AI agents are about to start shopping for UK small businesses. Most owners haven't noticed.

Why is this shift different from the last two?

The web has had two big shifts in how customers find businesses. Desktop search arrived in the late 1990s. Mobile took over in the early 2010s. Each rewrote the league table of who got chosen and who got skipped. The third shift, AI agents acting on the customer's behalf, is the first one where the customer never sees the search results. The agent reads. The agent picks. The customer accepts.

For a UK pub owner, a salon, a takeaway, a retailer, this matters in one specific way. In the previous two shifts, your job was to be findable. Your photos, your menu, your reviews, your location, all aimed at a human reading a screen. The new job is to be machine-readable. The agent visits your site, scans the structured data, pulls the information it needs, then reports back to its human in plain English.

If your site doesn't carry that structured data, the agent has to guess. Most of the time, it will skip you and recommend the competitor whose data is clear.

Three concrete signals the infrastructure is already shipping. ChatGPT shipped Instant Checkout in September 2025, allowing the assistant to complete a purchase end to end via the Agentic Commerce Protocol it co-developed with Stripe; the initial launch covered US Etsy sellers, with broader merchant onboarding planned by OpenAI thereafter. Perplexity has a free agentic shopping product integrated with PayPal, with merchant coverage growing through PayPal's Instant Buy programme. Google announced its Universal Commerce Protocol at the National Retail Federation conference in January 2026, co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, and endorsed by Adyen, American Express, Mastercard, Stripe, and Visa.

The bridge to Reeve Consult is already visible here. Agentic shopping collapses three things into one machine-readable path: discovery, booking, and payment. We work on the payment leg every day, and the same forces shaping how UK businesses get paid are now shaping how they get found.

How does an AI agent actually shop?

A useful way to picture it. An agent is given a job by its human. "Find me an Italian for two tonight near Hockley, no chains, gluten-free options, well reviewed, can take a card on arrival." The agent then visits dozens of restaurant websites, reads each one, cross-references against Google Reviews and TripAdvisor and the local press, and comes back with three options. The human picks one. The agent books it. The full path, in our test runs, takes seconds, not minutes [anecdote].

Notice what happens in those seconds. The website didn't need to convince a human. It needed to convince a machine that the venue exists, takes bookings, has gluten-free options, takes cards, and is genuinely well reviewed. Beautiful design did nothing. Clever copy did nothing. The only thing that mattered was clean, structured, retrievable information.

That extraction question, can I pull what I need cleanly from this site, is the only question the agent asks before it skips you or shortlists you.

For payments specifically, this already applies at the checkout end of the journey. An agent that finds a salon for its human customer needs to know whether the booking system supports a deposit on a card, how the no-show fee works, whether the venue takes Apple Pay or Google Pay. If those answers are not in your booking system's API or on the page in machine-readable form, the agent picks the next salon on the list. Payment legibility is search visibility now.

What signals do agents look for, in priority order?

This is the part most articles get wrong. They treat AI optimisation as a new science. It isn't. The signals an agent reads are mostly the signals Google's search index has read for years, with two differences. First, the bar is now hard, not nice-to-have. Second, machine readability matters more than human readability.

The five signals, in priority order:

1. Structured data markup. Schema.org JSON-LD on every important page: your menu, your products, your services, your opening hours, your reviews. This is the language the agent speaks. Without it, you exist as a paragraph of prose the agent has to guess at.
2. Content clarity. Short paragraphs. Direct answers. Tables where comparison helps. Headings that match how a customer asks a question. The same things that make a page easy to read for a human are the things that make it extractable for an agent.
3. API compatibility. Can an agent check your inventory, see your pricing, query your availability, complete a booking? If yes, you're a candidate for "buy from this one". If no, the agent has to scrape your homepage and may simply give up.
4. Cross-web brand presence. Reviews on Google, TripAdvisor, Yell, the trade press, local newspapers. Brand mentions on partner sites. Citations from authoritative sources. Agents triangulate. A site nobody else mentions doesn't get trusted.
5. Freshness. Visible last-updated dates, recent reviews, content that doesn't look like it was written in 2019. AI agents heavily weight current data. Stale pages signal an unreliable source.

Notice that none of these are exotic. Numbers one and two are 1990s-era SEO advice that most independent SMEs never finished implementing. Number three is the kind of thing an agency might quote four-figure project fees for, when in practice your booking system probably already exposes a usable API and nobody has switched it on. Number four is reputation work you should already be doing. Number five is a calendar reminder.

What concretely should a UK independent business change?

This is the playbook. Most of it is a week of focused work, not a six-month project.

Step one: add schema markup to every important page. The single highest-impact action. A simple Schema.org JSON-LD block on each product, service, FAQ, and review page tells every AI agent on the internet exactly what your business sells, what it costs, and who it serves. Implementation is an afternoon for a developer who knows the territory. Without it, you are a paragraph of prose; with it, you are a structured record.

Step two: rewrite key pages for clarity. Every page should answer four questions in plain English. What do you sell? Who is it for? What does it cost? How does it work? Headings, short paragraphs, tables where comparison helps. You are writing for a very smart, very literal reader with zero patience for marketing prose. The same edits make the page better for human readers too.

Step three: open your data to agent interactions. If you sell products, ensure your inventory and pricing are accessible through a clean data feed or API. If you take bookings (restaurants, salons, clinics, gyms), ensure availability is queryable through your booking platform's API. Several major UK booking platforms expose this directly to partner integrations: Phorest documents its booking and client APIs publicly. Equivalent integration paths are listed in the developer portals of Treatwell, Fresha, Zonal, Lightspeed, OpenTable, ResDiary, and SevenRooms; check whether your specific account has the right setting switched on, since defaults vary.

This step is where Reeve Consult's payments-consultancy lens is directly useful. The same APIs that let an agent see your availability also surface your card-payment options, your deposit policy, your refund rules, and your card terminal capability. We help clients map all of these into the customer journey already; agentic shopping makes that mapping visible to a non-human reader.

Step four: build brand signal across the open web. Get listed on relevant industry directories. Earn reviews on the platforms your customers and the agents both check. Local press, trade publications, industry partners. The agents triangulate; the more vouching the open web does, the higher you rank in their shortlist.

Step five: keep everything current. Add last-updated dates to every important page. Refresh core content quarterly. Update reviews and case studies as they come in. Stale content is a trust killer for agents.

There is a sixth move that compounds. Create content that teaches the agent which category you own. A pub doesn't say "we serve food". It says "we are the only Sunday roast in Hockley with three rotating ales and a vegan option, open from noon until four". The specificity teaches the agent. The more specific and consistent your category positioning, the more strongly an agent associates your business with that category, and the more often you become the default recommendation for that category.

This is the same compounding loop that built every successful SEO position of the last twenty years. The difference is the rules apply to a non-human reader.

How long is the window to act?

Every internet shift has had a window. A period where the early movers locked in a position competitors never caught up to. The businesses that learned SEO in 2003 still rank on page one of Google for their core terms. The brands that committed to mobile-first in 2010 captured customer behaviour they hold today. The ones that waited didn't catch up.

The AI agent window is shorter, for two reasons.

First, AI adoption is moving faster than any prior consumer technology. Gartner's November 2025 forecast projects 90 percent of B2B buying activity through AI agent intermediation by 2028, channelling over $15 trillion in spend. On the consumer side, Morgan Stanley's outlook puts 10 to 20 percent of US e-commerce spending through agent-mediated channels by 2030, with 23 percent of Americans already reporting at least one AI-assisted purchase in the past month. The mobile transition took roughly five years to reach mainstream behaviour. Industry observers expect this transition to compress that into roughly twelve to eighteen months [opinion].

Second, agent recommendations compound. Once an agent learns to recommend a particular pub, salon, or retailer in a category, that recommendation gets reinforced every time the agent's human accepts it. More recommendations lead to more bookings, which lead to more reviews, which lead to more authoritative cross-references, which lead to even stronger recommendations. The flywheel runs in favour of whoever gets on it first.

The businesses that wait are not just losing twelve months of bookings. They are setting themselves up to fight an uphill battle against competitors who are already the agent's default answer.

What does this actually cost a UK independent SME to set up?

Direct answer: less than most owners assume, if it's done in the right order.

Schema markup is an afternoon of developer time, often free if your booking platform or content management system supports it natively (most modern ones do). Page rewrites are an editorial weekend. Booking and inventory API exposure is usually a setting in your existing platform, not a new development project. Brand signal building is the work you should be doing anyway. The freshness discipline is a quarterly calendar reminder.

A reasonable budget for a UK independent business to do all of this properly, with someone who actually knows the territory, is in the low four figures [estimate based on Reeve Consult client engagements]. In our client work, the bookings recovered or won in a single quarter typically cover that several times over [anecdote]. The cost of doing nothing is the cost of becoming invisible to a buying agent eighteen months from now.

The decision is less about whether this matters and more about when you address it, before competitors make their sites easier for agents to read.