Intro
While running a Google Lighthouse scan on a webpage that’s still in development, I found that there’s been an update to the scan categories since the last time I used it; Lighthouse now has an option to check agentic browsing. This was actually added during Spring 2026, but I’ve only just noticed it myself.
How It's Different
While the four existing scan categories were graded on a scale of 1 to 100, this agentic browsing check seems to only have two main areas that can be passed or failed. Basically, you’re aiming for a two out of two on this one.
According to its own description, the “checks ensure high-quality, browsable websites for AI agents” and “is still under development and subject to change”. Makes perfect sense seeing as the AI landscape changes drastically in short periods of time; trying to optimize for a large language model is like trying to catch a fish with your bare hands.
Layout Shifts
The analysis it runs for this gets rather technical and I don’t know much at all about dynamic tool registration or the structure of site’s accessibility tree, the criteria for Cumulative Layout Shift (CLS) really resonated with me.
In short, CLS measures how often the layout of a page shifts as a result of content loading, whether that be images, advertisements, API calls or others. This is something that you have almost definitely come across before, and if you’re anything like me, it enrages you to no end.
In fairness, there are two types of layout shifts: expected and unexpected. CLS measures the latter but I’ll quickly give you an idea of what these both are.
Expect the Unexpected
Expected layout shifts are caused by some intentional interactivity/movement on a page or element. For instance, if you’re making an account and confirming your password, you might see a block of text appear with a warning that your passwords do not match. An unexpected shift might occur when an advertisement pops up in the middle of an article you’re reading which causes your place in the text to abruptly shift downwards. Bothersome, and a poor user experience. Google’s example of this on their CLS development article is amusing; it depicts a sudden shift in buttons, causing a user to press the wrong one and inadvertently place an incorrect order.
I see unexpected shifts a lot on sites like w3schools.com because those guides are riddled with ads. I get it — bandwidth is expensive, and helpful resources like W3 deserve to exist for the benefit of learning developers everywhere. It’s just frustrating that the user experience gets flooded with unwanted ad space both as side content and between modules. It hinders the actual performance of the page, which is a separate problem altogether.
A remedy to this comes from placeholder content which is incredibly common. I used to be under the impression that placeholders and loading spinners were only designed to let the user know their content is coming. While that is partially true, I now understand that it also prevents the content from appearing in a flash and changing the layout of the rest of the page. This is great for images, especially when you consider the amount of detail that has to be established properly; size & resolution, file type, fetch priority, and so on.
Wrapping Up
I won’t bore you with the exact mathematics but a layout shift score can be calculated for a site’s many pages by multiplying certain fractions related to positioning. Suffice it to say, the lower the score, the better.
This article was more an excuse to talk about how CLS, while quite technical as a metric, is actually something that internet-users see all the time — whether they’re conscious of it or not. And how unexpected layout shifts can be enormously frustrating, even more so on mobile devices that have smaller viewports.
It also sounded strange to me when I first read it because the positions of elements on a given page seem like something that would only be relevant to human-users given that they’re the ones who have to navigate to and use them. I didn’t think an AI agent would at all be sensitive to the position of an element in order to use it; just that it would scan markup and find that the element exists at all. Maybe this is done so that the AI can provide context based on a layout and this would prove difficult or inconsistent if the layout has a poor shift score.
The agentic browsing category, much like everything else in the current tech landscape, is subject to change. Google might update its spec to change how these categories are scored, or even what parts of a page are looked at for the scoring. Who knows? I just found it interesting.
Want to read more?
My related article about Google Lighthouse: What Is Google Lighthouse?
Google's development article regarding CLS: Cumulative Layout Shift (CLS)
August 31st, 2026