Agent Experience (AX)
Agent Experience is how well a website serves an agent acting on a person's behalf. More people now ask an agent to find a price, compare plans or locate the returns policy for them. If the agent cannot get in, cannot read the page, or gets lost halfway, the person never arrives either.
We are building a way to measure it, starting with a closed beta. Nothing is scanned without the site owner asking.
Request beta access →What AX is, and how it differs from SEO and GEO
Search engine optimisation is about being found: a crawler indexes your pages and ranks them for a query. Generative engine optimisation is about being cited: an AI answer quotes or links your content when someone asks a question.
Agent Experience is about being usable. The agent is not summarising your site for someone else; it is acting inside it, on behalf of a person who has a job to get done. A site can rank well and be cited often and still turn an agent away at the door, hide its prices behind a script the agent cannot run, or label its buttons in ways no machine can follow.
What an AX Score measures
An AX Score looks at a site through four lenses. Each one is reported on its own, with the evidence behind it, so you can see which part to fix first.
Reachability
Can an agent get in at all? We read your robots rules and AI directives, note any bot wall or challenge on the first load, and record the HTTP status and whether the content sits behind a sign-in.
Legibility
Once it is in, can the agent understand what it sees? We look at the accessibility tree, labels and accessible names, structured data, and how much an agent has to read before it finds what matters.
Declared surface
What does your site tell agents about itself on purpose? We check for llms.txt, the sitemap, API and MCP manifests, AGENTS.md, and any stated agent policy.
Task success
Does it actually work? A pinned reference agent attempts a fixed set of read-only tasks, and we record whether it finished each one and where it gave up.
How a run works
A pinned, version-stamped reference agent visits the site and attempts a fixed set of read-only tasks: find the pricing, find a way to contact sales or support, find where to sign up, and find the returns policy or the terms.
The agent never submits a form, never signs in with real credentials and never buys anything. It runs only on domains the requester owns and has verified, so nobody can point it at a site that is not theirs.
What the report shows
The report shows the findings for each lens and the point where the agent dead-ended on each task, with the page, the step and what it saw.
Blocked at the door
If your site challenges the agent on the first load, with a captcha, a bot wall or an interstitial it cannot pass, the report says so plainly. It is recorded as a Reachability finding, not as a low score. There is no composite number, and the other lenses are marked as not measured, because we could not see them.
What it means: a person's own agent meets the same door. If someone asks their assistant to check your pricing, it stops where ours stopped. The fix usually starts with how your bot protection treats declared, well-behaved agents, rather than with the page itself.
Methodology
The rubric is versioned, and every report is stamped with its provenance: when the site was captured, which agent version ran, and which rubric version judged it. The deterministic lenses come first, so most findings rest on what the page served rather than on a model's opinion.
The full methodology will be published with the rubric when the beta opens.
Questions
- Is this the same as agent readiness?
- No. Agent readiness is usually a checklist of files and settings. AX measures what actually happens when an agent tries to do something on your site, and the checklist items are only part of one lens.
- Which agents?
- One pinned reference agent, versioned, so two runs of the same site can be compared fairly. We name its version on every report.
- Will you scan my site without asking?
- No. A run happens only on a domain you own and have verified. Requesting beta access does not start a scan.
- What if my site blocks the agent?
- Then the report says it was blocked at the door. That is a Reachability finding with no score attached, and the report explains what it means and where a fix usually starts.
- Is there a score I can see now?
- Not yet. AX is in closed beta and there are no public scores. Request access below and we will be in touch when there is a place for your site.
Request beta access
Tell us which site you would like measured. We read every request and reply by email.