Claude chats exposed through Google and Bing search results have put Anthropic’s sharing controls under scrutiny after users found that some public Claude conversation links were searchable over the weekend, according to WIRED.
The chats were not ordinary closed sessions. Anthropic lets Claude users create public links to “snapshots” of specific conversations so they can share them with others. WIRED reported that some of those shared pages surfaced in search results, including chats about political affiliation, legal ethics reporting rules in Kansas, and erotic role play.
WIRED said the issue appears to have been first raised by a Reddit user. By the time WIRED checked, some of the linked conversations highlighted in the Reddit post had already been removed.
Why were Claude chats exposed in search results?
The short version: Anthropic told crawlers not to visit shared Claude pages, but WIRED found that the sampled pages lacked a separate instruction that Google and Bing say helps keep pages out of search indexes.
Anthropic used a robots.txt file to disallow crawlers from the shared-chat area of Claude, according to WIRED. A robots.txt file is a site-level instruction that asks crawlers, including search engine bots, not to access certain paths. It is old web plumbing, useful but easy to overestimate.
Search indexing has another layer. Google says in its developer documentation that a page blocked by robots.txt can still appear in search if another site links to it and the page does not also carry a noindex directive in the HTML or response header. Bing’s documentation also tells developers they can block crawlers with robots.txt and should use noindex tags on individual pages when they want to prevent indexing.
WIRED reviewed a sample of exposed Claude share pages and reported that they did not contain the noindex tag that Google and Bing say they consider when deciding whether to list a page.
What Google, Microsoft, and Anthropic said
Google spokesperson Ned Adriance told WIRED that Anthropic was responsible for whether the shared pages were public and indexable. He said search engines do not decide what web pages are made public, and that Google provides site owners with controls for crawling and indexing.
Microsoft, which owns Bing, did not comment before WIRED published. Anthropic did not respond to multiple requests from WIRED, including questions about why the shared Claude pages did not include noindex tags.
WIRED reported that Google no longer returned results for the Claude share search query that had worked earlier. Bing, however, still showed about 612 results for a site search targeting claude.ai/share at the time WIRED checked.
The same problem has come up before. Last September, Anthropic faced criticism after Forbes reported that hundreds of chatbot transcripts appeared in Google search. Anthropic told Forbes then that it used robots.txt to signal that crawlers should not access shared chats. Forbes noted that robots.txt alone does not guarantee that a specific page will stay out of search results.
How Claude users can check shared chats
Claude users who have created shared conversation links can review them inside the product. WIRED reported that users can go to Settings > Privacy > Shared chats to manage access.
The broader lesson is not especially glamorous: if a chatbot gives a conversation a public URL, treat that link like a web page. Search engines, crawlers, and link-sharing systems are very good at finding web pages, especially when the site owner relies on a polite crawler request instead of an indexing block.
This story draws on original reporting from WIRED.
