Social preview image
og-image.png · PNG · 1200×630
For social posts, link cards, and link unfurls.
Download →Agent readability for the web. Agents (ChatGPT, Claude, Copilot, Cursor) now read most of the web before humans do. a14y is an open spec for making any website discoverable, parseable, and comprehensible to those agents, a versioned scorecard that operationalizes the spec, and a CLI plus Chrome extension that score any site against it.
a14y scoring its own website. Captured from the open-source CLI; you can reproduce this in your terminal in under a minute.
Agent Readability Score: 92/100
scorecard 0.2.0 (released 2026-04-06) — mode page
33 pass · 3 fail · 0 warn · 0 error · 2 n/a (out of 38)
Site checks
Discoverability
✓ llms-txt.exists — Found llms.txt at https://a14y.dev/llms.txt
✓ llms-txt.content-type — text/plain; charset=utf-8
✓ llms-txt.non-empty
✓ llms-txt.md-extensions — 84 links checked
✓ robots-txt.exists — https://a14y.dev/robots.txt
✓ robots-txt.allows-ai-bots — Allowed: GPTBot, ClaudeBot, CCBot, Google-Extended
✓ robots-txt.allows-llms-txt
✓ sitemap-xml.exists — https://a14y.dev/sitemap.xml
✓ sitemap-xml.valid — urlset
✓ sitemap-xml.has-lastmod — 85 entries with lastmod
✓ sitemap-md.exists — https://a14y.dev/sitemap.md
✓ sitemap-md.has-structure — 8 headings, 85 links
✓ agents-md.exists — https://a14y.dev/AGENTS.md
✓ agents-md.has-min-sections — Found: installation, configuration, usage
Page checks — https://a14y.dev/
HTTP
✓ http.status-200 — 200
✓ http.redirect-chain — 0 hops
✓ http.content-type-html — text/html; charset=utf-8
✓ http.no-noindex-noai — no x-robots-tag
HTML metadata
✓ html.canonical-link — https://a14y.dev/
✓ html.meta-description — 195 chars
✓ html.og-title
✓ html.og-description
✓ html.lang-attribute — en
Structured data
✓ html.json-ld — 1 block(s)
✓ html.json-ld.date-modified — 2026-05-15T16:17:57.474Z
✓ html.json-ld.breadcrumb
Content structure
✓ html.headings — 10 headings
✗ html.text-ratio — 13.4%
✓ html.glossary-link
Markdown mirror
✓ markdown.mirror-suffix — https://a14y.dev/index.md
✓ markdown.alternate-link — https://a14y.dev/index.md
✓ markdown.frontmatter
✗ markdown.canonical-header — no Link header
✗ markdown.content-negotiation — text/html; charset=utf-8
✓ markdown.sitemap-section
Code
✓ code.language-tags — 2 blocks
API
- api.schema-link — not an API page
Discoverability
- discovery.indexed — no site index available
! Single-page review: only https://a14y.dev/ was audited.
For a full-site audit (crawls every reachable page) run:
a14y https://a14y.dev/ --mode site The longer description, written for journalists who need the full picture in a single read.
The problem. AI agents (ChatGPT, Claude, Copilot, Cursor, and the 50-plus coding assistants behind them) now read most of the web before humans do. Yet the web was built for human eyes: layout-driven HTML, JavaScript-rendered widgets, cookie walls, and broken markdown all degrade how well an agent can fetch a page, understand it, and act on it. There is no agreed-upon way to measure how readable a website is to those agents, and no public scoreboard pushing the industry to improve.
The spec. a14y (shorthand for agentreadability) is an open spec for making any website discoverable, parseable, and comprehensible to AI agents. The spec is versioned, every check is documented with the why and the how to fix, and the whole thing is licensed Apache-2.0 and lives on GitHub. Documentation sites are the high-value first target, but the scorecard works for marketing sites, product pages, help centers, and anything else agents might read.
The scorecard. Each spec version pins a fixed set of checks across three categories (Discoverability, Parsing, and Comprehension), so a score comparison across two sites or two points in time is always apples-to-apples. The current version (v0.2.0) ships 38 checks; the scorecard surface at a14y.dev/scorecards/ shows every one of them with detection rules and remediation guidance.
The tools. Three surfaces, one engine: a free CLI (npx a14y your-site.com), a free Chrome extension,
and a coding-agent skill (npx skills add timothyjordan/a14y) that
drives the audit-and-fix loop inside Claude Code, Codex, Cursor, and
any other spec-compliant agent. All three produce the same score for
the same URL and scorecard version. The CLI also emits a Markdown
fix-list (--output agent-prompt) that any agent can ingest
directly.
The proof. The public leaderboard at a14y.dev/leaderboard/ scores a cross-section of major websites against the current scorecard, with per-site scorecards, a score histogram, and category breakdowns. It updates as the dataset is refreshed, and every score on the leaderboard is reproducible by running the open CLI against the same URL.
The frame. YSlow made performance visible; Lighthouse made performance a checklist; a14y is the same kind of free, opinionated, embarrassing-in-a-good-way scorecard for the agent-readable web. Open implementation and an open dataset mean nobody owns the number. Anyone can verify, dispute, or extend it.
238 sites scored against scorecard v0.2.0. Top 10 below; the full ranking, score histogram, and per-site scorecards are at a14y.dev/leaderboard/.
| # | Site | Category | Score | Top failures |
|---|---|---|---|---|
| 1 | PostHog Docsposthog.com | Docs platform | 97 | api.schema-linkhtml.canonical-link +3 |
| 2 | Vercelvercel.com | Dev tool | 78 | html.glossary-linkhtml.json-ld +3 |
| 3 | Vercel Docsvercel.com | Docs platform | 78 | html.glossary-linkhtml.json-ld +3 |
| 4 | OpenAI Platform Docsplatform.openai.com | Docs platform | 76 | api.schema-linkhtml.canonical-link +3 |
| 5 | Latticelattice.com | Enterprise SaaS | 74 | html.glossary-linkmarkdown.alternate-link +3 |
| 6 | Leverlever.co | Enterprise SaaS | 74 | html.glossary-linkmarkdown.alternate-link +3 |
| 7 | Next.jsnextjs.org | Docs platform | 74 | markdown.canonical-headermarkdown.frontmatter +3 |
| 8 | Railway Docsdocs.railway.com | Docs platform | 74 | html.glossary-linkmarkdown.alternate-link +3 |
| 9 | Nuxtnuxt.com | Docs platform | 73 | html.glossary-linkhtml.json-ld.date-modified +3 |
| 10 | Typeformtypeform.com | Enterprise SaaS | 73 | markdown.alternate-linkmarkdown.content-negotiation +3 |
Logos as SVG (light + dark variants for any background) plus a social preview image for posts and link cards. Right-click any preview to save, or use the explicit download links.
For press inquiries, podcast bookings, or interview requests:
Typical reply time: under 24 hours on weekdays. a14y was created by Timothy Jordan.