Manifesto · rubric v0.7
The turn
The web is no longer only for people. It is becoming a working environment for AI agents, and this page reports how ready the sites we measured are for them.
The numbers this is based on
- Sites measured
- 12
- of 12 published candidates
- Highest
- 61.6
- vercel.com
- Median
- 34.2
- Agent-Integrated
- 0
- out of 12 reached L4
Read from the stored scan records of the 12 published candidates; the newest of them was measured 2026-10-12. Measured under rubric v0.7 · 61 checks · 8 site profiles.
The web was built for people with screens. Agents are becoming its users — and the turn is quiet, because nothing breaks the day you miss it.
We measured 12 of the 12 published candidate hosts, all under rubric v0.7. They are sites run by organisations that build the web. The highest score was 61.6 out of 100, on vercel.com. The median was 34.2. Not one of them reached Agent-Integrated.
Each of those sites is built for a person with a browser, and the gaps measured here do not show in one. An agent does not complain. It reads what a page serves without a browser, and when that is nothing, it moves on.
What that means in practice
| What the site does for a browser | What the agent gets |
|---|---|
| Renders a React storefront | 200 OK and an empty shell, because nothing executes |
| Returns a friendly 404 page | A 200 with HTML where a JSON document was promised |
| Publishes a marketing homepage | A summary it cannot use, and no answer to what the product does |
| Blocks scrapers by IP range | Indistinguishable from a site that blocks everybody |
| Has a pricing page a human can read | No Offer, no price, nothing to quote |
| Has a support inbox | No agent policy, so the agent cannot tell if it is welcome |
Why the gap is an advantage, for now
Of the 12 measured under rubric v0.7, 8 scored under 40 and 4 scored under 30. That is not a verdict on those teams. It describes an industry that has not been asked to do this yet. Standards are being written now, and each check page names the artifact to ship.
Once a competitor is legible to the agent choosing between you, the gap is no longer neutral. It becomes a routing decision made against you, and nothing tells you it happened.
What we will not do
- We will not give a score to a site we cannot read. A site that blocks us is reported as not measured, with the reason.
- We will not turn our failure to understand your site into work for you. Checks that do not apply are reported as out of scope, and the report says why.
- We will not tell you to publish a URL that does not resolve, or cite a standard that does not exist.
- We will not publish a score without the check outcomes behind it, and a fingerprint that binds those outcomes.
- We will not mark your site verified unless you put the token there yourself. An exemplar is our measurement, labelled as ours, and withdrawn on request.
Where to start
- Run a scan. It requests 103 fixed paths, plus the endpoints a site's own documents declare, up to a published ceiling.
- Open the fix list. Every failure carries the measured value, a command that reproduces it, and the artifact to ship.
- Paste the prompt into the agent you already use, or download the bundle and hand it to your agent.
- Re-scan after you ship, then open the fix list again.
What readiness actually is
AgenticWebScore fetches a domain from the outside, the way an agent would, and measures 61 checks across 5 dimensions. Readiness is not a ranking, and it is not AI visibility. It asks whether an agent can find you, read you, tell who you are, and act on your service without a person in the middle.
- Protocol Discovery: Do you expose your service as a callable tool and an authorisable API?
- Content Accessibility: Is your content genuinely readable as markdown, semantic structure and structured data?
- Discoverability: Can an agent find you, reach the right page and understand what you offer?
- Bot Access Control: Can you tell agent traffic apart by identity?
- Commerce Protocols: Which standard do you accept agent-initiated transactions over?
The full contract, including the site profiles that decide which checks apply to you, is on the methodology page. It is the reference for everything on this site.