What happens to your website when AI builds the interface?

A brand can spend months building the perfect product comparison page, only for an AI assistant to present its own comparison before the customer reaches it. The question is not simply whether the website loses a visit. It is whether the evidence that makes the brand worth choosing survives a change in who organises the decision.

On 7 October, OpenAI announced GPT-6 with Intelligent UI, enabling ChatGPT to combine answers with graphics, tappable buttons, forms, charts and interactive experiences. Its examples include an explorable 7-speed bicycle explanation and tools created within a conversation, such as a savings calculator or bill splitter. Rollout began globally for Plus, Pro, Business and Enterprise tiers, with expansion to Free and Go starting on 8 October. Those dates describe the start of a rollout, not proof that every account or response now uses the capability. [1]

For marketers, the important development is that AI can increasingly determine how information is organised and explored, as well as which information appears. That extends the AEO task beyond being visible in a response. The reasons to choose a brand need to remain understandable when the brand no longer controls the presentation.

What the interface announcement establishes

OpenAI says ChatGPT can choose an appropriate format and create tools for the task at hand. This is a capability in the Chat experience, not a declaration that every answer will become an application. Enterprise access depends on administrator settings, and the Pro reasoning option does not support Intelligent UI. The announcement does not establish that customers have stopped visiting websites or that interactive answers improve conversion. [1]

Some decision support could move into the assistant. A customer who wants an explanation, calculator or comparison may receive it without opening the branded equivalent. That is a plausible implication of the capability, not a measured change in purchasing behaviour. The earlier analysis of websites moving down the funnel examined traffic and advertiser-defined outcomes; this story concerns who constructs the experience in which a customer evaluates the evidence.

The retailer can keep the sale while the platform hosts the decision

Separate commerce developments show how presentation and commercial responsibility can divide. OpenAI's 24 March product-discovery announcement described side-by-side shopping comparisons and product information supplied through the Agentic Commerce Protocol, or ACP. It also said merchants could use their own checkout experiences while OpenAI focused on discovery. Those shopping capabilities are not features that the October Intelligent UI launch automatically enables. [2]

On 8 October, Newegg announced that US shoppers could buy a growing selection of eligible products directly inside Google AI Mode and Gemini through Universal Commerce Protocol checkout. The company says checkout happens in Google's experience through Google Pay, while Newegg remains the merchant of record and fulfils and supports the order. This is a company-reported live integration for eligible products, not evidence that the entire catalogue is available or that it increases sales. [3]

Together, these developments illustrate a possible division of the customer journey: a platform can host discovery and evaluation while the retailer retains transaction and service responsibilities. As the AI shopper analysis explained, recommendation, access and checkout are distinct stages. Keeping the order does not mean controlling the criteria used to win it.

The risk is losing the reason to choose the brand

Imagine a customer comparing three washing machines for a small flat. An assistant might organise the comparison around price, size and energy use. One brand's quiet operation matters only if accurate noise evidence enters the comparison and the assistant recognises its relevance. A carefully designed website cannot guarantee that the assistant will retrieve that fact or use it appropriately.

This is a hypothetical example, not a test of actual products. It highlights a strategic problem: an assistant may choose the comparison set, select attributes and help a customer eliminate options. A brand can appear in the answer while its most relevant advantage is omitted, misunderstood or applied to the wrong use case. Visibility alone would not reveal that failure.

The System 0 lens describes AI structuring options and evidence before a person chooses. Here it is an interpretation of the interface change, not proof that people are surrendering decisions. It also makes Machine Positioning more tangible: the degree to which intended brand positioning is faithfully and advantageously represented in AI-mediated decisions. That remains a diagnostic concept, not a separate positioning strategy, validated score or substitute for research into human choice.

Six things brands should do now

These are AEO Updates recommendations for evidence and measurement, not six interventions tested by the announcements. They extend the seven-action AEO operating model to the practical problem of AI-built presentation.

1. Publish the facts needed to make a choice

Audit the questions customers use to compare the brand with alternatives. Publish clear answers about specifications, compatibility, limitations, price, service and suitability, stating which product, market and date each fact applies to. If an important fact exists only inside a widget or image, provide an accessible textual equivalent. Google's guidance for its own AI search features supports making important content available as text, matching structured data to visible content and keeping merchant information current. It says no special AI schema or new AI text file is needed. These are Google Search guidelines, not universal rules for every assistant or a guarantee of inclusion. [4]

2. Make differentiating claims verifiable

Replace vague superiority language with evidence a reader can check. Explain the test conditions behind performance claims and the limits of comparisons. Quiet operation is more useful as a reason to choose when it has a measurement, a method and a relevant use case. The objective is to make a defensible advantage understandable wherever it is presented, not to promise that particular wording will increase rankings or recommendation frequency.

3. Check distribution as well as the website

Keep product pages, authorised merchant feeds and partner listings consistent. OpenAI's product-feed documentation separates discovery from Ads and checkout requirements, states that eligibility does not guarantee display, and requires checkout to be enabled through a separate integration. Variant identity, price and availability need particular attention because each variant can carry its own offer details. Approved-partner feed onboarding does not mean every merchant or checkout is immediately available. [5]

4. Make crawler access an explicit business decision

Check whether the content intended for discovery is accessible to the relevant search system. OpenAI distinguishes allowing OAI-SearchBot for ChatGPT summaries and snippets from GPTBot controls for potential training. Those are different access decisions. Make them deliberately with the teams responsible for content, rights and technical policy, rather than assuming that one robots.txt decision governs every use of the material. Access is necessary in the documented search flow, but it does not guarantee that a claim will appear. [6]

5. Test the experience customers actually receive

Use a stable panel of realistic decision questions across relevant models, markets and use cases. Measure visibility and recommendation separately, then inspect whether key claims are accurate and relevant advantages survive the response. Include human review of comparisons, controls and interactive presentations in the consumer product. Standardised text or API observations remain useful, but a text response alone cannot show how the consumer interface arranged the options. Record the interface, account tier, prompt and observation date so changes are interpretable rather than treated as one universal result.

6. Keep commercial evidence separate from visibility

Track referrals, qualified enquiries and sales where attribution permits. OpenAI documents utm_source=chatgpt.com in referral URLs to help identify inbound traffic. That tracks visits produced by referral links; it does not reveal choices made without a click. [6] Use customer research, longitudinal observation or controlled tests to investigate whether exposure changes consideration and choice. Rising visibility is a diagnostic result, not proof of incremental revenue, and unlike measures should not be directly subtracted or blended into a commercial score.

AEO Updates Takeaway

Brands should continue improving their websites, especially where customers need authoritative detail, reassurance, account access or service. The priority is not to abandon the branded experience. It is to reduce dependence on one presentation of the facts by maintaining reliable product information and defensible claims that can support a website, merchant feed, assistant response or partner experience. Commission expensive custom AI experiences when a demonstrated customer need and distribution opportunity justify them, not simply because the technology exists.

Intelligent UI makes the question more urgent: if an assistant builds the interface, can the customer still understand what makes the brand the right choice? AEO should protect the evidence behind that understanding, then test whether it survives the experience and matters to the person making the decision.

References

[1] OpenAI, GPT-6 and Intelligent UI for everyone, 7 October 2026; corroborated by the 7 October entry in ChatGPT release notes.

[2] OpenAI, Powering Product Discovery in ChatGPT, 24 March 2026.

[3] Newegg, U.S. shoppers can discover and purchase eligible Newegg products in Gemini and Google Search, 8 October 2026. Company announcement, also distributed by Business Wire.

[4] Google Search Central, AI features and your website, accessed 9 October 2026.

[5] OpenAI Developers, Products: Agentic Commerce and Get Started, accessed 9 October 2026.

[6] OpenAI Help Center, Publishers and Developers FAQ, accessed 9 October 2026.