AI Agents Are About to Start Booking Your Customers. Is Your Website Ready?
Until recently, AI assistants only suggested. In 2026 they have started to act. ChatGPT, Perplexity, Gemini and Claude can now book appointments, place orders and confirm transactions on a customer's behalf. The implication for UK small businesses is uncomfortable: soon your customers will not visit your website. Their AI agent will. And if the agent cannot read your site, the agent quietly moves on to a competitor that it can.
The four numbers that explain what just happened
The shift from AI-as-recommender to AI-as-buyer is documented. Four sources matter most.
+4,700% AI agent traffic to retail sites (Adobe, 2025). Adobe's 2025 retail report showed AI agent generated traffic to e-commerce sites grew 4,700% year-over-year. The starting base was small but the trajectory is exponential. By any plausible projection, 2026 to 2027 sees agent traffic become a meaningful percentage of all visits.
£3 to 5 trillion orchestrated by agents by 2030 (McKinsey QuantumBlack, October 2025). McKinsey's October 2025 QuantumBlack report forecasts AI agents will orchestrate between £3 and £5 trillion of retail spend by 2030. Even the low end is a meaningful single-digit percentage of total global retail. For a UK SME, even a fraction of a percent of that requires being agent-readable.
3x conversion gap, closing (Walmart, 2025). Walmart's 2025 in-chat shopping research showed agent-mediated purchases currently convert at roughly three times lower than on-site purchases. That looks like a deterrent until you flip it: the gap is closing rapidly. Today's 3x gap will not be 3x in 2027. Businesses ready when the gap closes capture disproportionate share.
71% of ChatGPT-cited pages use structured data (commercetools, 2026). commercetools 2026 research found 71% of pages cited by ChatGPT used schema.org structured data, versus a much lower baseline for uncited pages. The signal is one of the strongest predictors of whether an AI engine will cite or transact with a business.
Put those four together and the picture is clear. Agent traffic is exponential. The future spend pool is enormous. The conversion economics are still working themselves out. And the technical readiness that earns AI engagement is concentrated in a small minority of websites.
What an agent transaction actually looks like
Picture a customer asking ChatGPT (or any agent-capable AI) to "book me a plumber in Birmingham who can come this evening". This is what happens next, today, in 2026:
Step 1: The customer asks the question. They speak or type to the AI. They do not open Google. They do not open a directory site. They are asking the AI to handle this end to end.
Step 2: The agent reads candidate websites. The agent searches for plumbers in Birmingham, then visits a handful of candidate websites and reads them. The agent looks for clear answers (open this evening?), structured data (service area, prices, contact details in schema.org), and a way to act (booking endpoint, contact API, machine-readable availability).
Step 3: The agent runs 14 checks on each candidate. Implicitly. Can it find your page? Can it parse your structured data? Can it confirm you serve Birmingham? Can it find your evening availability? Can it identify a way to act? If too many checks fail, the agent moves on to the next candidate.
Step 4: The agent either books or escalates. If at least one candidate passes the checks, the agent either makes the booking (if the candidate exposes a booking endpoint or accepts agent-friendly form submissions) or hands back to the customer with "I recommend you contact [business] on [number] about [details]". Either way, the candidate that passed the checks gets the customer.
The losers in this flow are not the businesses with worse reviews or higher prices. They are the businesses the agent could not read.
The four AGENTREX categories, in plain English
AGENTREX is the free agent-readiness scanner at agentrex.aeo-rex.com. It runs 14 checks across four categories. Here is what each category measures and why it matters.
Discoverability
Can an agent find and crawl your business? This category checks llms.txt (the emerging standard for telling AI engines what your site is and how to read it), robots.txt (the established standard, but it needs the right entries for AI bots), URL hygiene (clean, semantic URLs), and homepage clarity (does your homepage answer the basic "what is this business?" question for a machine?).
An SME that fails this category is invisible to agents before any other check matters. The agent never gets past the front door.
Structured Data
Can the agent parse your business facts without guessing? This category checks for schema.org markup on the homepage, on service pages and on contact details. It looks for an Organization schema, LocalBusiness schema (with proper address, phone, opening hours), Service schema (with what you do, where you do it and what it costs), and a clean entity graph (one consistent identity across the page).
SMEs that fail this category often pass Discoverability but get skipped at decision time because the agent cannot extract the facts it needs to recommend the business confidently.
Agent Access
Are AI bots actually allowed to read your site? Many UK SME websites accidentally block AI agents at the CDN, firewall or server level. Cloudflare, AWS WAF and even WordPress security plugins often have default rules that block GPTBot, ClaudeBot, PerplexityBot and Google-Extended. The site loads fine for humans but returns 403 Forbidden to the agent. The owner has no idea this is happening because they never see the request fail.
This category checks each major AI bot against the live site and confirms it can fetch the homepage and key pages.
Transaction Readiness
If the agent wanted to buy or book, could it? This category checks for machine-readable pricing (schema.org Offer with price + priceCurrency), service availability (Service schema with serviceType and areaServed), contact endpoints (ContactPoint schema with telephone + email + contactType) and (where appropriate) .well-known/ai-plugin.json or equivalent agent-action manifests.
This is the newest and weakest category for most SMEs. It is also the highest-leverage one because as agent transactions grow, businesses that pass this category capture share from businesses that do not.
What an SME should do this quarter
Three steps. Each is feasible without a developer budget.
Step 1: Run the AGENTREX scan (free, 60 seconds). Go to agentrex.aeo-rex.com, enter your homepage URL, get a score across the 14 checks. The result tells you which category is weakest and which specific checks are failing. No card, no signup.
Step 2: Fix the cheapest wins first. llms.txt is a 30-minute job. Adding the right entries to robots.txt is another 30 minutes. Unblocking GPTBot, ClaudeBot and PerplexityBot at your CDN takes about an hour if your hosting provider has a sensible dashboard. These three together close most Discoverability and Agent Access gaps.
Step 3: Mark up structured data on the homepage and the top three pages. Organization, LocalBusiness, Service, ContactPoint. There are free schema generators that produce the JSON-LD. Pasting it into the page is straightforward. This single step typically lifts a low AGENTREX score by 20 to 30 points.
If after that work the AGENTREX score is still under 60, the gap usually needs the Transaction Readiness category addressed, which is where AEO-REX's done-for-you implementation closes the loop. But the first three steps are well within reach of any SME owner with a Saturday morning.
What this means for your business
The agent economy is not a 2030 problem you can plan for later. AI-generated traffic to retail sites already grew 4,700% in 2025. The fundamentals (structured data, agent-readable contact endpoints, machine-readable pricing) take effect the moment they are in place. UK SMEs that prepare now will be the ones agents transact with in 2027.
The first step is knowing whether your site even passes the basics. The free AGENTREX scan tells you in 60 seconds. From there the work is well-defined and the wins are concrete.
Frequently asked questions
What is an AI agent in the context of retail and booking?
An AI agent is a programme that acts on a person's behalf. The customer says "book me a plumber in Birmingham who can come this evening", and the agent reads websites, compares options, makes the booking and confirms it back. ChatGPT, Perplexity, Gemini and Claude all now offer agent features that go beyond pure recommendation into actual transaction.
How much retail spend will AI agents control by 2030?
McKinsey QuantumBlack research published in October 2025 forecasts AI agents will orchestrate between £3 and £5 trillion of retail spend by 2030. That is a midpoint of around £4 trillion, equivalent to a meaningful single-digit percentage of total global retail. For a UK SME, even capturing a tiny share of that requires being readable to agents.
What does "agent-ready" actually mean for a website?
Agent-ready means a website passes four categories of check. One, Discoverability: agents can find and crawl your business via llms.txt, robots.txt and clean URLs. Two, Structured Data: your pages carry schema.org markup that machines can read without guessing. Three, Agent Access: AI bots are not blocked at the firewall, CDN or robots level. Four, Transaction Readiness: machine-readable pricing and capability files so an agent can move from question to booking. AGENTREX runs 14 individual checks across these four categories.
Why do 71% of ChatGPT-cited pages use structured data?
Because structured data tells ChatGPT explicitly what a page is about. When a page carries schema.org markup describing a service, its price, its location, and its provider, ChatGPT can parse those facts without inferring them. commercetools research in 2026 found 71% of ChatGPT-cited pages used structured data, versus a much lower baseline for uncited pages. The signal is one of the strongest predictors of whether an AI engine will cite or transact with a business.
How does in-chat agent shopping compare to traditional on-site shopping?
Walmart published research in 2025 showing in-chat agent shopping converts at roughly three times lower than traditional on-site shopping. That sounds discouraging until you read it the other way: the conversion gap is closing rapidly as agents get more capable and as websites become more agent-ready. Today's 3x gap will not be 3x in 2027. Businesses that are agent-ready as the gap closes will capture disproportionate share.
How do I find out if my website is agent-ready?
Run the free AGENTREX scan at agentrex.aeo-rex.com. It runs 14 checks across the four agent-readiness categories and returns a score in about 60 seconds. No card, no signup. The scan tells you exactly which of the 14 checks pass and which fail, so you know which gaps to close first.
Run a free AGENTREX scan
The 14 checks take 60 seconds. The free AGENTREX scan tells you exactly which agent-readiness categories your site passes and which it does not, with no card, no signup and no commitment.