Agent experience: what AX means for your website

Most websites still assume their first reader is a person. Increasingly, it might be a system deciding how, or whether, you are shown to that person at all. AI systems are starting to sit between people and websites, reading pages, summarising options and, in some cases, taking action on a user's behalf. Before a human ever lands on your homepage, something else may have already read it and formed a view.

Websites have always been built for people. The problem is that an AI system doesn't read a page the way a person does. It cannot be relied on to interpret a slick layout, fill in the gaps around a vague tagline or sense what you probably meant. It works with what's actually on the page, and if that isn't clear, it draws its own conclusions.

That gap is what the AX conversation is really about: how well an AI agent can understand and act on your website. This post looks at what AX means, why it matters, and what you can do about it.

What is agent experience?

In a website context, agent experience, or AX, is how well an AI system can make sense of your site and act on it reliably. The term was introduced by Mathias Biilmann at Netlify, who frames AX as the experience AI agents have as users of a product or platform. We learned more about this from Mathias at Storyblok's Joyconf last year.

Most teams already think about UX, the experience people have on a site. Some also think about DX, how easy it is for developers to build on a product or use its tools. AX is the same instinct pointed at a different reader: the AI systems that now make sense of your content and act on it.

AX is not about replacing human users or building some cold, machine-first version of your site. If anything, it pulls in the opposite direction. Clear language, sensible structure and explicit information help an agent, but they help people just as much. AX is really a stricter version of good content and clean structure. The difference is that an agent has no patience for ambiguity and no instinct for filling in gaps.

Why AX matters now

Discovery is changing. People aren't always moving from search result to website anymore. They used to type a query, scan a list of links and click through to a few sites. Increasingly, they ask an AI tool a question and get an answer that has already done the reading for them.

People now lean on these AI tools to find a supplier in a particular field or location, or to compare a handful of providers and have the options summarised in plain terms. Ask the right question and you'll often get a specific company recommended back. A growing number go further still, handing the work to an agent that can act on it: checking availability, submitting a form or pulling together details on your behalf.

When that happens, an AI system becomes the first reader of your website. It shapes how your business is understood, summarized and surfaced, often before a human ever sees your homepage. If your site is vague or awkwardly structured, that system has less reliable material to work with, and it will work with whatever it can find.

Get this right and an agent reading your site quickly is far more likely to describe your business accurately to the person who asked.

AX starts with clear content

Agents need clarity, not cleverness. A clever tagline that hints at what you do might charm a human reader who already has context. An agent has none of that context, and it won't guess generously on your behalf.

This is where a lot of sites fall down. Vague positioning, service pages padded with filler, and calls to action that don't actually say what happens next all make a site harder to interpret. If a page spends three paragraphs talking about your "passion for excellence" before mentioning what you sell, an agent has to dig for the part that matters.

Good AX depends on content that answers the obvious questions plainly: what you do, who you help, what you offer, where you operate, what makes you credible and what someone should do next. That sounds basic, but it's exactly where sites tend to slip. None of this is separate from good human-facing content. It's the same discipline, applied a bit more honestly. The page an agent can parse cleanly is usually the page a busy human can scan in ten seconds and understand. If you've ever read your own service page and struggled to summarise it in one sentence, that's a signal worth paying attention to.

Structure matters as much as wording

Clear words help, but words alone aren't enough. Agents also need to understand how information relates across a site. A page can be well written and still sit in a confusing structure that makes its role hard to place.

A well-structured site makes the important things obvious: which pages matter, how your services relate, what someone should do next. Most of this comes down to a few habits that aren't hard, just easy to neglect. Headings should say what a section is actually about, not lean on clever labels that only land once you've read the whole page. Internal links should describe where they go, so "our approach" beats yet another "click here". And the same service shouldn't be called three different things across the site, which sounds obvious until you audit your own pages and find it happening. Where it fits, schema markup can spell out things like products, locations and reviews explicitly, so a machine doesn't have to guess at what a page is describing.

Underneath all of this sits your content model. On a lot of sites, a service page is one big block of formatting: the headline, the description, the pricing and the call to action all live in the same freeform field, held together by how it happens to look. An agent reading that has to work out which part is the price and which part is marketing. When each piece of content has its own defined field instead, the meaning is far easier to read reliably. This is where the way a site is built starts to matter. On a component-based or headless setup, you can plan that structure deliberately, so a "service" is always a service with the same fields, rather than something reassembled by hand on every page. The information lives in defined, reusable pieces rather than being trapped inside one layout. That helps agents, but it also helps the humans maintaining the site a year after launch. It's not the only way to get there, but it does make the structure something you decide on purpose rather than something that accumulates by accident.

AX and technical foundations

Content is only half of it. An agent has to reach your content, read it and trust that it won't move under its feet. Most of what makes a site agent-friendly comes down to those three things: reachable, readable, stable.

None of this is exotic. It's the kind of build quality websites should have had anyway. Content rendered on the server, rather than assembled later by scripts an agent may never run, is actually there to be read. Semantic HTML means a heading reads as a heading and a list as a list, which helps assistive technology and automated systems in exactly the same way. Stable URLs mean the references an agent forms today still work tomorrow. From there, the supporting details matter too: titles and metadata that describe the page honestly, structured data where it earns its place, and clear error states so a system can tell "not here" from "broken". And if you want agents to do something rather than just read, the APIs behind that need documenting properly.

The aim here is not to build a separate "AI version" of the site. That would be a maintenance headache and a quiet admission that the real site is hard to read. The aim is to make the real site easier to understand and interact with, whether the reader is a person, a search engine or an agent.

What poor agent experience looks like

Bad AX is easier to spot than good AX is to describe. You see it when a service page spends three screens on company values before getting anywhere near what's actually for sale. You see it on a location page that talks warmly about "serving the local community" without ever naming the towns it covers. You see it when the eligibility criteria that decide whether someone is a fit, lives in a downloadable PDF brochure rather than on the page itself, making them much harder for anything that doesn't open and parse the file cleanly. And you see it in the small inconsistencies that add up: the same thing called a "service" on one page, a "solution" on the next and an "offering" on the third, until a reader can't tell whether these are three things or one. Add weak headings, layouts that carry meaning the words don't, thin metadata and the odd broken link, and an agent is left guessing at the parts that matter most.

An agent working with this kind of site may summarise the business badly, miss details that would have made you a strong match, or simply pick a clearer competitor instead. It won't email you to ask for clarification. It will work with what's in front of it and move on.

What businesses can do now

You can start improving AX without chasing every new AI announcement. Most of the work is sensible web hygiene that's been worth doing all along.

Start with your core service pages, because that's where most of the damage hides. Read each one and try to summarise it in a sentence. If you can't, the page probably can't either, and no amount of schema will rescue a page that doesn't say what it's for. Next, get essential information out of PDFs and graphics and into readable text on the page. Then make the structure obvious: headings and internal links that reflect how things genuinely relate, and a content model where meaning lives in defined fields. Schema comes after the basics are clear, not before. And document your APIs only if you actually want agents to act, not just read.

Even if agent traffic stays messy and hard to measure, this is still work worth doing. Clearer pages, cleaner structure and accessible markup help the site you have now, for SEO, accessibility and the people using it today.

This is really just better web work

None of this displaces SEO, UX or accessibility. An agent reading your site cares about all three at once: it needs to find your content, make sense of it and trust that it holds together. That's why AX tends to expose the same gaps those disciplines were already trying to close.

AX isn't a fifth project competing for budget. It's a way to test whether the work you're already doing holds up when a machine is reading.

Where this leaves you

AX is likely to matter more as AI agents become a normal part of how people search, research and decide. You don't need to rebuild anything around it. The gains come from making your existing site easier to understand and act on. That is work worth doing whatever any model can or can't do today.

If you want a single principle to take away, it is this: if an AI agent cannot understand what your business does, it will struggle to explain it to the people you actually want to reach. The fix is not clever. It is clarity, structure and honesty about what your site is really saying.