Britain's Competition and Markets Authority today published a revised proposal that would require Google to present UK users with a search choice screen of up to 12 providers on Android devices and Chrome, and the regulator has rewritten the eligibility rules so that AI assistants could, for the first time, qualify for a place on it.

In Short

The UK competition watchdog wants Google to show people a list of search services to pick from when they set up a new Android phone or open Chrome, and it now says AI chatbots could be on that list if they work properly as search tools. This matters because whichever service sits in the default search box captures most of the queries, and therefore most of the search advertising money. If the plan goes ahead, you would also be able to try a rival for at least a week before committing, and you would get a small reminder about your choice once a year.

What the CMA changed

The document, titled "Consultation: User Choice Conduct Requirement", runs to 71 pages. It is the second consultation on the same measure. According to the CMA, the first version was put out for comment on 28 January 2026 as part of a package of conduct requirements under the Digital Markets, Competition and Consumers Act 2024, following the regulator's decision on 10 October 2025 to designate Google with strategic market status in general search services. PPC Land's coverage at the time dated the designation to 30 September 2025; the CMA's own document gives 10 October.

The January package contained four proposed requirements: publisher controls over AI content usage, fair ranking, choice screens on Android and Chrome, and data portability. Only one has so far been imposed. On 3 June 2026 the regulator imposed the Publisher Conduct Requirement, with its substantive obligations coming into force on 3 December 2026 and remaining in force until 10 October 2030.

The choice screen measure has now been sent back for a second round. According to the CMA, the revisions fall into four areas: the eligibility criteria a service must meet to appear on the screen, the process Google must run to decide who qualifies, the design of the "test-drive" and annual redisplay features, and changes to the choice architecture, the Device-Level Default Setting and the search default API. The regulator was careful to state, in a footnote, that the decision to consult again "should not be taken to amount to a precedent" for how it handles future revisions.

Why reopen a measure that was already drafted? The CMA's answer is that the market moved underneath it. According to the document, since January "Google has announced a number of potentially significant changes to the way its search service operates", and the regulator continues to see "signs that ever more users are adopting generative AI-based tools, such as AI assistants, for search". The footnote supporting the first point cites Google's own post, "A new era for AI Search", accessed by the CMA on 28 August 2026.

Why AI assistants are now in scope

The changes Google made to Search between the two consultations are extensive. At its I/O conference on 19 May 2026, Google moved its conversational search surface to Gemini 3.5 Flash and redesigned the Search box, a change the company described as the biggest upgrade to that box in more than 25 years. According to Stijn Huijts, a partner in the competition and antitrust practice at Geradin Partners, writing on LinkedIn today, Google "has upgraded Search with its newest Gemini Flash model, reimagined its Search box, and made it simpler to continue into an AI-assistant style conversation in AI Mode after an initial search." Huijts also noted Google's plan for "Search agents" that can "take action", along with agentic booking features that would initially be limited to Google AI Pro and Ultra subscribers, in some cases only in the United States. Google extended information agents to all AI Mode languages for Ultra subscriberson 12 June 2026.

Those developments undercut one of the arguments made against letting AI products onto the screen. According to the CMA, four respondents to the January consultation - Guardian Media Group, the Computer and Communications Industry Association, the Professional Publishers Association and the News Media Association - argued that because the regulator had excluded Google's Gemini assistant from the scope of the designation, AI services had no place on a search choice screen. Two respondents said AI services would fail to handle navigational queries. OpenAI took the opposite view, arguing that the criteria were not forward-looking enough and pointing out that Google's own search service already runs on AI models and an AI interface.

The regulator sided largely with the second camp. "The line between 'AI-based services' and other types of search service is increasingly blurred," the CMA wrote, adding that it does "not consider that because a service has an AI component or interface it should be excluded from the choice screen." Provided they meet the eligibility criteria, according to the document, "AI assistants could in theory be able to be listed on the choice screen." On the Gemini point, the CMA said its October 2025 analysis rested on Google's products and commercial decisions rather than on a sector-wide assessment, and that the designation decision itself had recognised AI assistants as an emerging competitive threat to Google Search.

The wording matters. Huijts headlined his post with the statement that AI assistants are "to be included" in the requirements. The CMA's formulation is more conditional: AI assistants become eligible to apply, and any individual service would still need to pass Google's assessment against three criteria.

The three criteria that survive

The January draft set four eligibility criteria. The revised draft keeps three and rewrites each of them.

Criterion 1 now requires a service to "adequately address the full range of use cases of a general search service on a wide range of subjects, including by searching the web." Two edits carry weight. "Search engine" became "search service", which the CMA describes as more forward-looking and a reflection of Google's own evolution through AI Overviews and AI Mode. "By searching the web" became "including by searching the web", acknowledging that answers can also come from data feeds and AI models. The interpretative notes define the current full range of use cases as four categories: navigational queries, transactional queries, simple informational queries and complex informational queries. Google had argued for at least nine, according to the document, including location-based queries, image search, news, and validation and accuracy checks.

A provider does not need its own index. According to the CMA, an eligible service must be able to search the web and deliver up-to-date information such as news, but it could rely on a syndicated index. Any generative AI summary must attribute web content "clearly, accurately", with a clear route for users to reach that content, though each provider may choose its own method. Error rates must not be "significant". The notes also observe that adding a word such as "link" to a navigational query may cause a service to return links it otherwise would not, and that Google's testing would need to account for such behaviour.

Criterion 2 swaps "the expected general search experience" for "a functioning general search experience". The CMA dropped the idea of reconciling expectations from users, device makers, publishers and search engines, calling it impracticable to do transparently. A functioning service, in the regulator's reading, works technically on each access point without crashing devices or returning errors, maintains Android compatibility, meets minimum security standards that Google would specify, and can show the ability or "credible plans" to operate on an enduring basis with a growing user base. Where one company offers several services, each must be "sufficiently distinct" at a technical level. The CMA rejected proposals from DuckDuckGo and one anonymous respondent to limit the screen to one service per corporate group.

Criterion 3 now asks whether a service is considered a general search service by "a significant proportion of UK users of the service", rather than by UK users generally. According to the CMA, "significant proportion" means "a meaningful or non-trivial proportion" and "does not mean a majority", with no numerical threshold. Evidence can be attitudinal, such as surveys, or behavioural, such as how people actually use the service.

The criterion that was deleted

The fourth January criterion required a service to be "operated and marketed with general search as a core and central part of the service." Google supported it, according to the document, as a guard against services "free riding" on the screen with a small search component bolted on. Which? said it required "considerable judgement", and Platform Leaders pointed out that AI assistants are not marketed as search engines.

The CMA deleted it for four reasons. It was not forward-looking, given how Google's own service has changed. It risked excluding a service that handled many search queries across all use cases but an even larger volume of non-search tasks. The regulator considers Google's free-riding concern irrelevant given its focus on end users. And the assessment overlapped heavily with Criterion 1. That fourth reason is the one most likely to matter for a general-purpose chatbot, whose search function may be a minority of its traffic.

How the screen would work

Twelve slots, five plus seven, fully random

If 12 or fewer providers qualify, every one of them appears on each screen. If more than 12 qualify, Google must select 12 "fairly". The interpretative notes spell out what fair means: the five most popular eligible providers plus seven drawn fully at random. For the first six months, popularity would be measured by a reasonable metric such as Google Play Store downloads. After that, it switches to a performance measure generated by the screen itself: how often each provider is chosen as a proportion of how often it is shown. Users must also be able to reach the full list from the screen.

This stratified "5+7" structure was Google's own proposal, according to the document, part of what the company called its "UK Choice Package". Other elements differ from the January draft. The list must now appear in "a fully random order", rather than a stratified order with the popular five on top, because the CMA found that lower positions are seen and chosen less often. Users must scroll through the full list before confirming a choice, a mechanism DuckDuckGo had requested. The User Research commissioned by the regulator from Thinks found that many participants recognised only two or three options and did not realise the list extended further down the page.

Each listing would carry a short description agreed between Google and the provider. Where a service charges users or caps free queries, the CMA expects the description to say so - a provision that seems written with subscription AI products in mind. An information screen explaining what a search service is and how the choice affects settings must appear immediately before the choice screen.

Once a user chooses, the selection must propagate. On Chrome, it becomes the address bar default. On an Android device it must become the default search application, the default Search Widget or the default engine behind Google's widget, and the default for the Chrome address bar.

Where it appears

The coverage expands well beyond the current Android screen. According to the draft, non-skippable screens would appear on first use or setup of new Android devices where the manufacturer pre-installs Google Search, a Search Widget or Chrome with Google as default; on first use of any new Chrome installation on Windows, macOS or iOS; and, for existing Chrome users on desktop and iOS, as part of the first major Chrome update after the requirement comes fully into force. Chrome installed on a relevant Android device counts as an access point on that device rather than as a separate browser.

The definitions were also tightened. Whether a device or browser falls under the rule now depends on whether Google Search was pre-installed or set as default from the outset, not on what the user later picked. Someone who switched to a rival years ago would therefore still see the screen. Circle to Search and other Android access points stay outside the scope, despite requests from DuckDuckGo, Microsoft and OpenAI; the CMA said it would monitor the share of searches flowing through those routes.

The test-drive

Every screen on Chrome and new Android devices must offer a "test-drive": setting a provider as default for a short, defined trial. Google may set the length and may let users pick from a range, but the CMA considers that the period "should not be shorter than a week". Two respondents proposed two weeks and another four. A confirmation screen must follow a test-drive selection, because users in the research needed extra explanation of what a trial meant. At the end, users receive a skippable, visually limited notification that takes them to the browser or device default setting. No notification is required if the user deliberately switched during the trial; deleting the search widget from a home screen does not count as a deliberate switch.

Compared with January, the test-drive has been narrowed. It now applies only to new Android devices, plus all Chrome installations on desktop and iOS, because Google said it has limited control over existing non-Pixel devices. The end-of-trial prompt no longer resurfaces the full choice screen, which means a user cannot chain multiple test-drives.

Annual redisplay

The January draft proposed re-showing the full screen every year. Google argued that this would interrupt users, cause "choice overload" and become ignored through repetition, and said it could not technically identify devices set up in the preceding six months. The CMA's compromise: the yearly redisplay becomes a skippable, visually limited notification tied to major Android or Chrome updates rather than to a fixed date, and it applies only to new Android devices plus all Chrome on desktop and iOS. A six-month exemption for users who recently made a deliberate choice remains in the text, in case Google finds a technical route. The option to re-trigger the screen from settings at any time has been removed.

Google would have to tell eligible providers at least three months ahead of the initial rollout and of each annual redisplay, so that, in the CMA's words, they "can prepare marketing campaigns in advance". Brave had asked for a fixed annual choice screen day.

Google runs the gate

Google currently decides which providers appear on the Android choice screen and on the screen it shows under the European Union's Digital Markets Act. The CMA will keep that arrangement, despite submissions from Mozilla, Which?, Platform Leaders and others calling for oversight, and one from the Knight-Georgetown Institute arguing that the regulator, not Google, should decide applications.

The concession is procedural. The draft requires Google to run an approval round every six months against the eligibility criteria only. Public notice must go out two weeks before the application window opens. Final decisions, with written confirmation to each applicant and to the CMA, are due within two months of the window opening, and rejections must carry detailed reasoning for the regulator. Applicants cannot be required to hand confidential information to Google. Existing eligible providers do not reapply each cycle, although Google may reassess one it doubts. This approval obligation would come into force one month after the requirement is imposed, while the rest would follow within six months, so that the list of eligible providers exists before the screens go live.

Settings, APIs and prompts

Three further obligations shape the plumbing around the screen.

On Android, Google must build a Device-Level Default Setting that changes the default across the widget, the search app and Chrome in a single step. Google had offered a central switch covering only the widget, arguing that propagating to Chrome would confuse users, according to the document. The CMA disagreed and kept its version, along with an expectation that searching the settings menu for terms like "default" or "default search" surfaces it. Where the setting only lists installed apps, Google must link to the full list of eligible providers, for example via the Play Store.

Eligible providers would receive an API that tells them whether they are a given user's default on Chrome or on an Android device, plus the ability to deep-link users directly into the relevant default setting. Sony opposed the API on privacy grounds, according to the CMA; the regulator's answer is that it could be built on the same basis as an existing Android browser API, with user consent likely obtained at setup. Google had said it would not restrict in-app prompting by rivals, and the CMA removed its earlier suggestion of limits.

Google's own prompts about default search must be "fair and balanced". Notifications about changes to default settings are permitted, but they must not ask users whether they want to switch to Google Search. One respondent asked the CMA to rule out dark patterns, and DuckDuckGo singled out Google's extension-based "switch-back" prompt. According to the document, Google said it is rolling out an updated "search hijacking confirmation dialogue" that it describes as "fully neutral". The regulator made no changes to this section.

The cost-benefit arithmetic

The proportionality section is where the numbers sit, many of them redacted into ranges.

On costs, the CMA kept its January estimates for Google: around £5 million over five years to extend the screen to Chrome on desktop and iOS, and up to £5 million for the test-drive. Google argued those figures were "preliminary and conservative" because they exclude server maintenance and technical fixes, and that annoyance from choice screens could push users towards Apple and Microsoft. It cited EU users losing the Discover feed under DMA choice screen arrangements, and estimated that if one in 1,000 Android users moved to iOS it could cost Google between $5 million and $15 million over five years. The CMA noted that Google produced no evidence that choice screens influence operating system or browser choice.

The time cost to users was recalculated. In January the regulator valued each showing at a little under 14p using the average UK wage. It now uses the Department for Transport's value of non-working time, £7.21 per hour, which puts the average 25-second journey through the screen at a little over 5p. Google's own research, with between 20 and 30 participants, put the setup screen at 10 to 20 seconds without a test-drive and said the test-drive button added 10 to 20 seconds. The CMA said it had not seen the methodology and contrasted the sample with the 12,000-participant Mozilla study it relies on.

Volumes are large. The CMA estimates between 5 million and 10 million non-skippable screens a year on Android, an initial 15 million to 35 million on Chrome for desktop and iOS, then 5 million to 15 million a year. Redisplay notifications would add 65 million to 135 million over five years. Google's counter-estimate, which counts every new Chrome profile, is 50 million to 80 million screens a year. Table 1 of the document puts total quantified five-year costs at £5 million to £15 million for the screens and redisplay combined, and £0 to £5 million for the test-drive, with an overall ceiling of £5 million to £20 million.

The benefit side rests on two claims. The first concerns consumer welfare from finding a better-matched service. The January figure of £149 per device was challenged by Google as biased towards existing Bing users. The CMA has replaced it with two lower estimates derived from the same Allcott et al. field experiment - £4.22 per device unadjusted and £14.77 adjusted for the US population - while noting Google's own commissioned research found the median user would need £39 a month, or £468 a year, to give up Google. Evidence points to switching rates of roughly 1% from choice screens. Allcott et al. found 1.1% moved from Google to Bing after an active choice; Decarolis et al. found Google's EEA share fell by 0.5 to 1.5 percentage points after screens arrived from March 2020; DuckDuckGo's 2020 research suggested a well-designed menu could cut Google's mobile share by around 20%.

The second, and larger, claim concerns competition. According to the CMA, Google has more than 60 million to 70 million logged-in mobile users in the UK, plus 20 million to 30 million on desktop, and generated between £10 billion and £20 billion in UK search advertising revenue in 2024. If increased competition cut advertising prices by between 0% and 0.1%, the CMA calculates, the benefit to advertisers alone would exceed the total cost of the requirement. Its calculations assume a one-year build, four years of benefits and a 3.5% discount rate from the Treasury's Green Book. Google maintains that the costs "still considerably outweigh the benefits", according to the document. The CMA's provisional view is that the benefits are "likely to significantly outweigh the costs".

What the plan does not do

A coalition of publishers and campaigners - the Independent Media Association, the Independent Publishers Alliance, Impress Media and the Movement for an Open Web, represented by Preiskel & Co - asked the CMA to widen the requirement so that users could control whether AI Overviews appear automatically in results. The regulator declined. It said it had not received "sufficiently strong evidence" of the problems such a control would address, and that the measure's aim is choice between providers rather than control over features within Google. It will keep the question under review.

For publishers, the June Publisher Conduct Requirement remains the relevant instrument. Google tested a Search Console toggle on 3 June 2026 letting sites exclude their pages from AI Overviews and AI Mode. The attribution language in Criterion 1, however, sets a floor for rivals too: any AI assistant hoping for a slot on Google's screen would need to link clearly to the web sources behind its answers.

Why this matters for marketers

Defaults decide where queries go, and queries decide where search budgets go. The CMA's own figures describe a UK search advertising market worth tens of billions of pounds concentrated on one platform; when the regulator designated Google in 2025, it found Google handled more than 90% of UK general search queries and that 200,000 to 300,000 entities bought search ads in 2024. Any measurable shift of default status towards other services - including AI assistants that sell or plan to sell advertising - would redistribute a portion of that inventory.

European evidence tempers expectations. Research covered by PPC Land found that DMA choice screens added roughly 6 million Firefox users while leaving Google Search near 90% share in the EU. The CMA's own switching assumption of about 1% is consistent with that picture. The UK draft differs from the EU model in several technical respects: it reaches Chrome on desktop, applies to existing Chrome users through an update, adds an annual reminder and a test-drive, randomises order fully, and gives rivals an API and deep links into settings. Whether those additions move share beyond the European pattern is the open empirical question.

The regulatory direction on both sides of the Channel is similar. On 16 July 2026 the European Commission adopted decisions that require Google to open Android to competing AI assistants and share anonymised search data. The EU has also shown the limits of choice screen obligations: on 2 September 2026 the General Court confirmed Microsoft Edge need not display one, citing its 5.8% share of European page views. Meanwhile smaller engines are working to get onto default lists by other routes; Brave Search secured a place on Chrome's global engine list in July 2026.

For advertisers running search campaigns in the UK, the practical timeline is long. Even if the CMA imposes the requirement shortly after this consultation, the approval round would start a month later, the screens would follow within six months, and redisplay notifications would begin only in the third year of the CMA's model. Google would report to the regulator every six months in the first year and annually thereafter, publishing non-confidential versions. New required metrics include how often users open the full provider list and, for each provider, how often it appeared in each position and how often it was chosen from there. Those figures, if published, would give the market its first detailed view of how UK users respond when an AI assistant sits next to Google in a random list.

Next steps

Responses are due by 5pm on 9 October 2026, to searchsms@cma.gov.uk or through the consultation portal - a window of 16 days. The CMA is asking stakeholders whether they agree with the revised eligibility criteria, the procedural safeguards, the frequency and timing of the screen, the test-drive design, the selection and ordering of providers, the link to the full provider list in settings, third-party access to default settings, and the updated proportionality assessment. It expects to publish responses, subject to confidentiality claims, before making a final decision under section 24 of the Act.

Timeline

Summary

Who: The UK Competition and Markets Authority and Google, with eligible rival search providers, AI assistant operators, Android device makers, UK users of Android and Chrome, publishers and search advertisers affected. Stakeholders cited in the document include OpenAI, DuckDuckGo, Mozilla, Microsoft, Samsung, Sony, Brave, Which? and publisher trade bodies.

What: A revised draft User Choice Conduct Requirement that would oblige Google to show a search choice screen of up to 12 randomly ordered providers on new Android devices and Chrome across desktop and iOS, with a test-drive of at least a week, a yearly skippable reminder, a single device-level default setting, APIs for rivals and a six-monthly approval process run by Google. Rewritten eligibility criteria allow AI assistants to qualify if they meet them, and the "core and central" marketing criterion has been deleted.

When: Published today, with responses due by 5pm on 9 October 2026. If imposed, the approval process would begin one month later and the remaining obligations within six months.

Where: The United Kingdom, covering Android devices meant for UK distribution and Chrome on Windows, macOS, Android and iOS where Google Search is set as default.

Why: According to the CMA, Google's changes to Search since January and the growing use of AI assistants for search required the measure to be technology neutral. The regulator estimates five-year costs of £5 million to £20 million against potential benefits it considers far larger, citing UK search advertising revenue of £10 billion to £20 billion in 2024.