ChatGPT, Claude, Perplexity, and Google AI answer buying questions every day, and they can only recommend sites they can actually read. They fetch from datacenter IPs, skip JavaScript, and give up fast. This scan shows you your site the way the machines see it.
Reads about a dozen public URLs once. Three scans per minute per visitor. Nothing stored.
The web most people browse and the web AI systems fetch are not the same web. Three gaps account for most invisible sites, and none of them show up when you check your own site from your own browser.
AI crawlers and the scrapers behind them fetch from cloud IPs. Bot-fight settings and WAFs challenge that traffic wholesale, so the site looks fine from your laptop and blocked from everywhere retrieval actually happens.
If your pricing, product facts, and copy only exist after hydration, a scraper sees an empty shell. The answer engines quote a competitor whose facts live in raw HTML.
Fetch pipelines run on budgets. Slow origins, fake 404s that return 200, and missing machine files like llms.txt and sitemap.xml all burn the budget before your content gets read.
The whole thing runs from a Cloudflare datacenter, which matters: it is the same vantage point AI retrieval uses, and the one your own browser can never test from.
About a dozen public URLs, once, in three polite waves: your homepage as a browser, as GPTBot, and as ClaudeBot, plus robots.txt, llms.txt, sitemap.xml, a deliberately missing page, a markdown-negotiation probe, and your trust pages.
Reachability counts most, because nothing else matters if agents get blocked. Then server-rendered content, metadata, strict JSON-LD, robots.txt access, the machine layer, honest 404s, and speed.
Each [FAIL] row explains what broke and what to change, in plain language. Rerun the scan after each fix and watch the score move.
The score is a weighted 0 to 100. Reachability failures cost the most because they hide everything else.
Agents can reach and read you. Remaining points are refinements: richer schema, a linked llms.txt, markdown negotiation.
Visible but leaky. You show up, but better-structured competitors get the citations for questions you should own.
Something structural is hiding you: a bot wall, JavaScript-only content, or no machine layer at all. Fix these first; they gate everything.
How well a website can be fetched, read, and quoted by AI systems: crawlers like GPTBot and ClaudeBot, answer engines like ChatGPT and Perplexity, and the agents acting for their users. It covers reachability from datacenter IPs, server-rendered content, structured data, robots.txt access, and machine-readable files like llms.txt.
No. robots.txt is a request, and it only governs crawlers that read it. The blocks that actually hide sites usually live elsewhere: bot-fight and WAF settings that challenge every datacenter IP, origin throttles that answer crawlers with 429s, and JavaScript-only pages that render empty for scrapers. This scan tests the real fetch path, not the policy file.
No. A scan reads about a dozen public URLs once, the same pages any visitor or crawler fetches, and writes nothing. Scans are limited to three per minute per visitor, and nothing about you is stored.
80 or above means agents can reach and read the site. 50 to 79 means visible but leaking citations. Below 50 usually means a structural blocker worth fixing before anything else.