The COO of a 120-person logistics business in Milton Keynes opens her AI assistant on Monday morning. She types: find three executive search firms that have placed finance directors in logistics, check their fees and track record, and book intro calls with the best two for next week.
By lunchtime there are two calls in her diary and a one-page summary in her inbox. She has not visited a single website. Her agent visited all of them.
This is already happening in the UK. In a CI&T survey of more than 2,000 UK and Irish shoppers, 68% have already used an AI agent when shopping. On the business side, Magenta Associates found 66% of UK senior decision-makers with B2B buying power use AI tools to find and assess suppliers, and 90% of them trust the recommendations. The Competition and Markets Authority thought it important enough to publish its own report on agentic AI and consumers in March.
Most firms I speak to are asking whether AI can find them. That question is already a step behind. The harder one is whether an agent can finish the job on your website: check you, compare you, contact you and book you, without a person clicking anything.
Your website was designed for someone who browses. Its next visitor reads everything, skims nothing and leaves the moment it can’t do what it was sent to do.
From finding you to booking you
An agent working for a buyer runs the whole early part of the sale. For a professional services purchase, that journey has six stages:
- Brief. The buyer sets the job: sector, size, budget, timing, what good looks like.
- Find. The agent builds a long list from AI answers, search, directories and the sources it trusts.
- Vet. It reads your site and what others say about you, checking each claim against the brief.
- Compare. It lines you up against the alternatives on the things the buyer cares about, including cost.
- Act. It sends the enquiry, fills in the form or books the call.
- Report back. It tells the buyer who made the cut, why, and what happens next.
Parts of this are already routine on the consumer side, where UK shoppers mostly use agents to compare brands, find the lowest price and find where to buy. Business buying is following, and the browser is being rebuilt for it. Google’s proposed WebMCP standard lets a website tell agents exactly how to use its forms and features, and it is now in an origin trial in Chrome.
Most of the web is not ready for stage five. Plenty of firms would survive the first four and lose the client at the contact form.
Designing for an agent that does the whole job
Each stage asks something different of your website. Design for all five that happen after the brief, not just the first.
1. Match the brief
The agent is matching your words against a specific brief. “We help ambitious growth businesses” matches nothing. “We place finance directors in logistics and distribution businesses with 50 to 500 staff” matches exactly.
Say plainly who you work for: sector, size, location, the problems you solve and the ones you don’t. Put it on the pages that matter, in text, not just in a logo strip.
2. Prove it
An agent can’t be charmed, but it can be convinced by evidence it can check. Write case studies as facts: the type of client, the problem, what you did, how long it took and what changed.
Name the people who would do the work and what they have done before. Make sure the same story appears elsewhere, on LinkedIn, in directories and on client sites, because the agent will look there too.
3. Price it
This is where most professional services sites fail. Agents are often given a budget, and a firm they can’t place against it is easy to drop.
You don’t need a rate card. An indicative range, a typical engagement shape or a “projects usually start from” figure gives the agent something to work with. “Every engagement is bespoke” gives it nothing. In the Magenta research, 71% of UK B2B buyers said they would avoid suppliers that lack clear, transparent information. An agent applies that rule every time.
4. Let it act
If the agent can’t complete the next step, the buyer never hears about you. Every route to contact should work for software acting on someone’s behalf:
- A plain enquiry form with clearly labelled fields, on the page rather than in a pop-up.
- The questions you actually need to qualify a lead, so the agent can answer them from its brief.
- Real-time booking for a first call, rather than an email chain to find a slot.
- A clear confirmation of what happens next, sent to a human.
WebMCP is worth watching here. Its declarative version works by adding annotations to standard HTML forms, so a well-built form today is most of the way there. It is still a proposal and subject to change, so treat it as a later step, not a reason to wait.
5. Hand back to the human
The agent’s last job is a summary for the buyer. Give it something worth quoting: what you would do, how long it would take, roughly what it costs and who would lead it.
Then make the human’s check easy. When the buyer clicks through from that summary, the page they land on should confirm everything the agent said in seconds.
What stops an agent on most firm websites
Most of what blocks an agent was put there to impress or capture a person. Common examples:
- Facts locked in pictures. Sectors, results and team credentials shown in images, sliders or video, with no text behind them.
- Gated proof. The best case studies hidden behind a download form the agent won’t fill in.
- “Contact us for pricing.” A dead end for any agent with a budget to work within.
- Contact routes that need a human. Pop-up forms, chat widgets as the only option, or “email us to arrange a time”.
- Category language. Invented labels for what you do, rather than the words a buyer would use in a brief.
- Inconsistent stories. Different services, sectors or team names on your site, LinkedIn and directory listings, which makes the agent less sure of all of them.
None of these needs a rebuild to fix. Most need a decision about what you are willing to say in public.
People still hire people
For a serious professional services engagement, a person still makes the final call. The government’s AI adoption survey, published last week, found UK businesses trust AI for routine tasks like drafting and summarising, but much less for high-stakes decisions or autonomous processes. The agent decides who gets the meeting. The meeting decides who gets the work.
That changes what the website is for. It no longer has to carry the buyer from first visit to enquiry on its own. It has to give the agent enough facts to put you forward, then give the human enough reasons to say yes when they check.
This is one site, not two. The facts and the story sit on the same pages: plain statements an agent can extract, and the judgement, people and proof a buyer wants to see for themselves.
It also raises the stakes on the first call. If an agent has booked it, the buyer arrives with a summary of you already in hand. The call has to go beyond what they have read, not repeat it.
A ten-minute test: send an agent to your own site
The quickest way to see your site as an agent does is to brief one. Use an assistant that can browse and act, then work through these steps:
- Write the brief your best client would give. Include their sector, size, budget and timing, in their words.
- Ask the agent to find and shortlist three firms like yours. Did you make the list? If not, who did, and why?
- Ask it about your firm specifically. Can it say who you work for, what you do, what it costs and who would lead the work?
- Ask it to book a first call or send an enquiry. Watch where it gets stuck.
- Read the summary it gives you. Is that the story you would want a buyer to hear?
Wherever the agent guessed, hedged or gave up, a buyer’s agent would too. That list is your design brief.
The website is becoming a place where work gets done
For twenty years, firm websites have been brochures that people read. They are becoming places where software does a buyer’s work for them: checking, comparing, booking.
The firms that win the next few years of new business will not necessarily have the best-looking sites. They will have the clearest ones: the plainest statement of who they help, the most checkable proof, an honest indication of cost and a next step that anything, human or not, can complete.
If you run the test above and want a second view on what it shows, a 30-minute diagnostic call is a good place to start.
Frequently asked questions
The buyer gives an AI assistant a brief, and the agent does the early work of the purchase itself: it finds suppliers, reads their websites and what others say about them, compares them on cost and fit, and sends an enquiry or books a call. The buyer only sees the result, usually a shortlist with a short summary of each option.
State plainly who you work for, what you do and how long it takes. Write case studies as checkable facts, give an indication of cost, and make the next step simple: a clearly labelled form on the page and real-time booking for a first call. Make sure the same story appears on LinkedIn, directories and other sites, because the agent checks those too.
Not necessarily a rate card, but an agent working to a budget needs something to compare against. An indicative range, a typical engagement shape or a “projects usually start from” figure is often enough. In UK research by Magenta Associates, 71% of B2B buyers said they would avoid suppliers that lack clear, transparent information.
WebMCP is a proposed web standard from Google that lets a website tell AI agents exactly how to use its forms and features, so an agent can complete a task such as an enquiry or booking reliably. It is in an origin trial in Chrome and still subject to change. A clear, well-labelled form today is most of the way there.
Not for serious professional services work. The UK government’s 2026 AI adoption survey found businesses trust AI for routine tasks but much less for high-stakes or autonomous decisions. The agent increasingly decides who gets the meeting, but a person still decides who gets the work, so a website has to serve both.
