Methodology
How a site's Agent Readiness tier reflects the experience an AI assistant encounters.
Last updated: 4 October 2026
We assess whether your site can be discovered, understood and used by AI assistants. The method observes what an agent can actually find from the outside: it does not reward a promise, or an intention declared in an internal document.
A four-dimensional reading
- Visibility — is your business presented clearly when an assistant arrives? We look for a summary, capabilities and understandable entry points.
- Connection — can an assistant establish a reliable relationship with your services? We verify access and identity guidance, not just a displayed link.
- Understanding — can your content be interpreted without ambiguity? We check whether useful information can be read without relying entirely on layout.
- Access — can authorised assistants actually reach what you make available? Declared rules are compared with the response actually received.
What we measure
An audit makes 27 independent HTTP observations of the domain. Each observation answers one precise question — is an element present, readable, reachable or verifiable? — and keeps its result and context. A finding that is absent is therefore not confused with a page that could not be loaded.
- 14 counted observations contribute to the tier because they currently provide genuinely discriminating information.
- 12 diagnostic observations are measured and shown, but do not affect the tier: they add context without manufacturing easy points.
- 1 control observation makes a small adjustment when AI crawler access rules are explicit, in either direction.
From finding to tier
Weights are not chosen because a standard is fashionable. They reflect a signal's measured ability to distinguish sites from one another. We also favour breadth: four clues concentrated on one axis are worth less than coverage spread across several dimensions.
The four tiers
| Tier | What it means |
|---|---|
| Not ready | Your site remains difficult for an assistant to discover or use. |
| Almost ready | One first element is in place, but the experience is incomplete. |
| Agent Ready | Several dimensions are covered: the site is beginning to work for an assistant. |
| Fully optimised | The experience is coherent across all assessed dimensions. |
Why some findings remain diagnostic
- A very common element cannot separate sites: counting it would give almost everyone an artificial floor.
- An emerging element should not receive a weight before it has been observed often enough to be interpretable.
- An ambiguous match is not proof: when a name can belong to several actors, we show the signal without turning it into a certification.
A measurement, not a promise
The tier is calculated from repeatable observations that can be compared across sites. It describes a technical surface at a point in time, not the quality of your business, and does not guarantee that an agent will complete every business journey. Commerce and performance are therefore measured in separate pillars.
The limits we accept
- An audit is still a snapshot: the measurement point, site availability and changing standards can affect the result.
- Few sites currently reach the top; we do not claim to derive a universal law from them.
- Thresholds and weights will be reassessed on a new panel as adoption of these conventions grows.
Method v1 is the current comparison baseline. Every important change should be documented, tested on real data and dated — so that a tier change remains explainable.