Private Claude chats turned up in Google and Bing results

Private conversations with Anthropic's Claude chatbot showed up in Google and Bing search results over the weekend, exposing chats that ranged from political questions to erotic roleplay. The root cause: Anthropic relied on a robots.txt file to block web crawlers from indexing shared chat URLs, but both search engines say that directive alone is not enough.

Claude lets users generate public URLs for specific conversations, creating "snapshots" they can send to colleagues or friends. Those links are not guessable. They are only discoverable if someone shares them. The problem is that once a link lands on a public forum, a blog post, or another indexed page, search engines can follow it and add the chat to their results, regardless of what the robots.txt file says.

Why robots.txt alone failed

Anthropic's robots.txt has blocked crawlers from the /share path since at least September 2025, according to Wayback Machine snapshots. But Bing's technical documentation explicitly warns developers: use a "noindex" tag on individual pages as well. Google's own developer guide states it will ignore robots.txt instructions if a page is linked from elsewhere on the web and lacks a noindex HTML tag or an x-robots-tag in the HTTP header.

WIRED reviewed several exposed chat pages and found none included the noindex tag. That gap left them eligible for indexing the moment they were linked from anywhere crawlable.

A redditor first flagged the issue. At the time of reporting, a Bing search for "site:claude.ai/share" still returned about 612 results. Google had already purged results for that query, though WIRED noted the underlying pages still lacked the noindex directive, leaving them vulnerable to re-indexing.

Both sides point fingers

Anthropic says it does not share chat directories or sitemaps with search engines. "When someone shares a conversation, they are making that content publicly accessible, and like other public web content, it may be archived by third-party services," a spokesperson told WIRED.

Google spokesperson Ned Adriance pushed responsibility back: "Neither Google nor any other search engine controls what pages are made public on the web. We give site owners clear controls to decide whether pages can be crawled or indexed, and we always respect those directives." Microsoft did not comment.

Anthropic did not answer questions about why it had not added the noindex tag. This is the second time it has faced criticism for the same issue. Last September, after a similar Forbes report, Anthropic said it uses robots.txt to signal crawlers. That measure clearly has not been sufficient.

A broader tension in AI indexing

Robots.txt has become a contested battleground. Anthropic, Meta, and OpenAI all include directives in their chatbots' robots.txt files blocking competitors' AI training crawlers from accessing their sites. OpenAI blocks Google-Extended. Meta blocks GPTBot. Google did not respond to questions about competitors blocking its training crawler.

The result is an odd asymmetry: AI labs treat robots.txt as a binding promise when it benefits them, but search engines treat it as optional when pages surface through other routes. Neither reading is wrong. Both reflect the file's ambiguous legal and technical status.

What Claude users should do now

Go to Settings > Privacy > Shared chats in your Claude account. Review every conversation you have shared and revoke access to any you no longer want public. Deleting the shared link removes the page, but cached copies may persist in search engines or the Wayback Machine until they refresh or honor a removal request.

For teams using Claude in production, assume that any shared link can end up indexed. Treat the share feature the way you would treat a public Pastebin: convenient for quick handoffs, unsuitable for sensitive information.

The exposed chats included users seeking legal advice and personal guidance, information most would not want searchable. Whether Anthropic patches the gap or not, the episode is a reminder: public URLs are public, and no text file can fully retract that.

Source: Feed: Artificial Intelligence Latest / Maddy Varner