Agent Readiness Audit

elmo lab audit checks a website the way AI crawlers and agents see it, then lists what to fix. It runs on its own: you don't need an Aeonav instance or Docker.

Pass every other domain your brand uses with --domain:

Each domain is resolved first, then handled by where it lands:

The site you pass first is always audited wherever it lands, even on a domain you didn't list.

The command exits with status 1 when no site could be audited. To gate a CI job on the result, pass --fail-under, for example --fail-under 80.

Pages are sampled from your sitemap, one from each top-level section, largest sections first, falling back to the homepage's links. Legal, careers, and translated pages are only sampled when nothing else is left. If your homepage answers on one host but every other page redirects to another (such as www), the audit follows your pages there. The sample is the same on every run.

Reach: Can Agents Get In? robots.txt, rules for AI search crawlers, AI user agents, and AI training crawlers, Content Signals, whether bot protection blocks AI user agents, noindex, snippet restrictions (nosnippet, max-snippet), pages behind a sign-in, meta-refresh and JavaScript redirects, alternate-domain redirects, and real 404s.

Read: Can Agents Read It? Text and headings without JavaScript, page size in tokens, markdown via Accept: text/markdown and at .md URLs, markdown alternate links, markdown frontmatter and code fences, llms.txt and its links, llms-full.txt, a section llms.txt for docs, the sitemap and its lastmod dates, discovery Link headers, and an agent view at ?mode=agent.

Understand: Can Agents Make Sense of It? Organization structured data and its sameAs links, page-level structured data, titles, descriptions, canonical links and Open Graph tags, publish and update dates, quotable passages of 40–200 words and named, linked authors in articles (docs, marked up as TechArticle, aren't held to these two), About, Contact, and Privacy pages, a pricing page, document structure, accessible names on controls, hidden text addressed to AI, and NLWeb schema feeds.

Checks marked i are optional improvements and never count as failures. Checks marked - weren't run, usually because the request budget ran out.

Each site gets a score out of 100: a pass counts fully, a warning half, and a failure nothing. Optional (i) and unchecked (-) results are left out, so a site isn't penalized for checks that don't apply to it. The exception is a site that blocks the audit, through robots.txt or by refusing requests: agents hit the same wall, so its unchecked results count as failures. With more than one site, the overall score is their average.

The audit identifies itself as AeonavBot/1.0 and links to this page. It is built to be cheap for the sites it checks:

To stop AeonavBot after its first homepage request:

To check whether AI user agents get through bot protection, the audit fetches the homepage once each with the user agents of ChatGPT-User, OAI-SearchBot, Claude-User, and PerplexityBot. Sites that verify those bots by IP address may treat the real bots differently, so that result is approximate.