Your website was built for humans. AI reads it differently.

When a human visits your website, they experience layout, colour, hierarchy, and motion. They read headlines, scan images, and follow the visual logic you have spent time crafting.

When an AI agent visits, it sees none of that. It reads the underlying code (raw HTML, text content, structured data) and builds its own understanding of what you do, what you know, and whether you are worth citing in an answer.

This is not a technical curiosity. It is a fundamental shift in who your website is actually for.

‍

Why this matters now

‍

This shift is not hypothetical. 37% of consumers now start their searches with an AI tool instead of Google, and AI referral traffic to websites is growing at well over 100% year on year. Some of that traffic never reaches a website at all - the AI answers the question itself, using content it has already read and judged credible. If your site is not built in a way that AI can read and trust, you are not just missing out on rankings. You are missing the conversation entirely, and increasingly, so is your business.

‍

What you see is not what it reads

‍

A human can interpret ambiguity. They can follow a visual metaphor, understand that a clever headline implies expertise, and feel the weight of a brand through its design choices. An AI cannot. It needs content that means the same thing in isolation as it does in context - because it will never have the context.

Here is where it gets uncomfortable for most organisations: the sites they are proudest of are often the least readable by AI. Visually ambitious, heavily animated, JavaScript-rendered experiences built for maximum human impression can be nearly empty from an AI's perspective. The content that matters - the expertise signals, the things that actually answer the questions people are asking - only loads after scripts run. AI agents read what is in the HTML at the point of arrival. If your key message comes later, it might as well not exist.

‍

What AI actually reads - and what it trusts

‍

The question most teams start with is the wrong one. 

They ask: can AI find our site? 

The better question is: when AI finds it, does it trust what it reads?

‍

Structure matters more than most people realise. Not visual structure - semantic structure. An H1 tells an AI this is the most important statement on the page. An article tag says this is editorial content with a defined purpose. A div says nothing. Most sites are built for how things look, not what they mean. For humans, that distinction is invisible. For AI, it is everything.

Beyond structure, AI reads for explicit credibility. This is the part that runs counter to a lot of brand instincts. The writing techniques that work on humans - implied meaning, brand personality, headlines that create intrigue without stating a claim - largely do not survive AI interpretation. An AI agent reads your content and decides whether it is credible enough to quote. Vague language does not make that cut. Named experts, cited data, direct answers to questions people ask: these are the signals AI treats as trust indicators.

There is a useful data point here. The A11y-CUA study, published at CHI 2026 by researchers at UC Berkeley and the University of Michigan, tested Claude on everyday web tasks under different access conditions. Under standard conditions, the agent completed 78% of tasks successfully. Using keyboard-only navigation - the same access pattern screen readers rely on - success dropped to 42%. That gap matters. And it points to something broader: the disciplines we treat as best practice for human users - semantic HTML, accessible markup, clear hierarchy - turn out to be exactly what AI needs too.

‍

SEO is not dead. But the goal has changed.

‍

For two decades, the metric that mattered was ranking. Get to page one. Drive traffic. Convert visitors.

That model is not gone, but it is no longer sufficient. When someone asks an AI a question today, they often do not click through to a website at all. The AI synthesises an answer from multiple sources and occasionally names where it drew from. The goal is no longer just to be the first one that appears in the results. It is to be the source the AI trusts when it constructs its answer.

This shift has its own name now.. Generative Engine Optimization (GEO) and the principles are different from traditional SEO. The most visited page about a topic is not necessarily the one that gets cited. The clearest, most directly useful answer to a question is. That favours content written with genuine depth over content optimised for keyword density.

One concrete signal of how quickly this has moved: the llms.txt file. Modelled on robots.txt, it is a plain text file that tells AI systems what your site covers and where the most important content lives. Two years ago, almost no one had heard of it. Now it is considered basic hygiene for any site that wants to be legible to AI.

‍

What to actually do

‍

Start by understanding what AI sees when it visits your site. Not what you think it sees,. what it actually sees. Tools exist that render pages the way a bot would, without JavaScript or styling. Run your site through one and the result is often surprising, and not in a good way.

From there, the priorities become clearer. Semantic HTML used for meaning rather than aesthetics. Content that stands on its own without visual context, if a sentence only makes sense alongside an image, it will be misread or ignored. Structured data that explicitly tells AI what a page is about, not just what words appear on it.

The harder shift is in how you think about content. The pages that get cited by AI are not the ones optimised for impressions. They are the ones that clearly and specifically answer questions people are actually asking. Writing for AI readability and writing genuinely useful content turn out to be the same thing. That is not a coincidence.

A quick checklist to work through:

• Render your site with JavaScript and styling turned off. If your key message disappears, an AI agent never saw it either.
• Replace generic containers (div, span) with semantic elements (article, nav, section, h1-h3 in order) - check your heading hierarchy doesn't skip levels)
• Make every sentence stand on its own, without depending on a nearby image or visual cue.
• Swap implied claims for stated ones - named experts, cited data, direct answers.
• Add or update an llms.txt file at your root domain.
• Add structured data (schema markup) to key pages.
• Server-render or statically generate anything you need AI crawlers without JavaScript to see.

‍

This is a UX problem. We just have not been calling it that.

‍

The question UX has always asked is: who is the user and what do they need to accomplish their goal?

The answer used to be one person, with a screen, reading what you put in front of them. Now there is a second user, .the AI agent reading your content before a human ever sees it, deciding what to pass on and how to represent it. That is a UX problem. It has always been a UX problem. We have just been calling it something else.

At Digital of Things, it is increasingly the conversation we are having with clients before we talk about anything else. Not because it replaces the human experience (it does not) but because getting it wrong at this layer means the work you do for human users never reaches them in the form you intended.

Your content was always for someone. Now, it is for two.