Claude shared chats appeared on Google after indexing gap

Claude shared chats Google results became a privacy problem after a Reddit user found hundreds of readable conversations through a site-specific search on July 25. For anyone who uses AI tools for work, investing, legal questions, or crypto, the episode is a reminder that “anyone with the link” can become much broader once search engines see that link.

The Reddit user searched Google for site:claude.ai/share, a query that asks Google to show indexed pages from that part of Claude’s website. The results included shared Claude conversations that users had made available through Anthropic’s link-sharing feature.

The thread on Reddit drew about 4,000 upvotes. According to the posts and examples discussed there, exposed chats included crypto wallet credentials, apparent Social Security numbers, legal discussions, and other private material. One example involved a crypto wallet seed phrase, which is the recovery credential that can give whoever has it control of the wallet’s funds.

Why were Claude shared chats on Google?

The issue centered on a missing noindex instruction, according to the reported findings. A noindex tag is a line on a webpage that tells search engines not to list that page in search results.

Anthropic’s Claude share links were meant to work more like unlisted pages: accessible to people who had the URL, rather than searchable by the public. But once a share link appeared on a public forum, social post, or other webpage, Google could learn that the URL existed and add it to search results.

Anthropic already had a robots.txt file blocking Claude share URLs from being crawled. Robots.txt is a standard file websites use to tell search engines which pages not to visit. Google’s own documentation says that blocking crawling does not necessarily block indexing if the URL is discovered somewhere else online. In that setup, Google may know the page exists even if it cannot read the page contents.

That explains why some results appeared with a message saying no page information was available. Google had found the link from outside Claude, but the crawler could not access the page to read more about it.

What else was exposed?

The same indexing issue also affected Claude’s public artifacts, according to the findings. Artifacts are things users build or publish through Claude, such as apps, dashboards, or financial tools.

A search for Claude public artifact pages reportedly surfaced payroll spreadsheets with employee names, internal customer relationship management chat exports, and unreleased product roadmaps. For retail investors and employees alike, that is the uncomfortable part: AI tools are now places where people paste business data, identity data, and financial credentials in one place.

OpenAI faced a related problem 11 months earlier with a “Make this chat discoverable” option for ChatGPT conversations. OpenAI Chief Information Security Officer Dane Stuckey later removed the feature, calling it “a short-lived experiment to help people discover useful conversations,” according to a Malwarebytes report cited in coverage of the incident. Some indexed ChatGPT conversations were saved by the Internet Archive before cleanup and remained accessible there.

What did Anthropic change?

Google began removing Claude shared-chat results by July 26 after Anthropic changed its configuration and added noindex tags, according to the reported timeline. Anthropic had not issued a public statement at the time of publication.

Bing was still showing some shared Claude links after Google’s results started disappearing. A public GitHub repository called Shared-Claude-Chats had also archived 453 Claude conversations and 519 Grok chats from exposed links, totaling 11,241 messages in plain text.

Claude users can revoke shared conversations through Settings, then Privacy, then Shared Chats. That can stop future access through those links, but it does not remove copies already saved elsewhere.

This story draws on original reporting from Decrypt.