Agent readiness levels explained: the five-level rubric
How the self-check turns your answers into a level, and what each level suggests you do next.
Am I Ready for Agents? editors · Reviewed September 2026
There are five levels: Hidden, Found, Readable, Actionable and Prepared. Each answer scores 2, 1 or 0 points, the five areas count equally, and your weakest area limits how high you can go. The rubric is our editorial judgement, published in full so you can score yourself by hand.
What are the five levels?
Level 1: Hidden
Overall 0–24
Agents are likely to have trouble finding or reading you at all.
What to do next
Start with the find and read questions. Ask your web team what your robots.txt and bot protection allow today, and whether key pages show their text without JavaScript.
Level 2: Found
Overall 25–44
Agents can probably reach you, but much of what they need is missing or unclear.
What to do next
Agents can get in. Make sure what they find is readable: prices, availability and policies as text, and structured data on product, event or article pages.
Level 3: Readable
Overall 45–64
Agents can read most of what matters, but completing a task is uncertain.
What to do next
Look at the act and pay questions. Walk through your main task as a guest and note every step a person has to rescue.
Level 4: Actionable
Overall 65–84
Agents can read you and complete some tasks; a few gaps remain.
What to do next
Decide how you treat agents that act for customers: verification, fraud rules and who owns the policy.
Level 5: Prepared
Overall 85–100
By your own answers, the foundations are in place. Confirm them with a scanner and your web team.
What to do next
Confirm your answers with an automated scanner and your own analytics, then revisit the self-check when standards change. This site is reviewed monthly.
How is the score calculated?
- Each answer scores points: yes 2, not sure 1, no 0.
- Area score: the area's points divided by the maximum for that area (2 points times its number of questions), as a percentage. Find, read and act have four questions each (maximum 8 points); pay and trust have three each (maximum 6 points).
- Overall score: the average of the five area scores. Each area counts for 20%, whatever its number of questions.
- Level: the overall score, rounded to a whole number, falls in one band: Hidden 0–24, Found 25–44, Readable 45–64, Actionable 65–84, Prepared 85–100.
- Weakest-area rule: your level can be at most one above the level your lowest area score would get on its own.
| Level | Name | Overall range |
|---|---|---|
| 1 | Hidden | 0–24 |
| 2 | Found | 25–44 |
| 3 | Readable | 45–64 |
| 4 | Actionable | 65–84 |
| 5 | Prepared | 85–100 |
Why does not sure score 1 point?
Not sure often means part of the work exists but nobody has confirmed it. A point recognises that, and the result lists every not sure answer as a question for your web team, so it is never hidden.
Why does the weakest area cap the level?
An agent's task only succeeds if every step works. A site that is easy to find and read but cannot take a booking or a payment will still lose the task at the last step. The cap stops a strong score in three areas from hiding a gap in a fourth.
Worked example
| Area | Answers | Points | Score |
|---|---|---|---|
| Find | 4 yes | 8 of 8 | 100% |
| Read | 2 yes, 2 not sure | 6 of 8 | 75% |
| Act | 1 yes, 1 not sure, 2 no | 3 of 8 | 38% |
| Pay | 1 not sure, 2 no | 1 of 6 | 17% |
| Trust | 1 yes, 2 not sure | 4 of 6 | 67% |
The overall score is (100 + 75 + 37.5 + 16.7 + 66.7) ÷ 5 = 59, which falls in the Readable band (Level 3). The lowest area, pay at 17%, would be Level 1 on its own, so the weakest-area rule caps the result at Level 2: Found. The first thing to take to the web team is the pay area.
How do these levels relate to scanner scores?
They do not map onto each other. Scanners test public pages automatically and use their own scales. ora.ai, for example, publishes grades from A+ (95–100, Leading) to F (0–27, Unusable) computed from automated checks. Our five levels score self-reported answers with an editorial rubric, so a Level 4 here is not equivalent to any grade elsewhere.
Source: ora.ai methodology · Read September 2026
What does the rubric not measure?
- Whether agents actually visit your site, or how often.
- Search visibility in AI answers.
- Traffic, conversion or revenue.
- Anything behind a login.
- Whether your answers are correct: only a check of the site itself can confirm that.