CMA Proposes Android Search Choice for AI Assistants

Britain's competition regulator has proposed a change that could put an AI assistant where Google Search sits today: behind the address bar, home-screen widget and search app used by millions of people.

The Competition and Markets Authority's revised user-choice consultation would let services such as ChatGPT or Perplexity qualify for the search-choice screens on Android devices and Google's Chrome browser. Its technology-neutral test would not disqualify an AI interface that can perform the work of a general search engine.

That is still a proposal, not an obligation. The consultation closes at 5pm on 9 October 2026 and the CMA expects a final decision by the end of the year. No user-choice conduct requirement has yet been imposed. If it is imposed in the form proposed, the eligibility process would start within one month and the rest of the requirement would take effect within six months.

A default that reaches beyond the browser

The architecture is broader than a one-off menu on a new phone. A new qualifying Android device would show a non-skippable screen at first use or set-up. New desktop and iOS Chrome installations would do the same, while existing users would see it with the first major browser update after the requirement took effect.

The chosen provider would become the default across the relevant access points. In Chrome that means the address bar. On Android it also covers the default search application and home-screen widget.

Google would also have to show a small, skippable annual notification leading back to the choice screen, unless the user had deliberately selected or changed a default in the preceding six months. The full provider list would remain reachable through browser and device settings.

The screen itself would carry no more than 12 providers. If 12 or fewer qualified, all would appear. If more qualified, five places would go to the most popular services and seven would be selected at random. For the first six months, popularity could be measured through a reasonable metric such as Google Play downloads. It would then be based on selections as a proportion of appearances on the screen. The 12 displayed services would be put in a fully random order, users would have to scroll through them before confirming, and a link would expose every other eligible provider.

There is also a trial mechanism. A user could make another service the default for a short period, which the CMA says should last at least a week, before receiving a skippable route back to the default settings. That matters because the regulator's qualitative research found that a trial option prompted some people to examine unfamiliar services more closely, although many did not initially understand what the trial would change.

AI assistants would have to behave like search

An applicant would have to satisfy three tests. It must address the full range of general-search uses across a wide range of subjects, including by searching the web. It must work technically across the relevant access points, meet minimum security requirements and show it can serve a potentially growing user base. A meaningful proportion of its UK users must also regard or use it as a general search service. The CMA says that need not be a majority.

The interpretative notes make the first test demanding. A provider must handle navigational, transactional, simple and complex informational queries, and supply up-to-date information. It may rely on a syndicated index.

For publishers, the operative language sits here rather than in a free-standing duty on AI companies. Any generative AI summaries would need to be configured so that web content is attributed clearly and accurately and users have a clear means of reaching the source. Providers could choose how to achieve that result. In other words, attribution and link access would become conditions of admission to the screen if the proposal is adopted, not a general statutory rule covering every AI answer in Britain.

The enforcement route also runs through Google. It would decide applications every six months, using only the stated criteria, and issue decisions within two months of a window opening. It would give the CMA every decision and detailed reasons for rejections. If Google failed to operate the process as required, the CMA could investigate it, order a remedy and, in principle, impose a fixed penalty of up to 10 per cent of worldwide turnover under the digital-markets regime. The proposal does not give the CMA the same direct route against an assistant provider.

A large screen with modest switching evidence

The addressable base is substantial. The CMA's workings assume about 30 million UK Android devices, excluding very old models, and an estimated 15 million to 35 million desktop and iOS devices using Chrome. It expects 5 million to 10 million non-skippable screens on new Android devices each year, plus an initial 15 million to 35 million on desktop and iOS Chrome. Those are device estimates, not unique people.

Demand has moved ahead of distribution. A government survey found that 56 per cent of UK adults had recently used AI that generates text or speech in response to queries, while 54 per cent had used natural-language digital assistants. Ofcom says Google Search still reaches 82 per cent of adults, but ChatGPT recorded 1.8 billion UK visits in the first eight months of 2025.

Choice screens are no guarantee of mass defection. The CMA cites switching of roughly 1 per cent in prior evidence. One field experiment found 1.1 per cent switched from Google to Bing after an active choice, while research on European screens estimated a 0.5 to 1.5 percentage-point fall in Google's share. The CMA's own interviews found limited appetite to leave the familiar default, even though the appearance of AI services made some participants pause and compare.

That makes eligibility the real competitive test. Google would still specify evolving minimum security standards, assess whether an assistant covers the required queries and decide what counts as a meaningful share of its users. Six-monthly application rounds could also make a fast-moving market wait. Reporting each rejection to the CMA, randomising seven of 12 slots and measuring later popularity by selection rather than installation all constrain that discretion, but they do not remove it.

The proposal could create a genuine new route to distribution for AI assistants and a pressure point for publisher attribution. Whether it changes the market will depend less on the neutral wording than on how quickly the CMA challenges Google when the company operating the gate also decides who may pass through it.