The businesses AI recommends (and how to become one of them)

We started with content. Then lead capture. Now: the infrastructure that puts your business in front of customers who never typed your name into a search bar.

Someone in Manchester opens ChatGPT and types "best CRM for a small recruitment agency." An answer comes back. It names three businesses. One of them gets a click, a visit, and a phone call that afternoon. The other two thousand CRM providers in the UK are not mentioned. They do not exist in that conversation.

Now replace the CRM with whatever you sell. An accountant in Bristol who handles contractor IR35. A bookkeeper in Leeds who specialises in restaurants. A web designer in Croydon who works with dental practices. The question gets asked. An answer comes back. Did it name your business?

This is already happening. Not in a pilot, not in a beta. Right now, today, real people are asking AI assistants for recommendations and acting on the answers. Greg Isenberg, who tracks distribution strategies for a living, noted that AI referrals for some businesses jumped from four percent to twenty percent in a single month. The channel is growing fast, and every day your business is absent from it, someone else is present.

Covering every question with pages you did not write by hand

There is a reason that one accountancy firm in Birmingham shows up when someone searches "VAT advice for online sellers in Birmingham" and yours does not. They have a page for it. You do not.

Not because they sat down and wrote five hundred pages by hand. Because they built a system that generates a useful, specific page for every variation that matters to their business. Bookkeeping for restaurants in Leeds. Corporation tax for contractors in Surrey. Payroll support for care homes in Manchester. Each one is a real page with genuinely tailored content, not a template with the city name swapped in.

This is programmatic SEO, and the reason it works now when it did not work three years ago is simple: AI can write content that is actually worth reading. The old version of this trick involved spinning the same paragraph with different place names dropped in, and Google got wise to it quickly. The current version involves generating pages where the content genuinely addresses the specific situation, because the cost of producing that kind of writing has dropped from hours of human time to minutes of machine time.

Why thousands of specific pages beat one good homepage

The arithmetic is straightforward, and it is worth sitting with for a moment.

Your homepage says "Accounting services across the UK." It competes with every other accountancy firm that says the same thing. Thousands of them. Your homepage is one voice in a very large room.

A page that says "VAT returns for Etsy sellers in Bristol" competes with however many other businesses bothered to create exactly that page. In most niches, the answer is: nobody. Zero. The specific page does not win by being better. It wins by being the only one that exists.

Google has always preferred pages that answer specific questions. AI assistants take that preference further. When ChatGPT or Claude is looking for something to recommend, it gravitates towards content that directly matches the question it was asked. A page that precisely answers "bookkeeping for restaurants in Leeds" is more useful to the AI than a generic services page, for the same reason it is more useful to the person who asked.

You are not tricking Google. You are not stuffing keywords. You are being the only business that bothered to answer the specific question someone actually typed.

That is the entire principle. It does not require cleverness. It requires coverage.

The part that sounds like science fiction (but is not)

The pages handle the search side. Now for the part most businesses have not heard of yet.

AI assistants are starting to use tools. Not just reading web pages and summarising them, but actively looking things up, checking availability, and taking actions. When someone asks an AI "find me an accountant in Bristol who handles IR35", the AI does not just guess. It goes looking. And increasingly, the way it looks is by connecting to structured sources of information that businesses provide directly.

An MCP server is one of those structured sources. Think of it as a menu for robots. It tells the AI assistant what your business does, where you are, what services you offer, what your availability looks like, and how to get in touch. It is written in the format that AI tools expect to read, the same way a website is written in the format that browsers expect to read.

Most businesses have not heard of MCP servers. That is exactly why early movers have an outsized advantage. Greg Isenberg compares building an MCP server in 2026 to building for mobile in 2010. The people who built mobile-friendly websites early did not just get a head start. They occupied the space before anyone else turned up, and some of them are still there.

Right now, the number of businesses with an MCP server is vanishingly small. The number of AI assistants looking for them is growing every month. That gap is the opportunity, and it is temporary.

What "being structured for AI" takes in practice

Neither of these is a weekend project. Both are worth being honest about.

Programmatic SEO, broken down plainly:

• A template: the layout and structure that each generated page follows. This gets designed once and refined over time.

• A data source: the combinations of service, location, and specialism that matter to your business. This might be a spreadsheet. It might be a database. It is the list of questions you want to answer.

• A generation system: the part that takes each combination, writes genuinely useful content for it, builds the page, and publishes it. This runs once and then tops itself up as you add new services or locations.

An MCP server, broken down plainly:

• Structured business information: your services, locations, pricing, availability, and contact details, organised in a way machines can read.

• A small server: a lightweight piece of software that speaks the protocol AI assistants expect. It sits alongside your website and answers questions about your business when AI tools come asking.

The programmatic SEO piece is a few weeks of focused work to build and then runs more or less on its own. The MCP server is a smaller build but requires someone who understands the protocol and how AI assistants discover and use these tools. Both produce results that compound over time. A page you publish this month still answers the same question next year. An MCP server you stand up this quarter is still feeding information to AI assistants next quarter.

Neither of these is something you would typically hand to your existing web agency, because most web agencies have not built them before. This is infrastructure work, not design work.

The window is open now. It will not stay open.

The businesses that set up programmatic SEO six months ago are already pulling in traffic from questions nobody else is answering. The businesses that stand up MCP servers this year will be the ones AI recommends in 2027. Both of these advantages exist for one reason: most businesses have not heard of either concept.

That will change. It always does. The businesses that built mobile sites in 2010 had a clear run for three or four years before everyone else caught up. The businesses that took social media seriously in 2012 had a similar window. Each time, the early movers did not just get a temporary bump. They established a presence that late arrivals had to fight to match.

Every month that passes, the window narrows. More agencies will learn about programmatic SEO. More developers will start building MCP servers. The gap between "first mover" and "one of many" closes by degrees, and once it closes it does not reopen.

The cost of building this infrastructure is measured in weeks. The cost of not building it is measured in every customer who asked an AI for a recommendation and got someone else's name.

This is the kind of infrastructure that takes a few weeks to build and runs for years. If you want to be the business AI recommends in your niche, talk to us about what that looks like for you.