Last month Google quietly turned on a new Lighthouse category: Agentic Browsing. It scores whether an AI agent — the kind that now browses, compares and increasingly buys on a shopper's behalf — can actually use your site. As of Lighthouse 13.3 it's on by default, which means it shows up for every developer, client and competitor who runs an audit, whether they asked for it or not.
We wanted to know one thing: does it score a product page any differently from a homepage?
So we ran it across six DTC brands — Ramraj Cotton, Aeropress, Olipop, Verve Coffee, Antler and Owala — auditing the homepage and a real product page for each.
The scores didn't move
| Brand | Category | Homepage | Product page |
|---|---|---|---|
| Ramraj Cotton | Apparel | 0.67 | 0.67 |
| Aeropress | Coffee gear | 0.67 | 0.67 |
| Olipop | Beverages | 0.67 | 0.67 |
| Verve Coffee | Coffee | 0.67 | 0.67 |
| Antler | Luggage | 0.65 | 0.67 |
| Owala | Drinkware | 1.00 | 1.00 |
Agentic Browsing uses a fractional pass ratio, not a 0–100 score.
Five of the six scored identically on both page types. The sixth moved by a single point, caused by minor layout shift on the homepage. That was the entire difference across the whole test.
That's the tell.
What the audit actually checks
Agentic Browsing looks at four things: the accessibility tree (can an agent read your buttons and forms), layout stability (do elements move before the agent can act), llms.txt (do you publish one), and WebMCP (do you expose tools an agent can call).
Every one of those is page mechanics. None of them look at the product. Which is exactly why a homepage and a product page come back with the same score — to this audit, they basically are the same page.
A few things stood out across the cohort:
- 5 of 6 fail the agent accessibility tree — the only check actually pulling scores down.
- 0 of 6 register any WebMCP tools — nobody in this group has adopted the standard most of the current hype is built on.
- 5 of 6 already pass llms.txt — so the "your site is broken, pay us to add an llms.txt file" panic doing the rounds is, for real DTC brands, mostly a solved problem.
What it doesn't measure
Here's what no Agentic Browsing score will ever tell you:
- Whether your price, availability and variants are structured so an AI can quote them correctly.
- Whether your product feed is eligible for Google Merchant Center or ChatGPT's checkout.
- Whether the agent that can now navigate your page can actually recommend and sell what's on it.
That's not a page-mechanics problem. It's a product-data problem. And it's the half that decides whether you even show up when a shopper asks an AI to find and buy something in your category.
An agent can pass every check in Chrome's audit and click your "Add to Cart" button perfectly — and still get your price wrong, miss that you're in stock, or pick the wrong variant, because the data underneath the button is broken or missing.
The bottom line
Run the Lighthouse Agentic Browsing audit. It's free, it's built into the browser, and fixing accessibility and layout stability genuinely helps both humans and agents.
But it's a floor, not the game. For ecommerce, the score that matters is the one Chrome doesn't show you — whether your product data is structured well enough for an AI to understand, recommend and transact.
That's exactly what Product Rank scores. Run a free scan of your store →