Real answers, not just setup
Other scores stop at technical setup. The open-source scanner, run with your own API key, lets an agent try to answer real questions from your pages and shows the full transcript, with the reason behind every miss.
Agent Readiness Score
Can AI agents find, read and answer questions from your docs? Get a free score for any documentation site, help center or knowledge base in about a minute.
How the score works
See the evidence behind each part of the score, from whether agents can reach your content to whether they can use it correctly.
01Access
Can an agent find and read the content?
- The 23 AFDocs checks: llms.txt, Markdown pages, page size, structure, real 404s, login walls
- Whether search crawlers and the AI assistants that honour robots.txt may read the docs
- Whether the sitemap lists the pages we found on our own
Give agents a direct path to your docs.
The report shows which discovery and reading checks pass, and what to change when they fail.
02Freshness
Is the content maintained and consistent with its sources?
- Links and section anchors on the pages still work
- Pages say when they were last changed, and sitemap entries still exist
- Endpoints in the docs exist in the API spec, and deprecated ones are marked
A working page can still have a broken link.
Findings identify the problem and give your team a concrete fix, including links to missing sections.
03Answerability
Can an agent answer real questions from it?
- Questions a real user would ask, written from your own pages
- An agent answers using only what it can fetch, with no JavaScript
- Each answer checked for reaching the right page, being correct, and citing its source
Open-source scanner only, with your own API key
Read the answer, its sources and the verdict.
Available in the open-source scanner with your own model API key. Not included in the hosted web scan.
Grading and scan coverage
Every check is weighted by how much it matters to an agent, on the same scale as the open AFDocs standard, and the score is the share a site earns out of 100. A check that could not run on a site is left out, never counted as a failure.
Grades: A+ from 97, A from 90, B from 80, C from 70, D from 60, F below. AFDocs’ own caps apply: a site with no llms.txt cannot score above 59. The web scan grades Access and Freshness; Answerability is added when you run the open-source scanner with your own API key.
Top scores
The best-scoring sites we have scanned so far, updated as scans settle.
| # | Site | Platform | Score |
|---|---|---|---|
| 01 | Amplifique.me ajuda.amplifique.me | 100A+ | |
| 02 | WP Travel Engine docs.wptravelengine.com | 100A+ | |
| 03 | Documentation.AI documentation.ai/docs | 100A+ | |
| 04 | Scrut help.scrut.io | 100A+ | |
| 05 | WAVjunkie docs.wavjunkie.com | 99A+ | |
| 06 | Shiprocket SOP shiprocket.documentationai.com | 98A+ | |
| 07 | Expo docs.expo.dev | 93A | |
| 08 | Mintlify mintlify.com/docs | 93A | |
| 09 | Documentation.AI documentation.ai | 89B | |
| 10 | Paperguide helpcenter.paperguide.ai | 87B |
What other agent scores miss
Help centers count too
Zendesk, Intercom, Freshdesk, Help Scout, Document360 and Confluence are scored on the same table as developer docs, so the numbers compare.
Freshness, not just format
Broken links and anchors, deleted pages the sitemap still lists, and endpoints your API spec does not have all lower the score.
A prompt your coding agent can run
Every report ends in a paste-ready brief for Claude Code, Cursor or Copilot, with a way to confirm each fix before you rescan.
Questions
How the score is made, and what it costs you: nothing.
What is the Agent Readiness Score?
A free 0 to 100 score for how well AI agents can find, read and use your documentation, help center or knowledge base. The web scan grades Access and Freshness; the open-source scanner adds Answerability.
How is it different from other agent scores?
Other tools only check technical setup, such as whether you have an llms.txt file. We run those checks too, using the open-source AFDocs checks, and then check whether the content is maintained: working links and anchors, live sitemap entries, endpoints that exist in your API spec. We also score help centers and knowledge bases, not just developer docs.
What is Answerability, and why is it not in my web score?
Answerability lets an AI agent try to answer questions a real user would ask, written from your own pages, using only what an agent can fetch. Every answer is checked against the sentence it came from. It needs an AI model, so it runs in the open-source scanner with your own API key rather than in the free web scan.
Do I need to sign up or give an email?
No. Enter a URL and the scan starts straight away. Every result, check and transcript is public. You can ask us to email the report if you want a copy.
What does "provisional" mean on a score?
That the scan could not stand behind the number: it read fewer than five pages, the site refused some of our requests, or too many checks could not be verified. The score still shows, but it stays off the leaderboard until a clean scan.
How long does a scan take?
Usually under a minute. Large sites can take a few minutes, and the page opens the report the moment it is ready.
How often can I rescan?
Results are saved for seven days, so repeat visits load instantly. Use "Scan again" to rescan a site, at most once an hour.
How do I stop my site from being scanned?
Our scanner identifies itself as DocumentationAI-AgentScore and follows robots.txt. Add "User-agent: DocumentationAI-AgentScore" with "Disallow: /" to opt out, or email us to remove a published report.
Agent-ready docs by default
Documentation.AI sites ship with llms.txt, Markdown pages, a sitemap and an MCP server, and our agent keeps the content current.
Live in under 5 min · No credit card required