"AI-visible" is the emptiest phrase in this industry right now. Everyone selling GEO uses it. Almost nobody says what would have to be true for it to be false.
Here is ours, and it is deliberately boring: six named crawlers have to be fetching the site, confirmed in server logs, before we will describe it as reachable by answer engines. Not robots.txt permitting them. Not a plugin claiming it. Fetched, with timestamps.
The six
Googlebot and Bingbot are the classical index, and they are on the list because AI Overviews and Copilot both draw from those indexes rather than from some separate AI corpus. GPTBot and OAI-SearchBot are OpenAI's two paths — the first trains and grounds ChatGPT, the second powers its search citations, and a site can be visible to one and not the other. ClaudeBot is Anthropic's. PerplexityBot is Perplexity's, and it matters disproportionately because Perplexity is one of the few answer surfaces that still sends a visible referral click back.
Six user agents. A log grep. It takes about four minutes.
Why it catches things
The check is cheap and it fails more often than you would expect. A robots.txt that allows everything still leaves plenty of ways to be unreachable: a Cloudflare bot-fight rule that does not distinguish GPTBot from a scraper, a WAF that rate-limits anything without a browser fingerprint, a staging password left on a subdomain, a CDN serving a JavaScript shell to non-browser clients.
None of that shows up in a crawl audit tool. All of it shows up in the logs as a user agent that requested a page and got a 403.
On the Osoba Renos build, all six are now fetching, and that is stated on the case study as a verified fact rather than an aspiration. On Valve Appliance Repair, the same structural work — coverage across all seven appliance categories, emergency intent phrased the way people actually phrase it, trust signals made machine-readable — ran alongside impressions up 631% and clicks up 331% in the most recent Search Console period against the one before it, from a standing start seven weeks earlier. Semrush puts organic traffic up 98% over its most recent window.
What the check does not prove
Being fetched is necessary. It is nowhere near sufficient, and we say so on every call where someone hopes otherwise.
A crawler fetching your page tells you the door is open. It tells you nothing about whether a model will name you when someone asks which contractor to hire, which depends on corroboration you do not control, on how resolvable your entity is, and on how quotable the passage is once retrieval reaches it. Plenty of thoroughly-crawled sites are never cited by anything.
So the six-crawler check is a floor, not a result. It is the thing we run first because it is the cheapest way to find out that everything downstream was going to be wasted.