Read · 53% of the score Read · 53% of the score
30.4%
of the whole web passes this
An agent can build a usable picture of your page from its markup.
The score · August 2026
What the AWCS measures
The AWCS is a percentage, from 0% to 100%, that says how ready a website is for the AI agents that read it. Nothing on the web is near 100% yet — the whole web averages 19.9% in August 2026 — so the interesting question is not who is perfect but who is moving. It is our score, not Google’s: Lighthouse does the measuring and we decide how the measurements add up. This page says exactly how.
Who is asking, and why it matters to you
More and more, the thing arriving at your site is not a person with a browser. It is an assistant somebody asked a question, and it has to read your page, work out what is on it and sometimes do something there — find a price, check availability, fill a form. A site built only for human eyes can be invisible to it: not slow, not ugly, invisible. The AWCS is an attempt to put a number on that, monthly, for the whole web, so that «are we ready for this?» has an answer with a figure in it.
The four things it looks at
Three of them add up to the score. The fourth multiplies whatever the other three came to, which is why a site can do everything right and still score zero. The percentage on each card is not yours — it is how much of the whole web passes that one thing this month.
Guide · 27% of the score Guide · 27% of the score
13.1%
of the whole web passes this
You serve an llms.txt at your root, and it parses.
Act · 20% of the score Act · 20% of the score
5.1%
of the whole web passes this
Your site exposes WebMCP tools an agent can actually call.
Reachable · multiplies the rest Reachable · multiplies the rest
96.0%
of the whole web passes this
Your robots.txt does not turn away the AI agents you want reaching you.
How the four become one number
Read is worth 53% of the score, Guide 27% and Act 20%. They are in that order because that is the order in which they stop an agent: one that cannot read the page never gets as far as your llms.txt. Reachable is different — it does not add anything, it multiplies. Pass it and your total stands; fail it completely and your total becomes zero, however good the other three were, because an agent your robots.txt turns away never fetches the page at all. Four scenarios, all of them the same site:
| Scenario | AWCS |
|---|---|
| An agent can read it, and that is all | |
| It can be read and it has a valid llms.txt | |
| All three, and every agent is let in | |
| All three, but the robots.txt turns every agent away |
What a number means
A percentage you have to calibrate for yourself is a percentage you cannot use, and this scale is not the one your instincts are set to: almost nothing on the web is above 40%, and no platform we measure reaches 60%. So every figure on this site carries a word, and the words come from this table. Where the boundaries sit is our decision and not a measurement — moving one changes what a figure is called and never what it is.
| Word | Range |
|---|---|
| Very poor | 0 – 10% |
| Poor | 10 – 25% |
| Weak | 25 – 40% |
| Fair | 40 – 60% |
| Good | 60 – 80% |
| Excellent | 80 – 100% |
Where the figures come from
From HTTP Archive, which crawls millions of sites every month with Lighthouse and publishes the results for anyone to read. We read a 0.1% sample of it — 14,373 sites this month, with Lighthouse 13.4.1 — compute the score in the database and publish it. Every figure on this site carries the sample it came from and its margin of error, and a movement smaller than its margin is published as a tie rather than as a change. Nothing here is measured by us, and nothing here is an estimate.