Google's new web.dev guidance advises developers to treat AI agents as a distinct visitor type. Recommendations include semantic HTML, stable layouts, proper label wiring and cursor hints. A new WebMCP standard lets sites register tools that agents can discover and call as functions.
Everything we suggest to make a site agent-ready also makes sites better for humans.
Google published new guidance on web.dev telling developers to treat AI agents as a distinct visitor type alongside human users. The advice reads like accessibility best practices reframed for a world where software, not people, is increasingly navigating websites on behalf of users.
The core message: some human users are pivoting from manual navigation to delegating goal-oriented journeys to AI agents. Your website needs to work for both.
Ways AI agents interact with websites: screenshots for vision models, raw HTML/DOM parsing, and site-provided high-fidelity maps
What Google recommends
The guidance is practical, not theoretical. It covers three interaction models that AI agents use to navigate sites: visual screenshots processed by vision models, raw HTML and DOM parsing, and structured maps provided by the site itself.
For each, the recommendations are specific. Use semantic elements like `<button>` and `<a>` instead of styled `<div>` elements. Keep layouts stable across pages. Link `<label>` tags to inputs via the `for` attribute. Set `cursor: pointer` on clickable elements.
None of this is revolutionary for anyone who has built accessible websites. The difference is the framing. Google is telling developers that these practices now serve two audiences: humans using assistive technology and AI agents completing tasks on behalf of users.
WebMCP: the bigger signal
Buried in the guidance is a link to WebMCP, a proposed web standard for agent-website interaction with an early preview program. WebMCP lets websites register tools with defined input/output schemas that agents can discover and call as functions.
This is the more significant development. It means Google is actively building infrastructure for AI agents to interact with websites programmatically, not just visually. Sites that implement WebMCP give agents structured pathways to complete tasks without needing to parse ambiguous UI.
Why it matters
If Google is publishing developer guidance and proposing web standards around agent interaction, they are planning for a future where a meaningful share of web traffic comes from AI agents acting on behalf of users. That changes how websites need to be built and optimised.
For Australian businesses, the practical implication is that technical SEO and web development are converging. A site that is well-structured for accessibility, well-labelled for screen readers and semantically clean will also be easy for AI agents to navigate. Poor technical foundations will mean being invisible to both.
What to do about it
The businesses that benefit most will be the ones already doing technical fundamentals well. For everyone else, this is another reason to stop deferring the cleanup.