Private Claude Chats Ended Up in Google's Index

Public links to Claude conversations ended up in Google Search — and it wasn't a hack but a simple oversight: the chat pages were open to search engines, and no one blocked them from indexing.

The mechanics are mundane. When a Claude user turned on «share this conversation», Anthropic generated a public URL. Those addresses were reachable by crawlers, with no noindex meta tag and no robots.txt disallow. The rest was predictable: as Search Engine Land puts it, if a URL is accessible to a search engine and isn't explicitly excluded, it will get indexed. The same thing had already happened with shared chats on other AI services — the same rake, stepped on again. A «share» button now ships with almost every service, and each such link is a separate page a crawler can reach and index.

What stands out here:

• links that looked private were actually public — that alone was enough to land a conversation in the index;
• the cause wasn't a «data breach» but the absence of noindex and robots.txt rules — basic indexation hygiene;
• one shared page is enough for it to surface on an ordinary Google query.

The lesson applies to any site with «not for everyone» pages: drafts, user accounts, internal search, share URLs. A search engine doesn't read minds — it indexes everything that's open and not marked noindex. Closing that off takes basic tools: robots.txt and noindex for share and service sections, an X-Robots-Tag header on private URLs. Checking is just as simple — Search Console's «Coverage» report shows what actually made it into the index. Controlling indexation comes down to regular checks; a single setting won't cover it.

The most unpleasant part of this story is that «private» in the interface and «hidden» from Google looked like the same thing to users, and it took search itself to show the difference.

Related reading: Indexed and Non-Indexed Pages: Common robots.txt and Noindex Mistakes · SEO site audit