AI Visibility Audit · inocta.io

Audit a website's AI-visibility and agent-readiness.

This extension reads the page you are on and the public files an AI crawler looks for on that domain (robots.txt, sitemap.xml, llms.txt), and tells you what an AI system can and cannot do with your site. It reports which AI crawlers your robots.txt allows, separating answer engines (blocking them costs you citations) from training crawlers (blocking them is a rights decision that costs little). It also detects whether your pages expose agent-usable tools via WebMCP or an agent card. Everything runs locally. Nothing is sent to inocta.io. If the site you are auditing declares a machine-readable file on another host (in its page head, its robots.txt or its Link header), the extension reads that one file, with your permission for that host. Beyond the standard file addresses, it guesses one address only: /sitemap_index.xml when a sitemap redirects. If you run a competitor comparison, the extension reads the same public files, the home page and the ARD manifest of each domain you enter, with your permission for each domain. Bilingual EN/FR. WHAT IT CHECKS Site fundamentals, fixed once for every page on the domain: robots.txt present and valid, sitemap.xml present and valid, no answer engine blocked, no training crawler blocked. Page content, reported as guidance rather than a pass mark: one H1, readable length, sub-headings, structured data, author and date. Agent readiness: the accessibility tree an agent actually reads (via axe-core, restricted to Google's 33-rule agentic set), llms.txt, WebMCP tool declarations, and the ARD manifest that tells an agent what it can use on the site. These checks are based on Lighthouse 13.5, and the report states where we differ from it. It also shows whether the site carries Chrome's WebMCP trial token, which decides whether an ordinary visitor's Chrome can see its tools. HOW IT REPORTS Counts of conditions we observed and can name. No single 0-100 quality score, and that is a decision rather than a gap: we score conformance to published practice and claim nothing about whether an AI will cite you. A check we could not run is excluded from the total and shown as unknown, never counted as a failure. Full report in a tab, a PDF for a client, and a JSON export that carries its own definitions and limits so it can be pasted into any AI assistant.