The first visibility decision may occur before a user asks anything. On September 23, 2026, the UK Competition and Markets Authority proposed strengthened rules that would require Google to present search-service choices on Android and Chrome, repeat the prompt annually, and allow qualifying AI assistants to appear alongside conventional search providers. The proposal is still under consultation; it is not a completed market change. Its GEO significance is immediate, however. A brand can be highly visible inside one answer engine and still reach few people if that engine is rarely chosen. Measurement must separate platform distribution from source selection inside the platform.

What the CMA proposed, and what it did not

The CMA designated Google as having strategic market status in general search and search advertising in October 2025. In its September 2026 consultation, the authority proposed that Android and Chrome users in the UK receive a choice of search services at setup and then annually. The proposal would permit AI assistants that meet technical and security criteria to participate. It would also require providers shown on the choice screen to attribute publisher content clearly and accurately.

The consultation is open until October 9, 2026, and the CMA said it expected a final decision by the end of the year. Details can change. No responsible article should tell businesses that annual AI-assistant choice screens are already operating across the UK.

The news is still useful because it recognizes AI assistants as potential search services at the distribution layer. GEO strategy has usually started after the user opens a platform. Choice architecture asks an earlier question: which platform receives the query at all?

Visibility is a chain of probabilities

A brand's realized exposure can be approximated as:

`Audience reach x query incidence x answer-surface incidence x brand inclusion x user attention`

For a source citation, add source selection and citation display. For a commercial outcome, add click, conversation, or transaction completion.

Each term describes a different system:

Term	Example question	Typical owner
Audience reach	How many people use this provider as a default or habit?	Distribution and product teams
Query incidence	How often do users ask the relevant question?	Research and category teams
Answer-surface incidence	Does the platform generate an AI answer for the query?	Platform behavior
Brand inclusion	Is the brand named or recommended?	GEO and evidence work
Source selection	Is the company's or another source cited?	Retrieval and source quality
User attention	Does the user hear, read, or open the result?	Interface and content design
Outcome	Does the user take a verified action?	Product, sales, or service operations

Improving brand inclusion from 20% to 40% matters very differently on a platform with 1% category reach and one with 50% reach. A single "AI visibility score" cannot express that difference.

Defaults shape behavior without determining it

A default reduces the effort required to use one service. Users can still switch, open an app, type a different URL, or use several tools for different tasks. The effect of a choice screen depends on its timing, wording, order, eligibility rules, explanations, and how easy it is to change the selection later.

That is why the CMA calls this a question of choice architecture rather than simply adding more logos. A screen can technically offer options while nudging people toward one of them. It can overwhelm users with unfamiliar providers or explain meaningful differences. An annual prompt can remind people that the market has changed, but it can also produce habitual confirmation of the existing choice.

For GEO planning, treat distribution as empirical. Do not assume that inclusion on a choice screen produces adoption, or that a selected default captures every search task.

One person can have several search defaults

The idea of a single search engine is already too simple. A user may rely on:

a browser address bar for navigation;
ChatGPT or another assistant for synthesis;
a marketplace for products;
YouTube for demonstrations and reviews;
Maps for local decisions;
an enterprise assistant for internal work;
a voice agent while driving;
a specialist database for legal, medical, or scientific research.

The effective default depends on the job. A browser choice screen can influence general web search while leaving shopping, local, video, and workplace behavior largely unchanged. GEO content should follow decision journeys across surfaces rather than assigning every query to one winner.

Distribution changes can look like content changes

Imagine a company whose mention rate remains 30% in Assistant A and 10% in Search B. If Assistant A doubles its share of category queries after a distribution change, the company's total AI mentions can rise even though neither platform changed its answer logic and the company published nothing new.

The reverse can happen as well. A strong GEO program may improve inclusion within a platform while total referrals decline because users move elsewhere or the interface reduces clicks. Without decomposition, teams may stop successful content work or claim credit for a market-share shift they did not cause.

Use a bridge analysis:

Change component	Question
Platform mix	Did the share of monitored queries by provider change?
Surface mix	Did more queries trigger summaries, agents, maps, or shopping modules?
Within-platform visibility	Did brand or source inclusion change under fixed conditions?
Interface behavior	Did citation placement or click opportunity change?
Demand	Did the category query volume or intent mix change?
Content	Did a published asset, correction, or evidence event plausibly affect selection?

Only the last row is directly attributable to content, and even then it requires a time-linked test.

Build a platform portfolio, not a universal rank

A platform portfolio starts with audience and task fit.

List the decisions that matter: discovery, comparison, troubleshooting, local visit, purchase, renewal, or expert verification.
Identify which services people actually use for each decision in each market.
Sample the answer surface with fixed prompts and documented account conditions.
Map the evidence formats each service can access: web pages, feeds, videos, maps, private knowledge, or partner data.
Invest in shared facts and platform-specific delivery where justified.
Rebalance quarterly as distribution, product features, and regulation change.

The shared facts should remain stable across the portfolio: entity identity, product attributes, policies, evidence, authorship, and corrections. Delivery may differ. A local business needs accurate map and location data; a video surface needs chapters and transcripts; a shopping agent needs current offers and variants; a research assistant needs primary documents and methods.

Measure platform reach with honest proxies

Exact query share is rarely available. Use a graded evidence stack:

Observed first-party data: referrals, app events, search-console reports, campaign or product logs.
Representative research: surveys with disclosed sample, geography, question wording, and dates.
Platform statements: useful for scale and feature availability, labeled as company-reported.
Panel testing: controlled visibility under known conditions, not population share.
Anecdote: useful for hypothesis generation, never a denominator.

Do not multiply a platform's global user count by a brand mention rate and call the result reach. Users are not queries, global use is not category use, and a mention may not be seen.

Where reach is unknown, present scenarios. For example: if Assistant A accounts for 5%, 15%, or 30% of category decisions, what would a ten-point inclusion improvement change? Scenario ranges reveal sensitivity without inventing precision.

Choice cohorts create a natural experiment

If a choice-screen policy is implemented, researchers may be able to compare cohorts exposed at different times or on different devices. A credible design would need to address selection bias: people who actively choose a new assistant may already have different habits, age, technical confidence, or needs.

Possible measures include:

provider selected at setup;
provider still used after 30, 90, and 180 days;
share of searches sent through the selected default;
use of additional search and AI services;
query categories by provider;
source-click and downstream action rates;
changes in publisher and brand exposure.

Use difference-in-differences, matched cohorts, or randomized interface tests where legally and operationally possible. Do not compare enthusiastic early adopters of an AI assistant with all default-search users and attribute every difference to the product.

Attribution rules become part of provider quality

The CMA proposal says qualifying providers on choice screens should clearly and accurately attribute publisher content. This links distribution eligibility to source behavior. A provider is not judged only on response speed or model quality; the way it represents the web may also matter.

For publishers and brands, monitor attribution at claim level:

Is the source named accurately?
Does the link lead to the supporting page rather than a home page?
Does the cited page actually support the nearby statement?
Is first-party marketing distinguished from independent reporting?
Are dates and corrections visible?
Can a user reach the source in the interface being tested?

A provider can expose many links and still attribute a specific claim poorly. Count support, not decoration.

What organizations should do now

The consultation does not justify a rushed platform bet. It justifies better instrumentation.

Establish the baseline

Record current traffic, mentions, citations, and verified outcomes by provider, surface, market, device, and query type. Preserve the unknowns.

Separate distribution from answer performance

Maintain one dashboard for estimated platform or surface mix and another for within-platform inclusion and accuracy. Connect them in a model without collapsing them.

Preserve portable evidence

Publish facts in accessible pages, structured records, feeds, transcripts, and primary documents that can serve more than one platform. Avoid content that depends on a proprietary widget as its only representation.

Watch final rules and implementation

Use the CMA consultation and final decision as primary sources. Record which devices, browsers, users, and providers are in scope before changing forecasts.

Test new cohorts after rollout

If and when the interface launches, sample actual choice screens and downstream behavior. Do not treat a press release as proof of adoption.

A better executive metric

Replace "average AI visibility" with an exposure-weighted, decomposable view:

Metric	Definition
Platform reach estimate	Share of relevant decision journeys using the provider
Answer eligibility	Share of tracked prompts that produce the relevant AI surface
Brand inclusion	Share of eligible answers that accurately include the brand
Supported inclusion	Included claims with adequate evidence
Source attribution	Supported claims with an accessible, correct citation
Verified outcome	Completed visit, lead, purchase, or service action

Keep the raw components beside any weighted total. When the total moves, the team can see whether distribution, product design, content, or demand changed.

Frequently asked questions

Are AI assistants already included on UK Android and Chrome choice screens?

The September 23 announcement describes a CMA proposal under consultation. It should not be reported as a completed rollout.

Does becoming a default search provider guarantee users will stay?

Users can switch services and use different tools for different tasks. Retention and task-level use must be measured.

Should GEO teams optimize for the largest platform only?

Use platform reach, audience fit, decision value, evidence requirements, and dependency risk. A smaller specialist surface may matter more for a high-value task.

Can referral traffic measure AI reach?

It measures only visits that preserve an identifiable referral. It misses answers without clicks, blocked attribution, app transitions, voice interactions, and later direct visits.

What can be done before the policy is finalized?

Build the baseline, separate platform mix from within-platform visibility, maintain portable evidence, and monitor primary regulatory documents.

Sources and evidence boundary

This article describes an open consultation as of September 28, 2026. It does not predict the final legal requirement, provider eligibility, interface design, adoption, or market effect. The probability model, portfolio framework, and measurement design are analytical recommendations, not findings published by the CMA.