Agentic Commerce: When AI Becomes the First Touchpoint in Your Customer Journey

An AI agent is increasingly the first thing that meets your brand, and it does not care about your homepage. It reads your data, checks your stock, compares your price, and moves on. Agentic commerce is the name for that shift, and this guide explains what it means for a UK business that sells online.

What Agentic Commerce Actually Means

Agentic commerce is a model of buying and selling in which autonomous AI agents research products, compare options, and complete transactions on a shopper's behalf, with limited or no direct human involvement at each step. The shopper sets a goal ("find me a waterproof dog coat for a medium collie, under £60, delivered before Saturday") and the agent handles the browsing, filtering, and often the checkout itself. Wikipedia files it under agent-based commerce, with agentic commerce as the more common name. It sits a step beyond the AI-assisted shopping most people have already used. A chatbot that answers questions is assistive. An agent that holds a budget, applies your constraints, and pays is agentic.

The distinction matters commercially because it changes who your marketing is aimed at. In assisted commerce, AI helps a human decide and the human still lands on your site. In agentic commerce, a machine shortlists on the human's behalf, and if your product never enters the shortlist, no amount of on-site persuasion helps. You are no longer optimising a page. You are optimising an input.

Three terms travel with it, and they are worth separating because suppliers use them loosely:

• Agentic commerce is the whole model: discovery, comparison, purchase and after-sale, carried out by an agent.
• Agentic payments are the narrower problem of letting an agent pay safely, using scoped, revocable credentials instead of a card number.
• Agentic commerce protocols are the shared rules that let any agent talk to any shop: here is the product, here is the price, here is proof I am allowed to buy it.

Where Agentic Commerce Stands for UK Sellers in Late 2026

Three things have converged. The models got good enough to handle multi-step tasks without losing the plot. The payment networks built rails that let an agent transact without a human re-entering card details. And the retail platforms opened structured product access so an agent can read live inventory instead of scraping a rendered page.

2026 has been the year that work reached real products, unevenly. Google launched the Universal Commerce Protocol on 11 January 2026 as an open standard covering discovery, buying and post-purchase support. Shopify's Agentic Storefronts, launched in December 2025, put merchants' products into ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta, and Shopify says the feature is active by default for eligible stores. OpenAI went the other way on checkout: in March 2026 it said the first version of Instant Checkout "did not offer the level of flexibility that we aspire to provide", and that it would let merchants use their own checkouts while it concentrated on product discovery.

The detail that matters most for a UK business is geography. In Shopify's own channel requirements, none of those AI checkout channels sells to UK buyers yet: ChatGPT needs a store that sells to customers in the United States, Copilot is US-only, Google covers the US, Canada and Australia, and Meta a list of six countries that does not include the UK. Google has said its UCP-powered checkout will reach Canada and Australia "in the coming months and later to the U.K." So in November 2026 a UK shopper can ask an assistant to find and compare products, but in most cases cannot yet have it pay you.

Discovery is the part that is live now, and the audience is large. The government's DSIT Public Engagement Survey 2025/2026 found that 59% of UK adults had used a form of generative AI in the previous three months. Business interest has cooled from its peak: UK searches for "agentic commerce" ran at about 1,900 a month from January to March 2026, around the UCP launch, and have settled at roughly 900 to 1,000 a month since the summer. Our read is that most of that volume is businesses working out what to do, not shoppers using agents every day.

The practical read for a UK SMB is this. Agent-led purchasing is not yet a meaningful share of your revenue, and anyone telling you otherwise is selling something. The work that makes you visible to agents, however, is the same work that improves conversion, feed quality and organic search today, and it is what decides whether an assistant recommends you during discovery. That is why it is worth doing now rather than later: the prerequisites pay for themselves regardless of when agent checkout arrives in the UK.

Six Shifts Agentic Commerce Forces on UK Brands

Discovery Moves From Search to Delegation

A shopper who delegates never sees a results page. The agent queries structured sources, weighs them against stated constraints, and returns a shortlist of two or three. Your job shifts from ranking on a page a human scans to being machine-legible enough to survive the filter. That means clean schema markup, unambiguous product attributes, and specifications written as data rather than buried in marketing prose.

Your Product Feed Becomes the Storefront

For agent-mediated purchases, the feed is the shop. Missing sizes, vague materials, inconsistent titles, and empty GTIN fields are no longer cosmetic problems: they are the reason an agent silently excludes you. If your Google Merchant Center feed has warnings you have been ignoring for two years, that is now a revenue issue rather than a housekeeping one.

Price and Availability Go Real-Time

Agents check before they commit. A stale price or a product listed as in stock that is not will get the transaction abandoned and, worse, will teach the agent that your data is unreliable. Cached feeds updated nightly were acceptable when a human was doing the checking. They are not acceptable when a machine is comparing you against a competitor whose stock updates in seconds.

Reviews and Reputation Feed the Decision

Agents lean on third-party signals because they are harder to game than on-site copy. Review volume, recency, and the specific content of reviews all feed the comparison. Detailed recent reviews give an agent attribute-level evidence ("true to size", "survived a wet winter") that a pile of old one-line ratings cannot. Under UK law since April 2025, fake or concealed incentivised reviews are banned outright, so the only safe way to build this signal is to ask every genuine customer.

Checkout Gets Tokenised

Agent-led payment does not work by handing a card number to a bot. It works through delegated, scoped payment credentials that carry limits and permissions, so the merchant can verify the agent is authorised to spend on that customer's behalf. If your checkout cannot accept a tokenised agent payment, the agent will complete the purchase somewhere that can.

Loyalty Becomes Agent-Mediated

When a machine buys on repeat, emotional loyalty stops doing the work. The agent re-evaluates every time unless something anchors it: an account-linked discount, a subscription, a genuinely superior returns policy, or a data relationship it can carry forward. Brands that rely on habit and familiarity will find both are surprisingly easy for software to override.

Agentic Commerce Protocols: ACP, UCP and AP2 Explained

Agents and merchants need a shared language for "here is the product, here is the price, here is the authorisation to pay". Several open standards now provide it, and the names come up in every supplier conversation, so it helps to know what each one is for.

• Agentic Commerce Protocol (ACP). The Agentic Commerce Protocol is an open standard for programmatic commerce flows between AI agents and businesses, developed by Stripe and OpenAI and published under the Apache 2.0 licence. It is the route by which merchants reach shoppers inside ChatGPT.
• Universal Commerce Protocol (UCP). Google's open standard, published at ucp.dev under Apache 2.0 with retailers and platforms including Shopify, Etsy, Wayfair, Target and Walmart listed as co-developers. It covers the whole journey from discovery to post-purchase support, and powers checkout in AI Mode in Google Search and in Gemini. Google's merchant guide is explicit that "You remain the Merchant of Record", so the sale, the customer contract and the consumer-law duties stay with you.
• Agent Payments Protocol (AP2). Announced by Google Cloud in September 2025, AP2 is a payment-specific layer for enabling AI agents to make payments on behalf of users securely. It is the agentic payments piece rather than a full commerce protocol.
• Card network and wallet frameworks. Visa Intelligent Commerce and PayPal's Agent Ready and Store Sync sit underneath, handling the question of whether an agent is allowed to spend a particular customer's money.

Agentic commerce protocol vs universal commerce protocol is the comparison most people search for, and the honest answer for a small merchant is that it is not your decision. ACP is the route into ChatGPT; UCP is the route into Google's AI surfaces. Your platform and payment provider implement the protocols, and a store on a mainstream platform will usually be exposed through both. You do not need to pick a winner, and you almost certainly should not try. What matters is the layer underneath, which every one of these standards assumes you already have:

• A product catalogue exposed through a stable API or a well-formed feed, not only through rendered HTML.
• Inventory and pricing that are accurate at request time, not at last-sync time.
• A payment stack capable of accepting delegated, scoped credentials rather than raw card details.
• Order, fulfilment and returns states that can be queried programmatically after the sale.

Get those four right and you are compatible with whichever protocol wins. Skip them and no amount of protocol adoption will help, because there is nothing underneath for the agent to read.

Who Is Actually Building This

The activity is concentrated in three groups, and knowing which group a supplier sits in tells you how seriously to take their timeline.

• The AI platforms. The companies that own the assistant surface (and therefore the customer relationship) are building shopping into the conversation. They also change course fastest: OpenAI's step back from its own checkout in March 2026 is the clearest example, and a good reason not to build anything bespoke around a single assistant.
• The commerce platforms. Shopify and its peers are exposing structured catalogue and checkout access so merchants become agent-readable by default. On the question of cost, Shopify's channel pages say there are no fees for selling through ChatGPT, Copilot, Google or Meta beyond standard payment processing, and its separate Agentic plan for stores on other platforms has no monthly subscription fee. If you are on a mainstream hosted platform, a meaningful share of this work lands on you through platform updates, which is a genuine argument for not building bespoke. If you are weighing a platform move for this reason, our Shopify web design page sets out how we approach it.
• The payment networks. Card networks, wallets and processors have all launched frameworks for agent-initiated payment, because the hard problem in agentic commerce is not finding the product, it is proving that the machine is allowed to spend the money. Expect this layer to mature fastest, since the networks already have the fraud and authorisation infrastructure.

For a UK SMB the honest implication is that most of this will arrive as a platform feature rather than a project you commission. Your leverage is in data quality, which no platform update can fix for you. Shopify is the clearest example of that pattern so far, having given Agentic Storefronts a dedicated dashboard in the merchant admin. It has since added Meta to that list of AI channels with product sharing switched on by default, so every merchant already has a setting to review, even before UK buyers can check out through any of them.

The Risks of Agentic Commerce, and What the CMA Expects

The UK has no agent-specific consumer law, and it does not need one: the existing rules already apply. The Competition and Markets Authority published guidance on complying with consumer law when using AI agents in March 2026, alongside research on agentic AI and consumers, and updated both on 30 October 2026 when the Digital Regulation Cooperation Forum (the CMA, FCA, ICO and Ofcom) published a joint statement on how existing legal and regulatory frameworks apply to agents. The guidance is written for businesses that deploy agents, and its core points are blunt:

• "you are responsible for what an AI agent does in the same way you are responsible for what an employee does", even if someone else designed or provides it.
• Breaches can cost up to 10% of worldwide turnover, plus possible compensation to affected consumers, now that the CMA enforces consumer law directly.
• If the fact that someone is dealing with AI rather than a person might affect their decisions, tell them, and do not overstate what the AI can do.
• Prices must include all unavoidable charges, paid endorsements must be labelled, and offers and price reductions must be genuine.
• Keep "a human in the loop", actively checking that the agent is making correct decisions.

For a merchant, the practical risks of agentic commerce fall into five groups:

1. Errors you own. An agent that orders the wrong size or quotes the wrong price does not move the liability off you. Refunds and cancellation rights under the Consumer Rights Act 2015 and the Consumer Contracts Regulations 2013 apply as they do to any online sale.
2. Your own agent saying the wrong thing. An on-site shopping assistant that invents a delivery date or a returns policy is a misleading statement made by your business.
3. Fraud and authorisation. Accepting agent payments without scoped, tokenised credentials invites disputes over whether the customer ever approved the purchase.
4. Data loss. A third-party agent keeps the browsing behaviour and, often, the customer relationship. You may get the order and nothing else.
5. Platform dependence. The rules of each AI channel are set by its owner and can change in a quarter, as OpenAI's checkout reversal showed.

Reviews deserve a specific mention because agents rely on them. Fake reviews and concealed incentivised reviews have been banned practices since the Digital Markets, Competition and Consumers Act's consumer provisions came into force on 6 April 2025, and the CMA's fake reviews guidance sets out what a business that publishes reviews must do.

On-Site Agents Versus Third-Party Agents: The Strategic Divide

These are two different problems and conflating them wastes money. An on-site agent is a shopping assistant you deploy on your own site, and it is a conversion-rate play: it helps a visitor who has already found you to choose faster and abandon less. A third-party agent is external, owned by an AI platform, and it is a discovery and distribution play: it decides whether you are in the consideration set at all.

Most SMBs get this backwards. They buy an on-site chat widget because it is purchasable and visible, then wonder why it did not change anything strategic. The on-site agent is worth having if your catalogue is complex or your products need fitting, sizing, or compatibility guidance. It is close to worthless if your real problem is that external agents cannot read your inventory. Diagnose which problem you have before you spend, and remember that an on-site agent speaks for you in law: whatever it tells a customer about price, stock or returns is your statement, not the software vendor's.

The uncomfortable part of third-party agents is that they sit between you and the customer permanently. You lose the browsing data, the email capture, and the merchandising opportunity. That is a real cost, and it is the same trade marketplaces have always demanded. The difference is that you can refuse a marketplace. Refusing to be readable by agents is closer to refusing to be indexed.

The Data and Brand Loyalty Problem

Agentic commerce compresses the funnel to the point where brand-building has fewer places to happen. No category page browsing, no considered comparison on your site, no exposure to your photography or your copy. If the agent decides on price, delivery window, and review score, then price, delivery window, and review score are your brand.

There are three defensible responses. First, own a product attribute that is genuinely comparable and genuinely better, because an agent can only prefer what it can measure. Second, build a direct relationship that survives the agent: subscriptions, accounts, loyalty balances, and first-party data that make repurchase cheaper through you than through a fresh agent search. Third, be the brand the customer names, because a shopper who says "order me another pack of the ones from that Cornish roastery" has removed the comparison step entirely. Grocery shows the same pressure in miniature, where the weekly shop is already split between discovery and transaction modes, as we set out in our grocery customer journey guide. Named demand is the one thing agents cannot arbitrate away, which makes conventional brand marketing more valuable in an agentic world, not less.

What does not work is hoping the agent will notice your values, your story, or your design. It will not. Those still matter, but they have to be converted into a named preference before the agent is invoked, not after.

What UK SMBs Should Do First

In order, and none of this requires a rebuild:

1. Audit your product data. Pull your feed and check completeness on titles, GTINs, materials, dimensions and variant attributes. Fix the warnings you have been dismissing. Google Merchant Center now also accepts conversational attributes such as question_and_answer, related_product and popularity_rank, which help AI systems understand your products; Google describes them as optional, so add them after the basics are clean, not instead.
2. Get schema right. Product, Offer, AggregateRating and availability markup, validated and accurate. Agents and AI search surfaces read the same structured data, so this work pays twice. Treat it as a way of being understood, not as a promise of any particular search feature.
3. Make stock and price truthful in real time. If your sync is nightly, find out what breaks if it is not, and shorten it. Include delivery charges in the price an agent sees, because an unavoidable charge added later is a consumer-law problem as well as a lost sale.
4. Check your reviews are genuine and machine-readable. Marked up, recent and specific, collected by asking every real customer. A review platform that renders reviews only in JavaScript with no structured data is costing you.
5. Review your platform's AI channel settings. On Shopify, Agentic Storefronts is on by default for eligible stores, so look at what is being shared and where, even though UK checkout is not live yet.
6. Ask your payment provider what their agent-payment roadmap is. You are not implementing anything yet. You are finding out whether they have one.
7. If you deploy your own agent, follow the CMA guidance. Disclose it where it matters, keep a human checking its output, and log what it tells customers.
8. Measure AI referrals now. Traffic from assistants is already arriving, and our guide to tracking AI traffic in GA4 shows how to see it.
9. Keep investing in named demand. The brand work that makes someone ask for you by name is the hedge against everything above.

If you want help working out which of these your site actually fails, our digital marketing team starts with a free audit, then works to a fixed monthly fee with no lock-in. The fixes usually sit with ecommerce development rather than with advertising. The same structured-data foundations underpin AI search optimisation, generative engine optimisation and getting your brand mentioned in ChatGPT, which is why we treat them as one project rather than three. If your paid channels are also underperforming on data quality, the pattern behind Performance Max failures is the same root cause: automated systems are only as good as the feed you give them.

The UK government's Small Business Commissioner guidance on agentic commerce is a reasonable neutral starting point if you want a non-vendor view before committing budget.

Frequently Asked Questions