This is the AI agent. And right now, most websites are functionally broken for it.
In this article, we explore:
- Why crawlability is the foundation everything else depends on
- Why AI agents interact with your site in a fundamentally different way than human visitors
- The three methods AI agents use to read your content — and what breaks each one
- What this means for your development, UX, and content teams
- Why non-commodity content and media are now GEO signals
- Why accessibility and GEO are now the same problem
- Where to start if you want your site to be visible in AI-driven search
Before anything else: can AI even find your content?
There's a question that comes before structure, accessibility trees, and semantic HTML. It's more fundamental — and more commonly overlooked.
Can Google's AI systems actually crawl your content in the first place?
Google is explicit: AI search features are built on publicly accessible, crawlable content. If your site blocks crawlers through robots.txt, CDN configurations, or hosting infrastructure, your content simply doesn't enter the picture. No crawl, no citation, no visibility — regardless of how well-written or well-structured the content is.
JavaScript is where most teams have a hidden problem. Google can process content rendered via JavaScript, but only when it's not blocked and only when it renders correctly. Many sites have critical content — product descriptions, category pages, key articles — that lives inside JavaScript components that load too slowly, render client-side only, or aren't accessible to crawlers at all. That content is invisible to AI search features even if it ranks reasonably well in traditional search.
What to check: Use Google Search Console's URL Inspection tool to see how Googlebot actually renders your most important pages. What you see there is roughly what AI systems see. If content is missing from the rendered view, it's missing from AI search too.
AI search features use publicly accessible, crawlable content to learn patterns and provide grounded responses. If your content can't be crawled, it can't be cited.
— Google Search Central, AI Optimization Guide, May 2026
The shift most teams haven't made yet
For two decades, we've built websites for human eyes. We've obsessed over visual hierarchy, hover states, motion design, and the emotional weight of a well-chosen typeface. That work still matters. But it now shares the stage with something else: the machine-readable quality of your site.
Google's AI optimization guidance and the web.dev framework for agent-friendly websites make this explicit. AI agents — the autonomous systems that browse, research, and act on behalf of users — interact with your site in a fundamentally different way than human visitors do.
If your site isn't legible to them, it simply doesn't exist in their world.
AI Overviews, AI Mode in Google Search, and agent-driven browsing are already shaping which content gets surfaced and cited. The gap between AI-readable and AI-invisible sites is opening right now.
— Caroline Gadd, Global SEO Lead, Valtech
This isn't a future problem. The window to act is now.
How AI agents actually read your site
Understanding the problem starts with understanding how these systems see your content. Agents use three primary methods — and most sites fail on at least one.
Screenshots
The agent captures a visual rendering of your page and uses a vision model to interpret it. It can recognize that a search bar in the top right is a global search, or that a large button in the centre of a product page is the primary action. Visual cues carry meaning.
But screenshot analysis is slow and computationally expensive. It's a fallback, not a primary method.
What breaks it: Shifting layouts. If the same element appears in a different position depending on product category or screen state, the agent loses confidence in what it's reading. Consistency is no longer just a UX principle — it's a visibility requirement.
Raw HTML
The agent reads your DOM directly. It understands nesting, hierarchy, and the relationship between elements. If a "Buy Now" button is inside a product container, the agent infers that button belongs to that product.
What breaks it: <div> soup. A layout built entirely from non-semantic elements gives the agent no structural signal. A <div class="btn"> tells it nothing. A <button> tells it everything.
The accessibility tree
This is the most important channel — and the least understood.
The accessibility tree is a browser-native API that distils your DOM into what actually matters: the roles, names, and states of every interactive element. It's what screen readers use. It's also what AI agents use as their primary, high-fidelity map of your site — filtering out visual noise to focus on pure functional intent.
What breaks it: Missing ARIA labels. Unlabelled icons. Interactive elements built from non-semantic HTML. Form fields without descriptive attributes. All of these create gaps in the accessibility tree — and where there's a gap, the agent is blind.
Modern agents combine multiple modalities — DOM, accessibility tree, and visual rendering. Our job is to provide clean signals across all three channels.
— web.dev, Build agent-friendly websites
What this means for your team
This isn't a problem owned by SEO alone. The accessibility tree sits at the intersection of development, UX, and content. Fixing it requires all three.
For development teams: Semantic HTML is no longer a best practice — it's a visibility requirement. Every <button>, <nav>, <article>, and <section> element is a signal to AI systems about how your site is structured. ARIA labels, descriptive alt text, and clearly named form fields feed directly into the accessibility tree that agents rely on.
For UX teams: Stability matters more than dynamism. Shifting layouts, progressive loading without clear state management, and interactive elements that change position based on context all degrade the agent's ability to build a reliable model of your page. Consistency is a new UX requirement — not just for returning human visitors, but for agents that need a predictable map.
For content teams: Structure is signal. Clear heading hierarchies (H1 → H2 → H3), concise link text, summaries at the top of long-form content, and well-labelled data tables make your content more extractable. Google's guidance is explicit: content that is well-structured, authoritative, and citation-friendly is content that gets surfaced in AI-generated answers.
But structure alone isn't enough. Google's AI optimization guide draws a sharp line between content that gets cited and content that gets ignored — and the dividing line is expertise. Generic, broadly available information has no advantage in AI search. AI systems already know the common answers. What they look for when deciding what to cite is content that offers something beyond common knowledge: original analysis, first-hand experience, expert perspective, and data that doesn't exist elsewhere.
This is what Google calls non-commodity content — and it's the single most important content signal for GEO performance. A well-structured page full of generic information will lose to a less perfectly formatted page that contains a genuine expert insight every time.
For content teams, this means one practical shift: before publishing, ask whether the content contains something an AI system couldn't have assembled from existing sources. If the answer is no, it needs more depth, more specificity, or a stronger point of view before it's ready.
Images and video are an underused GEO signal. Google's AI features support multimodal search — users can now search with images, not just text. Pages that pair strong written content with high-quality, relevant images and video have more surface area for AI systems to work with, and more opportunities to appear beyond standard web links. If you're already following image SEO best practices — descriptive filenames, alt text, and structured data — you're already optimising for AI search.
The same structural quality that makes a site accessible to users with assistive technology is the structural quality that makes it legible to AI agents. They share the same interface: the accessibility tree.
— Caroline Gadd, Global SEO Lead, Valtech
Accessibility debt is now GEO debt
Here's the insight that changes the conversation.
A site that scores poorly on accessibility criteria — missing labels, poor heading structure, non-semantic interactive elements — is also a site that AI agents will struggle to read, cite, and surface.
And conversely: investing in genuine accessibility improvements creates measurable GEO gains. Fix the label on a complex navigation element, and you've simultaneously improved the experience for screen reader users and made that navigation legible to every AI agent that visits your site.
This is not a coincidence. It's a design principle. Build for the broadest possible range of visitors — human and machine, sighted and not — and you build a site that performs across all channels.
Where to start: five practical steps
- Check your crawlability first. Open Google Search Console and run the URL Inspection tool on your five most important pages.
- Open Chrome DevTools and review your accessibility tree.
- Run a semantic HTML audit on your most important pages.
- Test your heading structure as a standalone outline.
- Audit your images for alt text and descriptive filenames.
These five steps won't fix everything. But they will show you, quickly and concretely, where your site is invisible to the systems increasingly deciding what gets found.
The opportunity
Most organisations are still treating AI search optimisation as a future consideration. The ones that move first will build a structural advantage that compounds over time.
AI agents don't just visit once. They return. They update their understanding of your site. A site that is clean, consistent, and semantically well-structured gets better treatment over time — cited more often, surfaced more reliably, trusted more readily.
The invisible visitor is already here. The question is whether your site is ready to receive it.
Ready to find out where your site stands?
Valtech's GEO audit maps your site's crawlability, accessibility tree health, semantic structure, and AI citation readiness — and translates it into a prioritised action plan your dev, UX, and content teams can act on.
Get in touch to book a GEO audit.
About the author
Caroline Gadd is Global SEO Lead at Valtech, working at the intersection of organic growth, AI search, and digital experience strategy. She leads Valtech's GROW Smarter initiative, building shared capability across AI, SEO/GEO, data, and accessibility.
