We shipped, then checked whether any AI assistant could actually see us. The answer was zero. The check cost one afternoon and I should have run it before building.

Reposting without the link, per automod. Happy to put it in a comment if anyone wants it.

Ride-along, warts included. We put a product live at the end of August. It works, the pages are up, Search Console is green. Last week I stopped assuming and measured whether an AI assistant would ever mention us. Posting the numbers because they were worse than I expected and the check is nearly free.

What I did

12 prompts through an assistant with web search turned on: 1 our exact brand name, 4 category queries, 7 questions a buyer in our space would actually type. Recorded every domain it cited. Then grepped 30 days of access logs for the AI crawler user-agents.

What came back

• Cited us: 0 of 12.
• On our own brand name it cited a .app version of our domain. Not ours. It 404s. Nobody maintains it.
• On the buyer questions the cited sources were OpenAI's own help centre (5) and Google's support site (2). Platform help pages, not competitors.
• GPTBot in 30 days: 0 fetches. OAI-SearchBot: 0. ChatGPT-User: 0. ClaudeBot: 4, all of them 404s on subdomains we do not serve.

The part that actually changed how I think

Google had indexed us. Search Console said the homepage was fine, no errors, nothing to fix. And it bought us nothing, because the crawlers that feed assistant answers had never visited. I had been treating "indexed" as a proxy for "findable" and those are now two different systems.

The dead domain is the bit I keep coming back to. It has age and some inbound links and that was enough to beat a live site the model had never fetched. You are not competing with your competitors for that answer. You are competing with whatever the model already believes.

What we changed

Nothing clever. We were serving a 404 on /llms.txt, so we shipped one. Our long-form guides were rendered client-side, so a crawler received a nav bar and nothing else – server-rendered them. Added Article, FAQPage and Organization JSON-LD to the pages that carry facts. Requested indexing per URL. Now I track two numbers weekly: crawler hits by user-agent, and how many of the 12 prompts cite us. Impressions I have stopped looking at.

Too early to claim a result. I will post the after when there is one, including if it does not move.

What I would tell you to do today

grep -Ei 'GPTBot|ClaudeBot|OAI-SearchBot|ChatGPT-User' access.log | wc -l 

If that returns 0 and you are building anything people would ask an assistant about, you have a distribution problem that no amount of content fixes, and you can find out in about ten minutes.

Happy to paste the exact 12-prompt set if anyone wants to run it on their own thing.