AI Agents and Your London Business Website: Are You Ready to Be Booked by a Bot?

Everything else in this series has been about being found. This post is about what happens next: the moment a machine, rather than a person, tries to use your website.

It sounds speculative, and some of it is. But Google has put it in its official guidance, and a good deal of the preparation is the kind of work that helps human customers anyway. So I want to separate what Google has actually said from the hype, and then give a London local business a short, sensible list.

What does Google say about AI agents?

Google describes AI agents as autonomous systems that can perform tasks on behalf of people, such as booking a reservation or comparing product specifications. Its guide says they take many forms and that browser agents may access your website to gather what they need, by analysing visual renderings such as screenshots, inspecting the DOM structure and interpreting the accessibility tree.

It frames this carefully. The advice is that if this is relevant to your business and you have extra time, look at the available agentic experiences and review its guide to agent-friendly website best practices. It also mentions emerging protocols such as the Universal Commerce Protocol that will let Search agents do more. This is the last item in the guide's list of next steps, not the first.

Where do Google's own search agents stand?

At its May 2026 announcement Google introduced search agents and said those capabilities would roll out to everyone in the United States that summer. The announcement did not give a UK date, and I am not going to guess one. For a London business the honest position is that this is arriving, in stages and in some markets first.

The same announcement said AI Mode had passed one billion monthly users, and that Gemini 3.5 Flash had become its default model. The direction of travel is that more of the comparing and shortlisting, the research stage of buying, happens inside AI systems that can eventually act as well as answer. Whether an agent books your clinic in 2027 or 2030 is unknowable. What you can do is make sure it would not fail on your site.

What does an AI agent actually need from a website?

An agent needs to understand what the page offers, find the action it is supposed to take, and complete it without ambiguity. The three ways Google says agents read a page, visually, structurally and through the accessibility tree, tell you exactly where to look.

• Visual rendering. What a screenshot shows. A booking button hidden under a cookie banner, a carousel that covers the phone number or a pop-up that can never be closed fails here as it fails a human.
• DOM structure. Real buttons, real links, real form fields. A clickable div with a script attached is harder to interpret than a button that says what it does.
• The accessibility tree. The layer screen readers use: labels, roles, names. If a form field has no label, neither a blind customer nor an agent knows what to type.

Google says perfect semantic HTML is not required for Search and that Google can understand imperfect code. It also says semantic HTML is a good idea where possible, because it helps other types of users such as screen readers and, by implication, agents. That is the right framing: it is optional for ranking, helpful for use.

How should a London local business prepare?

Keep the facts current, make the booking path simple and accessible, and keep the site fast and uncluttered. None of that is exotic. It is the practical version of Google's own advice on page experience, and it would improve conversion today whether or not an agent ever turns up.

Where does the Business Profile fit in?

Google says AI responses and other Search results can show local business information, and that Business Profiles help your business be visible in them. The profile is a structured, verified source an agent can read without scraping anything, which makes accuracy there more important than most owners realise.

A wrong opening time on the profile is a problem for a human who turns up to a closed door. It is a bigger problem for an agent that books a table based on it. I treat profile accuracy as the first agent-readiness task for any local client, because it costs nothing and it is the place errors do the most damage. The detail of how to keep it clean is in my post on Google's name and address rules.

What about automation on my side?

The other half of agent readiness is how fast your business responds when one makes contact. An enquiry that arrives instantly from a machine is the same as one from a person: it goes to whoever answers first. If your reply loop takes a day, an agent comparing three providers will simply move on.

• Missed-call text-back, so a lead that rings out gets an instant message rather than silence.
• Instant confirmation of every enquiry form with a clear next step.
• Lead routing to a named person with a response time target.
• Source tracking, so you know whether a booking came from the Map Pack, organic search or an AI answer.

That is the work behind my AI automation service, and it is useful today regardless of agents. Speed to lead is one of the few variables in local business that you control completely and that reliably moves revenue.

Does this change how I measure enquiries?

It makes attribution more important, and it should not make you over-read any single number. If a booking arrives from an agent acting for a customer, your analytics may record it oddly, as direct traffic or as an unfamiliar visitor, and there is no standard way yet to label it. I would not build a dashboard around something that cannot be measured reliably.

• Ask every new customer how they found you, on the form and on the phone. It is crude, and it is still the most reliable attribution a local business has.
• Keep the Search Console Generative AI performance report as your sanctioned view of AI feature visibility.
• Watch for unusual patterns in bookings, such as enquiries with no prior visit, without drawing conclusions from a handful.
• Track the response time to every lead. It is the one variable you control whatever the source.

What should I not do?

• Do not buy 'agent optimisation'. There is no agent ranking factor to buy. Google lists emerging standards as something to monitor, not a deliverable.
• Do not add hidden text or markup aimed at bots. That is cloaking territory and Google's own advice is that no special markup is needed.
• Do not block crawlers or agents out of fear. Check that robots.txt and your CDN are not blocking the access Google says is needed to be eligible for AI features at all.
• Do not neglect people. Everything an agent does, a human also does faster when the site is clear.

What is the ten-minute version?

1. Open your site on a phone and try to book from the first screen. Count the taps and the obstacles.
2. Check your hours on the site and the profile. Fix any difference.
3. Look at your forms. Every field should have a visible label and the button should say what it does.
4. Make sure prices or ranges appear as text where you can honestly give them.
5. Test what happens when a lead rings out or an enquiry arrives at midnight.

If all five pass, you are as ready as Google's guidance asks anybody to be, and your customers will already have noticed. If three fail, you have found your next month's work, and it would have been worth doing anyway.