Free test · no signup · under a minute
Assistants now visit sites on a buyer’s behalf, reading pages and filling forms.
Chrome ships a test for whether yours is usable by one. This runs that exact test
and shows you every result, including the elements at fault.
Every failing element is named, with the
line of code that broke it.
Why sites fail this
You know that button is a button because it looks like one. An agent gets
no picture. It gets a text outline of your page, the same one a screen reader uses.
What a person sees
Obvious. The logo goes home, the button gets a quote.
What an agent reads
navigation link ""no name button "Get a quote"main heading "What we do" paragraph
The logo link has no name, so the agent cannot tell what it does or
where it goes. It is a dead end.
A linked logo with no text inside it. An icon-only search button. A social icon in the
footer. Each one is invisible to an agent, and each one is a small fix.
What gets tested
Chrome’s agentic browsing audit is smaller than most articles suggest.
For a normal site it comes down to these three.
Fix this first
Whether every link and control has a name an agent can read. Fail this and an agent
cannot navigate your site at all, so nothing else matters yet.
Second
Whether the page holds still as it loads. If a banner drops in late and shoves
everything down, the agent clicks whatever moved into the gap.
Optional
A plain summary file at your root. Worth publishing so assistants can place you
quickly. Not required to pass, and a malformed one is worse than none.
The test also runs eight extra checks of ours on whether an assistant can read and quote
your page once it arrives: title, description, heading order, image alt text, language,
real links, crawlability and canonical.
Straight answers
Google, not us. This runs Chrome’s own agentic
browsing audit through Google’s official API, so every pass and fail is Google’s verdict
on your site rather than our opinion of it. What we add is the part Google does not give
you: each result in plain English, the exact elements at fault with the code that broke
them, and the order to fix them in so you are not guessing which one matters.
No, and be wary of anyone who says otherwise. This is
not a ranking factor. Google has said the audit reflects emerging standards rather than
search signals. What it measures is whether an agent acting for a buyer can actually
complete a task on your site. Treat it as usability for a new kind of visitor, not SEO.
Not to pass, and not for rankings. Chrome leaves the
check out of your score when you have no file, so it is never counted against you. Google
has confirmed llms.txt has no effect on Search or AI Overviews, and independent monitoring
of hundreds of millions of AI crawler visits found the file is almost never fetched. Anyone
telling you it will lift your traffic is guessing.
We recommend it anyway, for a narrower reason. Google’s own agent audit checks for it,
and its stated rationale is that without one an agent spends longer crawling your site to
work out what you do. It costs very little to publish. Just make sure it is well formed,
because a malformed file is the only version that actually fails the check.
If you want to determine which AI bots actually reach your site and what each one does, match your traffic against the specific user agents and behaviors of verified AI crawlers.
Because Chrome only counts checks that apply to you.
Three of the six audits cover WebMCP, an experimental protocol that returns a result only
if you have registered for Google’s origin trial. Almost nobody has. Those three drop out,
and so does llms.txt if you do not have one, which is why a typical site is scored on two
or three checks.
The address, no. The email, only if you give it. Running
the test sends your URL to Google’s PageSpeed Insights API and nowhere else. If you ask for
the fix list, your email and your result are recorded so I can follow up once. No list, no
newsletter, and it is not shared or sold.
Usually 30 to 60 seconds. Google renders your page on
its own hardware before scoring it, and large or slow pages take longer. Runs past 90
seconds happen. If Google does not answer within two and a half minutes the test stops and
tells you rather than spinning forever.
You get the full result whether or not you ever talk to me. If it comes back clean,
that is a useful thing to know too.
I run paid media for industrial and commercial companies. Every month I watch budget
go to pages that a buyer, or increasingly a buyer’s assistant, cannot get through.
When Chrome added this audit I ran it on my own site first. It failed, on a single
logo link with no text inside it. That took about two minutes to fix and I would never
have found it by looking at the page, because on screen it looks perfect.
So the test is free and complete. Run it, take the results, fix them yourself if you
want. If you would rather not, that is what the $750 is for.
André Rosdahl, Synthesis Insights
Also free: see which of your competitors are running Google and Meta ads right now, with the Competitor Ad Check.