Asking an assistant "can you see my website?" is not a test. The answer depends on whether that session browsed the web or answered from memory, and both happen. What you want is evidence that a crawler can reach the page and get something usable back. That is measurable.
1. Read your own robots.txt
Open yourdomain.com/robots.txt in a browser. Search for the names that matter: GPTBot,
OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended.
A block looks like this — and it means the assistant never sees the page, no matter how good it is:
User-agent: GPTBotfollowed byDisallow: /- or a wildcard:
User-agent: *withDisallow: /, which blocks every crawler including the AI ones
2. Fetch your homepage as the crawler
If you can run a terminal command, this is the honest test — it asks for the page the way GPTBot would:
curl -A "GPTBot" -I https://yourdomain.com
A 200 means you were served. A 403 or 503 means something in front of your site
— a firewall, a bot rule, a security plugin — is turning crawlers away even though robots.txt allows them.
That case is the one owners never find by reading files, because the file says "allowed".
3. Check for a map
Open yourdomain.com/llms.txt. If it 404s, an assistant crawl has to guess which of your pages
matter. It will guess wrong: usually the homepage, sometimes a blog post from 2019.
4. Ask the question your customer would ask
Not "do you know my site". Ask what a buyer asks: "who does [your service] in [your area]". Then read the answer for names.
Three outcomes, and each means something different:
- You are named — you are readable and being used. Work on staying current.
- Competitors are named, you are not — the most common result, and the fixable one. This is a readability problem, not a quality problem.
- Nobody local is named — the category may be too thin for the model to answer. Try the question with the city included.
5. Ask the same question twice, a week apart
Answers move. One mention is a snapshot; the useful number is whether it happens again. This is also the reason we built the scan to store results over time instead of showing a single score.
What this test cannot tell you: how often you are cited across all conversations. No tool can see inside someone else's chat. Treat any product claiming otherwise with suspicion — including ours.