Framework
The Maturity Model
Six levels describe how far an organization has come with agents, from invisible to trusted. Each domain has its own level, because an organization is rarely at the same level in all six.
The six levels
- 1
Closed Level 0
Agents are blocked or can’t make sense of you, so they leave you out of their answers.
- 2
Visible Level 1
Agents can reach your public pages and find you in the places they look, but they guess at the details.
- 3
Described Level 2
Your facts, offers and terms are published in forms agents read reliably, and they match what a person sees.
- 4
Conversational Level 3
Agents can ask you questions and get accurate answers, quickly, with a person behind the answers that need one.
- 5
Callable Level 4
Agents can do things with you: book, buy, apply or request a quote, through actions you define.
- 6
Trusted Level 5
Verified both ways, measured end to end and improved continuously. Agents prefer you because dealing with you reliably works.
Levels are cumulative. To be at level 3 in a domain, you meet the criteria for levels 1 and 2 as well. Levels describe what is true today, not what is planned.
Domains against levels
The short criteria for each cell. The full description is on each domain page.
| Domain | 0. Closed | 1. Visible | 2. Described | 3. Conversational | 4. Callable | 5. Trusted |
|---|---|---|---|---|---|---|
| Discover | Agents are blocked, often by accident, or cannot load your pages at all. | Public pages are reachable. Access rules exist but nobody has reviewed them for agents. | A reviewed access policy and a current sitemap. Key sources outside your site are listed and correct. | An agent guide such as llms.txt and plain-text versions of key pages. Your services are listed in agent registries. | Discovery runs as a routine: access reviewed quarterly, listings checked monthly, unintended blocks alerted within a day. | Agents find you first in your category, and you adjust access policy from evidence about which agents bring outcomes. |
| Describe | Facts live in scattered documents, images and PDFs. Agents contradict each other about you. | Facts are on the site as text, but vary from page to page and channel to channel. | A fact register feeds the site. Key pages carry validated schema.org markup, and offers are in a feed. | Documentation, catalogs and policies are all machine-readable and checked for consistency across channels. | Descriptions are produced from the source of truth by automation and tested on every release. | Agents describe you correctly in sampled checks, and a mistaken description is corrected within days. |
| Converse | No way to ask you anything except a contact form no agent can use. | Agents can find answers on pages but cannot ask. Contact details are public. | FAQs are structured and current. Common questions have a documented answer. | A question endpoint answers from approved sources, with live data and a route to a person. | Agents hold multi-turn conversations with you, in scope and with logs. Handoffs are timed and staffed. | Answers are verified as accurate by sampling, and failed questions are reviewed weekly and turned into improvements. |
| Transact | Every task needs a person on a website. Agents hand people back. | Agents can read how to do things but cannot do them. Anything transactional is manual. | Requests such as quotes or bookings can be submitted in a structured way, but handled by hand. | At least one end-to-end action is available to agents, with limits, and orders land in the normal systems. | The main tasks are callable, in open protocols, with change, cancel and dispute handling. | Agents choose you for action because completion is reliable. Problem rates match or beat those of human customers. |
| Trust | No policy. Agents are blocked or allowed by habit, and nobody can say which is which. | Basic rate limits and bot protection exist. Agent traffic is treated like any other automated traffic. | A written agent policy exists, and legal and the business have approved it. | Agent actions require scoped, revocable permission. Every action is logged to an identity. | Verified identity is checked on requests. Provenance and disclosure are routine, and incidents are rehearsed. | Trust runs both ways: you verify agents and agents can verify you. Reviews and audits are routine. |
| Measure | No agent traffic is separated from other traffic. Nobody knows what agents do. | Agent visits are counted from logs. Nothing is reported beyond that. | Visits, questions and the first failures are counted with documented definitions. Accuracy is tested occasionally. | A monthly report covers the core metrics, including failed tasks and accuracy from a regular sample. | Outcomes are attributed to agents and share of recommendations is tracked on a fixed panel. Executives see a quarterly summary. | Metrics steer investment, definitions are versioned and the figures are benchmarked against peers. |
Self-assessment
Twelve yes or no questions. Answer for the whole organization. Nothing is stored or sent: your answers stay on this page.
1. Do you know which AI crawlers and agents your website allows and which it blocks, and was that a deliberate choice?
2. If you ask a general-purpose assistant about your organization, does it name you and describe you roughly correctly?
3. Are your key facts (names, locations, hours, products, prices, policies) kept in one place that someone is responsible for keeping current?
4. Does your website publish machine-readable data about those facts, such as schema.org markup, a sitemap or a product feed?
5. Has someone checked what agents say about you against what you want said, in the last six months?
6. Can an agent get an answer to a common question (stock, opening hours, eligibility) without a person reading it?
7. Is there a named team, and a published route to a person, for questions an agent cannot settle on its own?
8. Can an agent complete at least one real task with you, such as a booking, a purchase, an application or a quote request?
9. Do you have written rules for what agents may and may not do with you, approved by legal or risk?
10. Can you tell which agents are dealing with you, by verified identity rather than a guess from the traffic?
11. Do you report agent visits, questions answered, failed tasks and accuracy to leadership every month?
12. Can you name outcomes (sales, bookings, enquiries) that came through agents, and a person accountable for what agents commit you to?
0 of 12 answered. Answer every question to see an indicative level.