Google Lighthouse is starting to check if AI agents can use your website

Google just added “quietly” a new Lighthouse / PageSpeed Insights check called Agentic Browsing.

This is interesting, because it means Google is no longer looking only at how fast a site loads for humans or how well Googlebot can crawl it, but they now also start checking if an AI agents can actually use the site well and understand it well.

The Agent score is not the normal 0-100 Lighthouse score, it looks more like a step-based validation, something like 1/3, 2/3, 3/3 or 1/2 2/2, depending on how many checks pass and what type of device you check.

From what I understand, the audit looks at things like:

WebMCP support - if the site exposes actions/tools in a way agents can understand.

Accessibility structure - buttons, labels, forms, names, roles, etc. Basically, if the page is OK for screen readers, it is also OK for agents.

CLS / visual stability - if the layout moves while loading, an AI agent might click the wrong thing.

llms.txt - the file some sites are starting to use as instructions for LLMs.

I think that this is worth watching…

Also, something to watch out for is that a lot of plugins and themes work well visually, but the HTML underneath is a real mess… divs over divs, fake buttons, custom forms with no labels, this can confuse AI agents.

I still don’t think this is something to panic about today, as it looks early and experimental. But the direction is: websites will not be built only for users and crawlers anymore, they will also need to be built for AI agents.

2 Likes

Interesting,

Though I can’t help but feel it’s optimising for the wrong thing. Shouldn’t the agents be the ones smart enough to navigate websites like a human does?

2 Likes

Makes no sense to me the world is going insane!

3 Likes

Exactly! This tells us that the agents should be smart, but are not smart. Currently they seem to be some sort of web crawlers which pretend they are smart and can do more stuff on the crawled links, but because they have more power and are “smarter”, they can actually end up doing dumb things, especially if served a messed up HTML.

2 Likes

This will not work not with what we have now. AI is cool but they push it too hard is not ready for real world, all companies that use AI for coding end up fixing all ‘productive’ code by a senior dev and I mean all cases…

Is funny that now with AI even though most of the code is written by AI we work more than ever as devs…

1 Like