Questions? Answered.

AI search assistants (ChatGPT, Perplexity, Claude) increasingly answer questions instead of sending search results. If their crawlers are blocked — often accidentally, by a single robots.txt line — your content is invisible in those answers and you lose that referral traffic.

Common causes: a blanket Disallow: / added for staging that leaked to production, security plugins that block unknown user agents, CDN bot-fight rules toggled on by default, or copying a robots.txt template meant to stop training bots but which also blocks the search-class crawlers that send referrals back.

llms.txt is an emerging standard (like sitemap.xml was in 2006): a markdown file at /llms.txt that maps your most important pages for AI systems and coding agents. Adoption grew 8.8x in the last year. It does not replace robots.txt — it complements it.

No. Every finding is deterministic: we fetch your robots.txt, llms.txt, sitemap, and pages with our verifier user-agent (and a probing-as label for the crawler being tested), then apply fixed rules. The same site produces the same report every time. That also means zero AI API costs, which keeps pricing low.

We fetch a page twice — once as a normal browser and once with our verifier probing as GPTBot or ClaudeBot. We never impersonate those operators. If the browser sees hundreds of words but the crawler-shaped request sees almost nothing, your page probably renders client-side only, meaning many AI systems read an empty shell.

No. Blocking GPTBot or other training-class crawlers is a legitimate business decision. Our score only penalizes blocks on search-class crawlers (the ones that send referrals) and agent-class crawlers acting on behalf of real users.

Attacking AI agents recon public origins for leaked API keys, unauthenticated JSON APIs, and MCP files. Visiting AI browsers can also be steered by hidden page text (the EchoLeak / Atlas class). Avuril fetches the same public URLs those agents would, then records kinds and prefixes — never secret values, and never exploit payloads. A high agent-surface finding is evidence of a public configuration issue, not a penetration test.

Yes — a surface layer only, from the same public HTTP fetch. We check security headers (HSTS, CSP, nosniff, clickjacking, Referrer-Policy), cookie flags (name only, never values), mixed content, third-party scripts without SRI, security.txt, robots.txt that advertise secret paths, and a short allowlist of well-known public files such as /.env and /.git/HEAD. There is no AI, no paid scanner, and no exploit payload. This is not a penetration test and a high score is not a security certification.

No. Avuril only fetches public URLs. There are no scripts, plugins, middleware, or DNS changes required.

Yes — that's who we built the Agency tier for. Monitor up to 25 client domains from one dashboard and share the readiness reports with clients as your own audit artifact.