Your Next Website Visitor Might Not Be a Person

For twenty-odd years, websites have been built for two audiences: people, and the search engine crawlers that send people. There's now a third. AI assistants read your pages to answer questions, compare suppliers and, increasingly, carry out tasks on someone's behalf. Some of the most important visits your site gets this year will come from software that never sees your design.

How the buying journey is changing

Think about how someone researches a pallet truck, a law firm or a hosting provider today. Plenty still start with a search engine. But a growing number ask an assistant instead: summarise the options, compare these three suppliers, tell me which one supports what I need. The assistant reads the sites and presents its answer. The buyer might visit the winner's site afterwards, or might not visit any of them at all.

Agentic browsing takes this a step further. Assistants can now open pages, fill in forms and work through multi-step flows. That means your contact form, product configurator or quote request might be completed by an agent acting for a real customer.

None of this replaces human visitors. But it adds a layer between you and your customer, and that layer rewards a very particular kind of website.

What AI readers struggle with

Assistants are good at reading text and remarkably bad at guessing. Information that only exists in an image, a PDF behind a form, a tab that loads on click or a clever animation is often information the assistant simply doesn't get. If your key specifications live in a downloadable datasheet, an assistant may well tell a buyer that you don't publish them.

Vagueness hurts too. If your pricing page says "contact us" and a competitor's says "from £49 a month", guess which one makes it into the comparison. If three pages describe your service in three different ways, the assistant has to pick one, and it might not be the one you'd choose.

Then there's speed and reliability. Automated visitors are less patient than people. Pages that depend on heavy client-side rendering, cookie walls that hide the content, and forms that only work with a perfectly placed click are all places where an agent gives up and moves on to the next supplier.

The good news: it's mostly just good engineering

There's a lot of noise right now about optimising for AI, and a fair amount of it is snake oil. Most of what makes a site work well for AI assistants is exactly what made it work well for search engines, screen readers and people on slow connections: fast, server-rendered pages, proper headings, accessible forms, structured data and content that says what it means.

If you've been putting off accessibility work, this is one more reason to get on with it. The same semantic structure that helps a screen reader navigate a page helps an agent understand it. Fix one and you tend to fix the other.

There are a couple of newer pieces worth watching. Conventions like an llms.txt file give assistants a plain-text map of your site. Protocols such as the Model Context Protocol go further, letting you expose things like a product catalogue or stock availability directly to AI tools, in a structured way that you control. For businesses with large catalogues or dealer networks, that second one is particularly interesting: rather than hoping an assistant scrapes the right number from the right page, you hand it the right number.

Where to start

You don't need a rebuild to get ready for this. A few practical checks will tell you a lot about how your site looks to an AI reader:

• Load it without JavaScript. Turn scripts off and open your key pages. Whatever disappears is probably invisible to some agents too.
• Move facts out of files. If specs, prices or service areas only exist in PDFs or images, put them on the page as text.
• Add structured data. Schema markup for your organisation, products, services and FAQs gives assistants unambiguous facts to quote.
• Test your forms the hard way. Try completing them using only the keyboard. If that's painful, an agent will struggle as well.
• Ask the assistants. Ask a few popular AI tools about your business and your competitors. Note what they get wrong and trace it back to your content.

Built once, read by everyone

It's tempting to treat this as yet another channel to optimise for. We'd frame it differently. A website that's fast, well structured, clear about its facts and accessible to everyone works for people, search engines and AI assistants alike. That's the kind of site worth building anyway.

The businesses that do well here won't be the ones chasing every new AI trick. They'll be the ones whose websites were built properly in the first place.