Alex Yarosh Get Free Snapshot

Source record

@tjrobertson52 TikTok profile avatar

@tjrobertson52

2026-07-10

Audit what browser agents actually extract from key pages instead of assuming raw HTML represents the complete experience.

Share source record
Tiktokexcerpt onlyen1

Source Text

i

We are quickly moving into a world where most interactions on your website will be done by an agent. Instead of a human visiting your website themselves, an agent will be visiting your website on their behalf. So I just want to share some new information we have about how agents see your website and how you can prepare for this new future.

Dan Petrovich recently put out a very interesting blog post talking about how Google's Chromium uses something called a page content agent to understand the content on your website. They ran a bunch of experiments and extracted exactly how Google sees your website. And this is all happening inside of a Chrome browser.

I'm sure other browsers are starting to do the same thing, but I doubt any of them are as ahead on this as Chrome is. Keep in mind that most browsers do run on chromium. Using Microsoft Edge, for example, you're still using Chrome.

And it's true of most headless browsers as well. If you're using playwright or puppeteer, you're using chromium. And if you're using Gemini in Chrome, Chrome will pass this data directly to Gemini, but it could be passed to any large language model.

So since essentially every AI has access to some version of chromium, I think it's safe to say that all AI will have access to this data from chromium. And what Dan Petrovich revealed in his blog post is that Chrome's page content agent has become remarkably Good at taking the content of a webpage and stripping it down into markdown for the agent. And at first that sounds like a very simple task, right?

You should just be able to look at the HTML mail, strip out the text, and there you have your markdown. But it's actually a lot more complicated than that. First issue is that not everything that's rendered on the page is actually included in the HT

own. But it's actually a lot more complicated than that. First issue is that not everything that's rendered on the page is actually included in the HTML.

Some elements are loaded by javascript and won't be shown in the HTML. The page may have retrieved information from your specific account, and that won't show up in the HTML. Or there may be form fields on the page that you've entered information into.

Chrome's page Content Agent can retrieve all of that information. Just a few months ago, the only way for AI to get his information was to actually take screenshots of the page. Then they might scroll or click a button and take another screenshot.

And every screenshot would take a couple seconds to process. It was incredibly slow and consumed a lot of tokens. What's perhaps most remarkable about Chrome's Page Content Agent is it runs entirely on your browser, so the page extraction happens almost instantly and essentially for free.

And because it's on your hardware, it can also retrieve sensitive information. In Dan's blog post, you're surprised to see that it even retrieved information from your bank account, for example, or from password fields. But then before that, Information leaves your device.

Google strips out the sensitive information. But the Page Content Agent doesn't just help AI read the content on your page, it also helps it interact with the page. Every element that it documents on the page also comes with coordinates, so the AI knows exactly where the buttons are and compress them without having to look at the page.

So the big news here is that AI agents just got way better at interacting with websites. What can you do about it right now? I recommend visiting your website on Chrome and then opening the Gemini for Chrome in the side panel.

Just ask it to g

can you do about it right now? I recommend visiting your website on Chrome and then opening the Gemini for Chrome in the side panel. Just ask it to give you the raw output from its page content agent when viewing your website.

Then look through the output and make sure it's understanding your website correctly. In my experience, it does a really good job even if you haven't prepared your website for AI agents. But if you haven't yet set up webmcp, it might have trouble understanding your forms.

Source Intelligence

i

Audit what browser agents actually extract from key pages instead of assuming raw HTML represents the complete experience.

AI agent website readability · asserts

  • Open Gemini for Chrome on important pages, request the raw page-content-agent output, and compare it with the page's intended meaning and forms.

Questions this source answers

i

What is this source mainly about?

Audit what browser agents actually extract from key pages instead of assuming raw HTML represents the complete experience.

What should an operator take from it?

Open Gemini for Chrome on important pages, request the raw page-content-agent output, and compare it with the page's intended meaning and forms.

Which topics does it connect to?

This source is connected to AI agent website readability.

What public evidence supports the record?

I recommend visiting your website on Chrome and then opening the Gemini for Chrome in the side panel. Just ask it to give you the raw output from its page content agent when viewing your website. Then look through the output and make sure it's understanding your website correctly.