Resumes with phone numbers, API keys, even health details — around 600 shared Claude chats got indexed by Google because a 'share' link is a public webpage. It's fixed now, but the lesson applies to every AI chatbot you use.
That AI conversation you shared with a colleague? A search engine may have read it too. In late July 2026, Reddit and X users discovered that Google had indexed around 600 publicly-shared Claude conversations and Artifacts — and when people started searching, what turned up was alarming: resumes with real names and phone numbers, API keys, crypto wallet details, a lawyer's notes on an ethics case, financial spreadsheets, and content that appeared to include Social Security numbers. Anthropic moved fast and had the results wiped from search by July 28. But the underlying trap isn't specific to Claude — it's built into how 'share' links work on nearly every AI chatbot, and it's worth understanding so you don't fall into it.
Why 'anyone with the link' quietly means 'anyone on Google'
The mechanism here is mundane, which is exactly why it keeps happening. When you share an AI conversation, the app generates a public URL — a normal webpage that anyone can open without signing in. That's the intended feature: you paste the link to a friend, they see the chat. The problem is that 'a public webpage anyone can open' is also precisely what a search engine crawls and indexes. Unless the site explicitly tells crawlers to stay away, those shared pages get treated like any other public content and can surface in search results. The Axios report frames it as an avoidable flaw, and it is: the gap was between what users understood 'share' to mean and what it technically did.
A share link is a public URL. Unless the platform blocks crawlers, 'anyone with the link' quietly includes anyone who searches. · Unsplash
How to check and clean up your own shared chats
Open your AI chatbot's settings and find the 'Shared links' or 'Shared conversations' section — most now list everything you've published.
Review the list and delete/unpublish anything sensitive, or anything you shared once and forgot about. If in doubt, unpublish it.
Search Google for site:claude.ai (or the relevant domain) plus your name or a phrase from a chat, to see if anything of yours is still indexed.
Going forward, never paste secrets — API keys, passwords, personal data — into a conversation you might share. Treat share links as public.
The good news is that fixing your exposure is quick: unpublishing a shared conversation removes the public page, and once the platform has told search engines to drop it, the result disappears from Google over time. But prevention beats cleanup. The habit worth building is simple — assume anything you put in an AI chat could one day be shared, and anything you share could one day be searched. It's the same discipline as not putting secrets in a public GitHub repo: the tool will happily publish exactly what you tell it to.
Quick answers
Were my Claude chats exposed?
Only chats you explicitly shared via a public link could have been indexed — your private conversations were never public. Around 600 shared chats were found in search. To check, open Claude's settings, review your shared links, and unpublish anything sensitive. You can also Google 'site:claude.ai' with your name or a distinctive phrase to see if anything of yours surfaced. Anthropic has since blocked search indexing of these pages.
Is it safe to share AI conversations at all?
Yes, as long as you treat a share link as fully public. Sharing a chat that contains nothing sensitive is fine. The mistake is sharing (or forgetting you shared) a conversation that includes personal data, credentials, or confidential work. Before sharing, reread the conversation for anything you wouldn't post publicly. And periodically clear out old shared links you no longer need.
Did Anthropic fix this?
Yes — Anthropic updated its robots.txt and related measures to stop search engines indexing shared conversation pages, and the existing results were cleared from Google and Bing by July 28. The company noted that shared links are public by design and urged users to review and unpublish sensitive content in their settings. The indexing vector is closed, but the general advice to treat share links as public still stands.
Has this happened with other AI tools?
Yes. ChatGPT had the same issue — shared conversations were indexed by Google until OpenAI changed how the feature worked. It's a recurring pattern because the root cause is universal: a 'share' feature creates a public webpage, and public webpages are crawlable. Any AI product with link-sharing is susceptible unless it explicitly blocks indexing, so the caution applies across the board, not to one vendor.
The fix took Anthropic a few days; the lesson is permanent. Every AI chatbot's 'share' button creates a public page, and public pages get found — so review your shared links, unpublish anything sensitive, and never put secrets in a conversation you might share. Sources: Axios and TechSpot. It's the digital-hygiene version of the same lesson behind this month's bigger breaches: the exposure is almost always something that was left public by accident.
