Private conversations people had with Anthropic's Claude chatbot turned up in ordinary Google and Bing searches over the weekend, after users discovered that shared chat links, meant to be seen only by whoever received the link, were instead being crawled and indexed like any other public webpage on the internet.
The exposure was first flagged by a Reddit user, and the conversations that surfaced included people asking Claude for advice on which political party they should join, questions about whether attorneys in Kansas are required to self-report when they believe they have committed an ethical violation, and, in some cases, erotic role play chats, all conversations most people would assume were meant for a single recipient, not the open web.
How Claude's sharing feature works
Claude lets a user share a snapshot of a chat with someone else by generating a public URL for that specific conversation thread. That URL is meant to work like a private link passed between two people rather than a page built for search traffic. The trouble is that a link being technically public, even if nobody ever advertises or posts it widely, is enough for a search engine's crawler to stumble onto it once it is linked to from anywhere else on the internet, which is exactly the collision of ordinary web mechanics and generative AI sharing tools that produced this exposure.
Why the robots.txt block did not hold
To stop crawlers from reaching shared chats, Anthropic relies on a file called robots.txt, the long-standing standard method websites use to tell crawlers, including those run by Google and Bing, which parts of a site should not be accessed. A snapshot from the Wayback Machine shows Anthropic's robots.txt has instructed crawlers to stay away from shared chats since at least September 2025.
But robots.txt alone has never guaranteed that a page stays out of search results. Bing's own technical documentation says developers should not rely on robots.txt by itself and should also add a noindex tag to individual pages. Google's developer guidance goes further, stating that it will ignore robots.txt instructions altogether if a page is linked to from somewhere else on the internet and that page does not separately carry a noindex HTML tag or an x-robots-tag in its response header. A review of a sample of the exposed Claude chat pages found that none of them carried the noindex tag that both companies say they check for before deciding whether to index a page.
Google and Bing point the responsibility back at Anthropic
At the time of this report, searching site:claude.ai/share on Bing still returned about 612 results. Some of the chats originally flagged in the Reddit post had already been deleted by the time they were reviewed, and the specific share search that still worked on Bing no longer surfaced any results on Google.
Microsoft, which owns Bing, did not provide comment. Anthropic did not respond to multiple requests for comment either, including questions about why it never added a noindex tag to its shared chat pages. Google spokesperson Ned Adriance said responsibility for what gets indexed sits with Anthropic, not with search engines. “Neither Google nor any other search engine controls what pages are made public on the web, and these pages were indexed across many search engines,” Adriance said, adding that Google gives site owners “clear controls to decide whether pages can be crawled or indexed” and that it always respects those directives.
Not the first time this has happened
This is not a new problem for Anthropic. The same issue surfaced last September, when the company said it uses robots.txt to tell crawlers not to access shared chats, while also acknowledging that the file offers no guarantee against a page still being indexed. Nearly a year later, the exact same failure mode has repeated itself, exposing a fresh batch of private conversations to anyone running a basic search.
Robots.txt still matters, just for a different fight
Even though robots.txt cannot reliably stop a determined search engine from indexing a page, AI companies keep using the file for a separate purpose, keeping rival companies' crawlers out of their training data. Many AI labs promise website owners that their content will not be used to train models as long as the site disallows specific crawlers in its robots.txt file, and the labs apply that same logic to each other. Anthropic, Meta and OpenAI all list instructions in their own chatbots' robots.txt files that disallow competitors' crawlers from accessing any part of the site that hosts their chatbot. OpenAI and Meta did not respond to requests for comment on the practice, and Google did not address questions about the fact that its rivals are blocking its Google-Extended training crawler from their chatbot sites.
What Claude users should do
Shared Claude chats have stopped turning up in Google's results for now, but the pages reviewed still do not carry a noindex tag, meaning they could resurface in search results again at any point without warning. Anthropic has not explained why the tag was never added to these pages in the first place. Claude users who want to check or restrict who can find their shared conversations can go to Settings, then Privacy, then Shared chats, to manage access and keep sensitive exchanges out of public view.
