evidence shows fatigue, it should adjust the frequency. This would match the regulator's own stated commitment to evidence-based remedies.
A third risk is the attribution rule. It is sensible in principle, but attribution must be defined by outcome, meaning users can reach the source, rather than by a rigid format that forces every product into the same interface. Otherwise a rule meant to help publishers could freeze design choices in a fast-moving category. Publisher groups such as the Professional Publishers Association and the News Media Association have warned that AI assistants operate outside existing licensing frameworks, and clear attribution is only a partial answer to that concern. Attribution links do not replace compensation, and regulators should be honest that this proposal does not settle the licensing question.
What to watch before year-end
Three things will determine whether this is a good remedy. The first is whether the eligibility criteria are objective and independently verifiable. The second is whether the CMA builds in a review of the annual prompt after the first cycle of data. The third is whether the attribution standard is outcome-based. The consultation closing on 9 October is the moment for developers, publishers and consumer groups to press on all three.
The broader lesson is that the Digital Markets regime works best when it is used as a scalpel. The choice-screen remedy targets a specific default-setting mechanism, applies to a firm with a documented dominant share, and now extends to the products that are actually changing how people search. If the CMA keeps the criteria neutral and the review evidence-led, it will show that pro-competition regulation and innovation can be pursued together.