Private Claude Chats Were Indexed by Google and Bing: What Happened

Shared Claude chat URLs surfaced in Google and Bing search results over the weekend, exposing conversations ranging from legal ethics questions to erotic role play. The root cause: Anthropic relied on a robots.txt file to block search engine crawlers from shared chats, but both Google and Bing document that robots.txt alone is not enough to prevent indexing. The shared pages also lacked a "noindex" HTML tag, which both search engines say they respect. Anthropic did not respond to requests for comment from WIRED.

What happened

Detail	Fact
Engines affected	Google and Bing
Bing results (at time of writing)	About 612 results for site:claude.ai/share
Anthropic robots.txt blocking shared chats since	At least September 2025 (per Wayback Machine)
Missing tag on exposed pages	“noindex” HTML tag
Previous incident	Same issue flagged in September (reported by Forbes)
Anthropic response	No comment to WIRED

Claude lets users generate a public URL, called a “snapshot,” for any chat thread and share it with others. Anthropic has instructed web crawlers to stay away from those shared URLs via its robots.txt file since at least September 2025. The problem is that robots.txt is not a guarantee.

Google’s own developer documentation states it will ignore robots.txt instructions if the page is linked to from elsewhere on the web, unless the page also carries a “noindex” tag in its HTML or an “x-robots-tag” in the HTTP response header. Bing’s documentation makes the same recommendation. WIRED reviewed a sample of the exposed pages and confirmed that none of them included that tag.

The issue was first flagged by a Reddit user. Among the conversations that appeared in search results: a user asking which political party to join, a question about whether Kansas attorneys must self-report ethical violations, and erotic role play.

Who said what

Google spokesperson Ned Adriance told WIRED that indexing those pages is Anthropic’s responsibility. “Neither Google nor any other search engine controls what pages are made public on the web,” Adriance said, adding that Google gives site owners clear controls and always respects them when they are set correctly.

Microsoft, which owns Bing, did not comment. Anthropic did not answer questions about why the “noindex” tag was absent from shared chat pages.

By the time WIRED published, Google results for the share query had cleared and some individual chat URLs had been deleted. Bing still returned roughly 612 results. Importantly, the pages WIRED reviewed still lacked the “noindex” tag, meaning the same problem could recur.

Why does robots.txt fail here?

robots.txt tells crawlers not to fetch a page. It does not tell a search engine to remove a URL it already knows about from another source, such as a backlink. If any page on the web links to a shared Claude chat, Google can list that URL in its index even without crawling the page directly. Only a “noindex” directive on the page itself reliably prevents that.

This is not a new or obscure distinction. It is documented in Google’s and Bing’s public developer guides. Anthropic was warned about this exact gap in September, according to the Forbes report WIRED referenced, and apparently did not add the missing tag in the months that followed.

Separately, WIRED noted that Anthropic, Meta, and OpenAI all use robots.txt to block each other’s AI training crawlers from their chatbot sites. Google did not respond to questions about its competitors blocking its Google-Extended training crawler.

Why it matters

Users who share a Claude chat link assume their audience is whoever receives the link. Discovering that a conversation about sensitive legal, political, or personal topics is one Google search away is a significant breach of that expectation, even if the user technically made the URL public.

This also illustrates a gap that affects any web product with user-generated shareable URLs. Relying on robots.txt alone is insufficient for pages that should not be indexed. As teams build AI-powered features that generate shareable content, the indexability of those pages needs to be explicitly managed at the HTML level, not just the crawler level. Our earlier coverage of AI security gaps at major labs showed this pattern repeating: security and privacy controls added after the fact, rather than baked in from the start.

Our take

Anthropic’s response here is hard to defend. This was not a novel attack or an obscure edge case. Google and Bing both publish clear guidance: robots.txt is not enough; you need “noindex.” The company was told about this in September. Months passed. The tag was still missing.

The broader lesson for any developer shipping user-facing features with shareable links: treat indexability as a product requirement, not a footnote. Audit your robots.txt, check your HTML, and test with “site:” queries on both Google and Bing before you ship. If you are building features on top of LLM APIs or third-party AI tools, the infrastructure decisions those vendors make can directly affect your users’ privacy too.

What to do about it

1. If you use Claude, go to Settings, then Privacy, then Shared chats, and revoke any shared links you no longer want accessible.
2. Search “site:claude.ai/share” on both Google and Bing to check whether any of your own shared chats are still indexed.
3. If you run a web product with shareable user content, audit every public URL type to confirm a “noindex” tag is present on pages you do not want crawled, and verify with a “site:” search.
4. Do not rely on robots.txt alone for privacy-sensitive URLs. Add “noindex” in the HTML head or as an HTTP response header, and test both.

For any team building products where user data and public URLs intersect, getting your technical SEO foundations right is not optional. The same controls that keep sensitive pages out of search results also determine what your users can expect to stay private.

Source: WIRED · AI