AI in your website

Most conversations about “AI and your website” mash together two things that need separating.

Can AI find and quote you correctly? When someone asks ChatGPT, Gemini or Perplexity for a supplier like you, does your name come up? Is what’s said about you accurate, current, and yours — or is it three years out of date and lifted from a directory listing you forgot existed?

Can an AI agent actually use your site? Increasingly, software is doing the visiting. Browsing agents book, buy, compare and fill in forms on behalf of real people. If your booking form is a wall of unlabelled inputs, or your pricing lives in an image, the agent gives up and recommends someone else.

The first is a content and reputation problem. The second is a technical one. We handle both.

In May 2026, Google shipped an Agentic Browsing category in Lighthouse 13.3 — the same tool that already scores your performance, accessibility and SEO. Agent-readiness went from a talking point to a number you can be measured against, by anyone, in about thirty seconds.

Around the same time, Google and Shopify launched the Universal Commerce Protocol, and WebMCP — a way for websites to hand AI agents a proper list of what they can do, instead of making them guess — entered public trial in Chrome. The plumbing for an agent-readable web is being laid right now.

Most sites are not ready. That’s the opportunity.

Where we’re honest with you

There’s a lot of noise in this space, and a lot of people selling llms.txt files as a magic ranking fix. The evidence doesn’t support it. Independent analysis of over 137,000 sites found that 97% of llms.txt files are never requested by an AI system at all, and having one shows no correlation with being cited more often.

We’ll add one, because it costs nothing and the Lighthouse audit looks for it. But we won’t build your strategy on it, and we’d be wary of anyone who does. The things that genuinely move the needle are duller and more durable: clean semantic structure, proper schema markup, pricing and FAQs in readable text rather than graphics, fast stable pages, and content that answers real questions in full.