TL;DR: AI shopping agents are moving from research assistants to checkout surfaces. Google and OpenAI/Stripe have each published an open protocol for agent-led purchasing, and Adobe forecasts a 130% rise in AI-driven traffic to US retail sites this holiday season. Neither protocol is a UK launch announcement, but the groundwork they demand (clean product data, accurate pricing, honest promotions, clear returns) is the same work UK retailers need to do anyway, and the CMA is already enforcing the consumer-law side of it. Our view: do not chase a checkout integration first. Fix the data and the pricing hygiene first.
Why this matters now
For two decades, UK ecommerce teams optimised for a human with a browser: a search ranking, a product page, a basket, a checkout. Agentic commerce changes who is on the other side of that flow. An AI assistant researches, compares and, increasingly, helps complete the purchase on the shopper's behalf.
The numbers are still modest but the direction is clear. Adobe reported that traffic to US retail sites from generative AI tools rose 693.4% year on year across the 2025 holiday season, while noting that the base of users "remains modest" (Adobe, January 2026). For 2026, Adobe projects AI traffic to US retail sites to grow 130% year over year, with the biggest increases forecast for Thanksgiving Day (159%), Black Friday (95%) and Cyber Monday (88%), within a record $275.1 billion online holiday forecast (Digital Commerce 360, 29 September 2026).
Those are US figures, and we are not going to pretend otherwise. We have not found equivalent UK figures we can verify, so we will not quote any. What UK retailers can reasonably take from them is the shape of the curve: small base, fast growth, concentrated on peak trading days.
What has actually been launched
There is a lot of noise about "agentic commerce". Two concrete things have been published.
OpenAI and Stripe: Instant Checkout and the Agentic Commerce Protocol (ACP). On 29 September 2025, Stripe described ACP as an open standard co-developed with OpenAI that creates a shared language between businesses and AI agents (Stripe). At launch, US ChatGPT users could buy from US Etsy sellers, with Shopify merchants listed as coming soon. Per VentureBeat's report, merchants remain the merchant of record through fulfilment, returns and support, can accept or decline orders, and pay a small fee on completed purchases, with no amount stated (VentureBeat). Stripe says businesses not processing with Stripe can still adopt the standard.
Google: the Universal Commerce Protocol (UCP). Announced on 11 January 2026, UCP is described by Google as an open standard for agentic commerce across discovery, purchase and post-purchase support. Google says it was co-developed with Shopify, Etsy, Wayfair, Target and Walmart, and is compatible with existing protocols including Agent2Agent, Agent Payments Protocol and the Model Context Protocol (Google). Google said UCP would "soon" power checkout on eligible listings in AI Mode in Search and the Gemini app for eligible US retailers, with retailers remaining the seller of record, and that it would work with retailers to expand globally in the coming months.
Notice what is common. Both are US-first. Both keep the retailer as the merchant or seller of record. Both depend on the agent being able to read accurate, structured information about what you sell and on what terms. That last point is the one UK retailers can act on today.
The consumer is more cautious than the hype suggests
It is worth tempering the narrative with what shoppers say they want. Research from Commerce and PayPal Australia, fielded by Logica Research among 1,000 Australian online shoppers, found 63% interested in trying an agentic AI shopping tool, but only 4% wanting AI involved at the point of payment (IT Brief Australia, 26 June 2026). One in five already use AI tools when shopping online, and 79% expect agentic tools to match or exceed today's payment-security standards. In the words of Commerce's APAC lead, shoppers are "warming up to the idea of AI helping them shop smarter" but "not ready to let it shop for them just yet".
This is Australian data, not UK data. We cite it because it is the best verified signal we found on shopper appetite, and because it matches what we would expect: assistance in discovery is welcome, delegated payment is a trust leap. For UK retailers it suggests the near-term prize is being the product an assistant recommends, not winning a fully automated purchase.
Our point of view: the storefront is becoming your data
Here is the position we take with clients. In an agent-mediated journey, your product detail page is no longer the first thing a buyer sees. An agent reads your catalogue, your price, your delivery promise and your returns terms, then summarises them. If those are inconsistent, thin or wrong, the agent either skips you or misrepresents you.
An analyst quoted by American Banker put it bluntly in the context of Google's launch: product detail pages are "the fuel" the agentic engines need, and if checkout moves into the assistant, the merchant "loses the last touch point" (American Banker). Those are one analyst's opinions, but they point at a real strategic tension: you gain reach and lose some direct relationship.
So the question for a UK retailer is not "which protocol do we implement?" It is "if an agent read our catalogue and policies right now, would it describe us accurately?"
Four AI use cases worth your attention
1. Machine-readable catalogue and policy data. Titles, attributes, variants, stock, delivery times and returns terms held as structured, consistent data rather than marketing prose. This is the prerequisite for any agent integration and it also improves conventional search and marketplaces. A headless or API-first commerce architecture makes this easier because the same product and policy data feeds every channel; see our work on headless commerce.
2. AI-assisted customer service for post-purchase. Google's UCP explicitly spans post-purchase support, and shoppers rate returns highly: in Australia Post's October 2026 quarterly update, 75% of shoppers said an easy returns process makes them more likely to buy again (Australia Post). Again, Australian data, but the lesson travels. Order status, returns initiation and warranty questions are bounded, high-volume tasks where an assistant connected to your order systems can help. See AI chatbots and assistants and our returns and warranty hub.
3. Measuring AI-referred demand. If AI assistants send you traffic, you need to see it. That means tagging referrers, separating the segment in analytics, and watching conversion and basket behaviour for it separately. We have not verified published conversion benchmarks for AI-referred traffic, so we would not plan around any; measure your own. Our data analytics and BI team builds this kind of reporting.
4. Personalisation and lifecycle marketing. Customer data platforms and automation remain the unglamorous place where AI pays back first, because you own the data and the channel. See marketing automation and CDP.
The risk most teams underweight: UK consumer law
If you sell to UK consumers, the regulatory clock is already running, independent of any AI roadmap.
The Competition and Markets Authority gained direct consumer enforcement powers under the Digital Markets, Competition and Consumers Act 2024, in force from April 2025. In its one-year review the CMA said its public enforcement had targeted three areas: drip pricing, fake reviews and online choice architecture, including default optional charges and false "time limited" offers (CMA, 17 April 2026).
The cases are concrete. According to the CMA and law firm Cooley's October 2026 summary, AA Driving School and BSM Driving School admitted drip pricing, with refunds over £760,000 and a £4.2 million fine; Marks Electrical settled over consumers being automatically opted in to services (£600,000 redress, £720,000 fine); and StubHub UK settled over mandatory fees not shown in the headline ticket price (£590,000 redress, £889,200 fine) (Cooley). The CMA also reported advisory letters to 54 firms in a fake-reviews sweep and opened investigations into five companies in March.
Now the AI link. The CMA lists how businesses use AI, including agents, as a year-two priority and points to its guidance on complying with consumer law when using AI agents. Neither source we reviewed describes an AI-specific investigation yet, so treat this as direction of travel, not precedent.
Why does this matter for agentic commerce? Because an agent will surface the total price, the fees and the terms you publish, and will do so at scale. A dark pattern that a hurried human might miss is something a machine reads literally. Drip pricing, false urgency and misleading discounts become both a legal risk and a discoverability risk. We are not lawyers; take advice from your own counsel, particularly on promotions and subscription terms, where our sources disagree on timing of new rules (the CMA blog says spring 2027, Cooley says January 2027).
Risks to name plainly
- Misrepresentation by agents. An assistant may summarise your offer inaccurately. You cannot fully control it, but you can reduce the surface area by keeping data consistent.
- Loss of the customer relationship. If checkout happens inside an assistant, you may see less first-party data. Decide what you need back through the protocol and what you will accept losing.
- Fraud and authorisation. Delegated payment raises questions about who authorised what. Stripe's design uses scoped payment tokens tied to a merchant and cart total; whatever route you take, plan for disputes and chargebacks involving agent-initiated orders.
- Fees and dependency. Reported fees exist but amounts were not published in the sources we reviewed. Do not model margins on guesses.
- Over-building. Both protocols were launched US-first. Building bespoke UK integrations before availability is confirmed is a waste of runway.
What to do this quarter
- Audit your catalogue as an agent would. Take your top 50 products and check that title, attributes, price, stock, delivery time and returns terms are present, consistent and machine-readable across your site, feeds and marketplaces.
- Run a pricing-and-promotions review against CMA themes. Is the total price clear from the first screen? Are optional extras opt-in? Are "limited time" claims true? Are reviews genuine and clearly handled? Fix before peak trading.
- Add an AI-referral segment to analytics. Tag known assistant referrers and start a baseline now, before the holiday peak, so you have your own numbers rather than someone else's.
- Pilot one bounded assistant use case. Order status or returns initiation is a sensible start. Keep a human fallback and log failures.
- Pick a watching brief on checkout protocols. Assign one person to track ACP and UCP availability for UK merchants and revisit next quarter. Integrate only when there is a confirmed route and commercial terms you can accept.
- Write your agent policy. Decide what you permit, what data you share, and who owns incidents.
Where Syslabs fits
Syslabs is a software and AI consultancy that builds ecommerce platforms, integrations and analytics. In this area we typically help with catalogue and API architecture, assistant pilots connected to order and returns systems, and reporting for new traffic sources. Scope and pricing depend on your stack and are agreed on a consultation, and we will tell you when the right answer is to wait.
Next step
If you want a practical review of your catalogue readiness, pricing hygiene and AI pilot options, book a consultation. More perspectives are in our AI insights library and on our ecommerce industry page.
