Find out what your docs actually score.
Twenty-three criteria, seven groups, every one scored 0–3 — against a rubric we publish rather than hide. Free on any public corpus, including your competitors'.
Seven groups, published in full
This is rubric v1. It is our opinion, argued for in the open — if you think a criterion is wrong, we would rather have the argument than the benefit of the doubt.
Structure & Diátaxis
A + B = 40% (substance)
- Coverage of all four Diátaxis quadrants — tutorial, how-to, reference, explanation
- Mode purity: pages that don't try to be a tutorial and a reference at once
- A golden-path quickstart a newcomer can finish in under fifteen minutes
- Task coverage mapped to the jobs your users actually came to do
Reference quality
A + B = 40% (substance)
- Reference generated from the spec and kept in sync (OpenAPI or equivalent)
- Per-language snippets, a try-it surface, and sandbox keys that work
- A complete error catalogue — every code, what causes it, what to do
- Versioning and a changelog with migration notes, not just release dates
Deep-linkability
C + D = 25% (trust)
- Stable, human-readable anchors — with redirects when they change
- Coherent URL hierarchy, breadcrumbs and a sitemap that matches it
- Sections that are self-contained enough to quote without the surrounding page
Freshness
C + D = 25% (trust)
- Last-updated derived from version control, against a stated staleness SLA
- Automated link-rot measurement rather than an annual manual sweep
- Style-as-code in CI — linting and readability checks that block a merge
Search visibility
E = 10% (a shrinking channel)
- Canonicals, server-rendered or statically generated pages, Core Web Vitals passing
- Structured data limited to markup search engines still support
- hreflang correctness where you publish more than one language
AI-answer readiness
F + G = 25% (a growing channel)
- Citation, quote and statistic density in the places answers get lifted from
- Answer-shaped headings, with nothing important hidden inside tabs or accordions
- A deliberate AI-crawler policy — a decision you made, not a default you inherited
RAG & agent access
F + G = 25% (a growing channel)
- A .md mirror per page, plus llms.txt and llms-full.txt
- An MCP endpoint that returns cited chunks rather than a blob of prose
- Token efficiency — how much cheaper your markdown is to read than your HTML
Scoring is 0–3 per criterion. The weight bands above are how the rubric groups them — substance first, then whether the corpus can be trusted to be current, then the two distribution channels. We'd rather show you the working than round the number — and we version the rubric in public when it moves.
What lands in your inbox
A scored corpus
A 0–100 score with the group breakdown, so you can see whether you have a substance problem, a trust problem or a distribution problem.
Per-page defects
Which pages fail which criteria, with the specific defect named — not a generic recommendation to improve your documentation.
A prioritised fix queue
Ordered by weight and effort, so the first ten fixes are the ten that move the number. Useful on its own, whether or not you work with us.
Give us one URL
We'll crawl outward from there — sitemap, reference, changelog, whatever is publicly reachable — and score the corpus against the rubric above.
Public materials only
We score what a reader — or an agent — can reach without a login: public docs, public APIs, public academies, public MCP endpoints. We never crawl gated content, and we never ask you for credentials.
Batched, not instant
The audit runs on a queue and a human reads the report before it goes out, so expect a reply rather than a number appearing on screen. Rate-limited per domain.
A rubric, not a law
No industry-standard docs score exists — we checked. This is our rubric, v1, published openly so you can argue with it. Criteria get merged and split as we calibrate, and we'll say so when they do.
Judgement, not just crawling
Some criteria are measurable (link rot, Core Web Vitals, token ratio). Others need a person to read the page and decide whether a tutorial is actually a tutorial. We tell you which is which in the report.
Prefer to read first? Tour the platform or see the whole loop.
Docs Score FAQ
What does the audit cost?
Nothing. The Docs Score on a public corpus is free and rate-limited. It's the front of our funnel and we'd rather you find it useful than gated — the paid work is the private internal-enablement assessment and the platform itself, and pricing for those isn't published yet. Talk to us.
Do I have to talk to sales to get the report?
No. Submit a URL and we send the report. If you want to talk about what to do with it, that's a separate conversation you have to start.
Will you crawl anything private?
No. Public materials only — anything behind a login, a paywall or an allowlist is out of scope, and we don't accept credentials to get around that. The company-level internal pillar is assessed only in a paid engagement, with your participation.
How is this different from an SEO audit?
Search visibility is one of the seven groups and carries ten percent. The other ninety percent is whether your documentation actually teaches, whether it can be trusted to be current, and whether an AI agent can read and cite it — which is where the traffic is going.
What happens to my data?
The URL and your email go to our inbox so we can send the report. We don't sell contact data, and you can ask us to delete the request at any time.
Does a low score mean you'll pitch me a rebuild?
It means you'll get a fix queue. Most of it is work your own team can do, and the report is written so they can. We only make money if you decide the loop — regenerate, compile, certify — is worth buying.
The score is the start of the loop
Ingest and score are stages one and two. If the report makes the case, stages three to seven turn the same corpus into a living academy with real exams and verifiable credentials.