Agentic Browsing: How well can an AI read your store? Here's how to measure it

People increasingly no longer search themselves, they let an AI assistant do it. ChatGPT, Claude and Gemini open stores, read product pages and present a recommendation. The only question is: can the AI even read your store cleanly? There is now a score for exactly that, and you can measure it yourself in a few minutes.

What Agentic Browsing measures

Google Lighthouse, the tool behind PageSpeed Insights, has a fifth, experimental category next to Performance, Accessibility, Best Practices and SEO: Agentic Browsing. It rates how well an AI agent can read and operate your page. So not how pretty the page is for people, but how clear the underlying structure is for a machine.

The number surprises many. I measure it regularly on real stores, and classic WooCommerce or Shopware sites often sit at 7 to 30 out of 100 on desktop, while the same stores score 90 or more on SEO. Translated: found by Google, missed by the AI.

How to measure it yourself

Through the normal PageSpeed interface the experimental category is not always visible. The most reliable way is the free PageSpeed API. All you need is a free API key from the Google Cloud Console.

Then open this URL in your browser, once for mobile and once for desktop:

https://www.googleapis.com/pagespeedonline/v5/runPagespeed
  ?url=https://YOUR-STORE.com
  &strategy=desktop
  &category=performance
  &category=accessibility
  &category=best-practices
  &category=seo
  &category=agentic-browsing
  &key=YOUR_API_KEY

In the JSON response the value is at lighthouseResult.categories.agentic-browsing.score (0 to 1, so multiply by 100). If you prefer a script, grab the response with curl and filter with jq:

curl -s "https://www.googleapis.com/pagespeedonline/v5/runPagespeed?url=https://YOUR-STORE.com&strategy=desktop&category=agentic-browsing&key=YOUR_API_KEY" \
  | jq '.lighthouseResult.categories["agentic-browsing"].score * 100'

Always measure both, mobile and desktop. The values often differ a lot, and the weaker one shows you where the work is.

The three most common causes of a low score

A poor score almost always has the same roots:

1. No structured data. Without a Product schema (JSON-LD) carrying price and availability, the AI has to guess. Check your product pages with the Schema validator.
2. Content only appears in the browser. If the product page is fully assembled by JavaScript, an agent often sees nothing but an empty shell. Server-side rendered HTML fixes this.
3. No llms.txt. Open your domain plus /llms.txt. If you get a 404, the AI is missing a map of your store. A short text file with your most important pages is enough to start.

What to do with the number

The Agentic Browsing score is still experimental, so don't treat the exact figure as gospel. As a direction indicator it is worth its weight in gold: it shows you in black and white whether your store is readable for the new generation of AI search or not. Measure it today, note it down, and measure again after every optimization. That is the difference between a gut feeling and proof.