Is your site ready for AI agents?

What is agent readability?

Agent readability is how easily AI agents — ChatGPT, Claude, Perplexity, Google Gemini, coding assistants, MCP clients — can discover, parse, and act on a website. It spans three surfaces: discovery files (llms.txt, robots.txt, sitemaps), structural signals (semantic headings, canonical links, structured data, markdown mirrors), and protocol manifests (MCP Server Cards, A2A Agent Cards, agents.json, agent-permissions.json).

Why does AI agent readability matter for SEO?

AI agents crawl what loads cleanly and cite what parses correctly. The incentives are sharp: a July 2025 Pew Research study found users who encounter a Google AI Overview click on a source link only about 8% of the time — roughly half the rate of searches without an AI summary. Princeton’s GEO study (KDD 2024) measured that adding source citations to a page lifted its inclusion in AI answers by roughly 40%, with statistics and quotations close behind. Sites that score well get summarised accurately and referred qualified traffic; sites that score poorly get paraphrased (badly) or skipped entirely. Unlike traditional SEO, you don’t need to rank on page 1 — structured, citable content gets pulled even when organic rank is low.

What does the agent readability scanner check?

• Vercel Agent Readability Spec — 15 site-wide checks (llms.txt, robots.txt, sitemap.xml, sitemap.md, AGENTS.md, HTTPS, OpenAPI) plus 23 per-page checks (meta tags, JSON-LD, headings, markdown mirrors, content negotiation, code-block language tags, JS-rendering dependency).
• llmstxt.org — 10 checks against the llms.txt format (H1 present, blockquote summary, H2 sections, link format, content-type, llms-full.txt).
• Agent protocols — 15 checks covering MCP Server Cards (SEP-2127 / RFC 9728 OAuth Protected Resource metadata), A2A Agent Cards (a2a.proto v1.0.0), Wildcard agents.json, agent-permissions.json, UCP (Universal Commerce Protocol), x402 (HTTP 402 Payment Required), and NLWeb (natural-language /ask endpoint).
• Accessibility — 23 WCAG 2.2 and layout-stability checks over your homepage (image text alternatives, form labels, control names, ARIA validity, iframe titles, page title, language tag, landmarks, heading order, tab order, zoom, tables, lists, autofill tokens, captions, explicit media dimensions). The accessibility tree is what assistive tech and AI agents parse to act on a page, so it’s reported as a separate accessibility sub-score — WCAG is not the Vercel spec, so it never moves your agent-readability score.

How is the agent readability score calculated?

score = round((passed checks / total checks) × 100). The denominator compounds: 15 site-wide + (23 per-page × number of pages scanned). A systemic issue like a missing canonical link on every page compounds significantly. Ratings: 90-100 Excellent, 70-89 Good, 50-69 Fair, 0-49 Needs Improvement.

Why choose Agent Ready over an SEO scanner or manual audit?

Agent Ready is built specifically for AI-agent readability — not a human-search SEO tool with an “AI” tab bolted on. It is the only scanner that validates llms.txt, the full Vercel Agent Readability Spec, and every agent-protocol manifest (MCP, A2A, agents.json, agent-permissions.json, UCP, x402, NLWeb) in a single pass.

• vs general SEO crawlers (Lighthouse, Screaming Frog) — they optimise pages for human search engines and never check the agent-protocol surfaces AI agents read.
• vs manual audits — all 70 checks run in seconds, every deploy, instead of hand-verifying five specs by hand.
• vs single-spec llms.txt validators — those lint one file; Agent Ready covers the other ~50 conditions too, with a plain-English fix for each failure.

See the full breakdown: Agent Ready vs the alternatives.