--- title: "Agentic Browsing: What Google's New Lighthouse Test Checks" description: "Agentic browsing in PageSpeed Insights: what the new category checks, how the score works and what business owners should do about it now." url: https://www.hamed.de/en/agentic-browsing/ language: en author: Hamed Farhadian published: 2026-10-25 updated: 2026-10-25 --- # Agentic browsing – Two kinds of visitors – people and AI agents. Agentic browsing is a new, still experimental category in Google Lighthouse and PageSpeed Insights. It checks whether AI agents can use a website reliably: whether buttons and forms are properly labelled, whether the layout stays stable, whether there is an llms.txt file and whether WebMCP is set up correctly. It isn’t a ranking factor, but it is a good measure of how ready your website is for the next kind of visitor. PageSpeed Insights has had a new category for a few months now. Next to Performance, Accessibility, Best Practices and SEO, there’s “Agentic browsing”. Most people haven’t even noticed. In this article I explain what the category is about, what exactly Google checks, how the score works and what you should do about it as a business owner. Using hamed.de as an example, I show how a website passes every check. ## What is agentic browsing? Agentic browsing means an AI agent visits a website on its own and gets things done there: finding information, comparing offers, filling in a form. The Lighthouse and PageSpeed Insights category of the same name checks how well your website is built for that. More and more people let AI search, compare and enquire for them. The agent then clicks its way through your website by itself. To do that, it has to understand it: what’s a button, what’s an input field, where does a link go? A person can tell by looking. An agent reads the structure of the page or works from screenshots. Google introduced the category with Lighthouse 13.3 in May 2026. Lighthouse is Chrome’s testing tool and the engine behind PageSpeed Insights. Google itself describes the category as experimental and “still under development”, so the checks may still change. ## What does PageSpeed Insights check for agentic browsing? Four areas: whether the page is built so agents can understand it, whether the layout stays stable while loading, whether there’s a usable llms.txt file and whether WebMCP tools are set up correctly. | Check | What it looks at | What you can do | |---|---|---| | Accessibility for agents | Do buttons, links and form fields have clear names? Are roles and page structure valid? Are controls visible? | Use real buttons, links and labels instead of look-alike elements. | | Layout stability | Does the page jump while loading? Measured as Cumulative Layout Shift (CLS) – how much content moves after it first appears. | Give images fixed dimensions; don’t push banners or notices in later. | | llms.txt | Is there an llms.txt file in the root directory – a kind of table of contents for AI? If so: does it have a heading, a sensible length and links? | Add a short llms.txt with a heading, two sentences about the site and links to your key pages. | | WebMCP | Does the website offer tools to agents? Are forms marked up accordingly and the descriptions valid? | Only relevant if you use WebMCP. Watch it for now rather than rebuilding. | Three of the four areas aren’t new. Clear labels have been part of accessibility for years, and a stable layout is part of the Core Web Vitals. What’s new is that Google now groups them under one question: can a machine use this website? ## How is the score calculated? Unlike Performance or SEO, there’s no score from 0 to 100. PageSpeed Insights shows a fraction: passed checks out of all checks that apply to your website. If your website doesn’t use WebMCP, the WebMCP checks drop out. If there’s no llms.txt, that check counts as “not applicable”, not as failed. That’s why results look different: a site without llms.txt can reach 2 out of 2, a site with one 3 out of 3. Add WebMCP and you’re also measured against the WebMCP checks. What that means for you: don’t just compare the number after the slash – look at which check fails. In PageSpeed Insights, every check shows you the elements affected. ## Is agentic browsing a ranking factor? No, not at the moment. The category comes from Chrome’s testing tool, not from Google Search. Google says explicitly that Search ignores llms.txt and that AI features need no special files. Google set this out in its [guide to optimising for AI features](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) and clarified it in June 2026: creating an llms.txt file is fine, but it neither helps nor harms you in Google Search. The [Lighthouse documentation](https://developer.chrome.com/docs/lighthouse/agentic-browsing/scoring) also describes the category as a way to gather data and give pointers, not as a final verdict. My view: this is exactly how it starts. First a metric, then a standard. Google introduced the Core Web Vitals as metrics in 2020 and only made them part of ranking in 2021. Build it properly now and you won’t have to retrofit later. I explain more about the role of AI search in [GEO vs SEO](https://www.hamed.de/en/geo-vs-seo/). ## How does hamed.de score? I built these checks into the relaunch of hamed.de from the start. The result on mobile: 100 four times over in Performance, Accessibility, Best Practices and SEO, and 3 out of 3 for agentic browsing. The three passed checks are accessibility for agents, layout stability and llms.txt. I don’t use WebMCP yet, so it doesn’t count. There are no tricks behind it, just a few fixed rules: - **Real controls:** every button is a button, every link a link, every form field has a visible label. - **Nothing jumps:** every image has fixed dimensions, and there’s no cookie banner sliding over the page later. - **Content in the HTML:** the full text is in the source code. JavaScript only adds to it and isn’t needed to read the page. - **An llms.txt plus text versions:** the [llms.txt](https://www.hamed.de/llms.txt) links to every page. Each page also has a plain-text version that AI systems can read without any design. How the other scores come about is explained in [improving PageSpeed Insights](https://www.hamed.de/en/improve-google-pagespeed-insights/). ## What should you do as a business owner now? Test your website once, fix any problems with labels and layout, and add a short llms.txt. WebMCP can wait for now. 1. **Test:** search for “PageSpeed Insights”, enter your domain and open the mobile view. The new category is on the far right. 2. **Check labels:** if accessibility for agents fails, PageSpeed Insights shows you the elements affected. Usually it’s buttons without text, icons without labels or form fields without a label. Fixing them helps people using screen readers too. 3. **Calm the layout:** a high CLS costs you twice – with visitors and with agents. Common causes are images without dimensions, late-loading ads and notices that push content down. 4. **Add an llms.txt:** a short file with a heading, two sentences about your business and links to your key pages is enough. Don’t overrate it: it doesn’t count for Google, but it’s an offer to other AI systems. 5. **Keep an eye on WebMCP:** the standard is an early draft. It gets interesting once agents can use it to request appointments or calculate quotes directly. What that means for you as a managing director: your website will soon have two kinds of visitors. People and AI agents. Both should find what they’re looking for. How to get mentioned in AI answers is covered in [getting visible in ChatGPT](https://www.hamed.de/en/how-to-get-visible-in-chatgpt/). If you want to know where your website stands today, try the free [SEO Compass](https://kompass.seoeffekt.de/?quelle=hamed.de-en-agentic-browsing-md&lang=en) – or see how I set websites up for AI search as a [GEO expert](https://www.hamed.de/en/geo-expert/). **SEO Compass: check for free where your website stands on Google and in AI search:** https://kompass.seoeffekt.de/?quelle=hamed.de-en-agentic-browsing-md&lang=en ## Questions about agentic browsing ### Is agentic browsing a Google ranking factor? No. The category belongs to Lighthouse, Chrome’s testing tool, not to Google Search. Google itself says that Search ignores llms.txt and that AI features need no special files. The results show how well AI agents can use your website – not how well you rank. ### Why does PageSpeed Insights show me fewer checks than other sites? Because only the checks that apply to your website count. If you don’t use WebMCP, the three WebMCP checks drop out. If there’s no llms.txt, Lighthouse marks that check as “not applicable” rather than failed. That’s why one site shows 2 out of 2 and another 3 out of 3. ### Do I need an llms.txt file? Not for Google. For other AI systems and agents it’s a cheap addition: a text file in the root directory that explains in a few lines what your website is about and links to the most important pages. Lighthouse then checks that it has a heading, isn’t too long and contains links. ### What is WebMCP? WebMCP is a proposed web standard that lets a website offer tools directly to AI agents – “request an appointment” or “calculate a quote”, for example – instead of the agent clicking through forms. It’s an early draft that Chrome is still trialling. For most businesses, keeping an eye on it is enough for now. ## Contact - Contact form: https://www.hamed.de/en/agentic-browsing/#kontakt - Email: hamed@hamed.de - Phone: +49 40 3556 7813