AI StrategyChoosely EditorialEvidence-based analysis

Are Claude Chats Private? How to Check Whether Your Shared Chats and Artifacts Are Public

Private Claude chats were not leaked. Publicly shared conversations and Artifacts became searchable, exposing a gap between what Share technically meant and what users reasonably expected.

← Back to AI Radar
A person at a desk reviewing Claude shared chats flowing from a private workspace into public search results.

Last verified: July 30, 2026. Search visibility changed quickly during the reporting period, and results differed between Google, Bing, shared chats, and published Artifacts.

Choosely reviewed Anthropic's sharing documentation, Google's indexing guidance, and reporting from WIRED, TechCrunch, Axios, and Forbes. We did not search for, access, or reproduce exposed personal conversations.

Your ordinary Claude chats were not dumped into Google.

The pages that became searchable were conversations and Artifacts that users had turned into public links using Claude's sharing or publishing features. Private, unshared chats remained private from the public.

That distinction matters, but it does not get Anthropic completely off the hook. Reporting found shared pages containing sensitive medical, company, contact, financial, and access-related information. This was not a breach of Claude accounts. It was a preventable privacy and product-design failure involving content that users had technically made public without necessarily understanding how public it could become.

The practical rule is simple:

If you have ever created a Claude share link or published an Artifact, treat it as a public webpage and review it now.

Public by design, searchable by surprise

The issue drew widespread attention over the weekend of July 25 after users found that searches targeting Claude's sharing URLs could surface public conversations and Artifacts. TechCrunch reported shared pages containing sensitive records and private business material. WIRED reviewed Claude conversations found through search, while Axios was still able to locate indexed Artifacts on Google after conversation results had begun disappearing.

The search-engine status moved quickly. TechCrunch could no longer reproduce the Google conversation results by Monday afternoon. Axios could still find public Artifacts on Google, while WIRED reported that its Claude chat query continued returning results on Bing at the time of publication. Those observations should be treated as timestamped findings, not a permanent description of what either search engine shows today.

This also was not Anthropic's first warning.

In September 2025, Forbes reported that Google had indexed just under 600 Claude conversations created through the same sharing system. Anthropic said those pages had become visible because users posted their links online, although one identifiable user told Forbes they had not knowingly done so. The results later disappeared.

A second round of the same problem is harder to dismiss as an obscure edge case. The sharing system had already demonstrated that public links could become search results.

What Claude actually shares

Anthropic says normal Claude chats are private by default. On Free, Pro, and Max accounts, choosing to share a chat creates a public snapshot that anyone with the link can view.

Claude contentPublic by default?What happens when you share or publish it?
Normal consumer chatNoIt remains private until a public share link is created
Shared consumer chatYesAnyone with the link can view the saved snapshot
Later messages in the same chatNoMessages added after the snapshot remain private unless the chat is unshared and shared again
Uploaded fileNoThe original file is excluded, but Claude's visible summary or extracted content may appear in the conversation
Raw MCP tool-call dataNoThe raw retrieved data remains hidden, but Claude's final output is visible
Published Artifact on Free, Pro, or MaxYesAnyone with the link can view and interact with it
Team or Enterprise chatNo public sharingAccess is limited to authenticated members of the organization
Team or Enterprise ArtifactNo public publishingIt can only be shared inside the organization

A shared chat snapshot contains the messages and Claude responses created before the link was generated, including any Artifacts in the conversation. Messages sent afterward remain private unless the user later refreshes the public snapshot by unsharing and sharing it again.

Attached files are handled differently. Anthropic says the original uploaded file is not placed inside the public snapshot, and raw information retrieved through MCP tool calls remains hidden. That protection only goes so far. If Claude quoted a document, summarized a spreadsheet, extracted names, or turned private source material into a report, the visible output can still be shared. The file may remain private while its useful contents sit neatly on the public page.

Artifacts carry their own rules. On consumer plans, publishing makes an Artifact publicly available to anyone with the link, including people without a Claude account. Other Claude users can also copy its code and build from it. On Team and Enterprise plans, sharing keeps access inside the authenticated organization. The wording is doing a fair amount of work there.

Anthropic's search controls were not good enough

Anthropic told reporters that it does not provide search engines with directories or sitemaps of shared conversations. It also said the URLs are difficult to guess and generally become discoverable only when someone posts the link somewhere a crawler can see, such as a forum or social network.

That may explain how individual links were discovered. It does not resolve the product problem.

WIRED found that sampled Claude share pages lacked a noindex instruction, which search engines use to keep pages out of their results. Anthropic had blocked its share path in robots.txt, but Google explicitly warns that robots.txt is mainly a crawling control and should not be used to hide confidential web pages from search. A blocked URL can still appear when another page links to it.

The technical fix is slightly messier than simply applying both controls at once. Google must be allowed to crawl a page before it can read a page-level noindex tag or response header. Blocking the crawler and adding noindex can therefore prevent the crawler from seeing the instruction that says not to index the page.

That nuance does not excuse the outcome. It strengthens the broader point.

Search-engine instructions are not access controls. Google recommends password protection for genuinely private material. Removing a page, requiring authentication, or preventing unauthorized access is more reliable than asking crawlers to behave themselves.

Anthropic's interface did describe the option as creating a public link. Fair enough. But many users understand "anyone with this link" as an unlisted handoff to a colleague, client, or friend. They do not necessarily understand it as permission for the page to appear beside ordinary websites in Google.

A clearer warning would have left less room for confusion:

This creates a public webpage. It may be indexed, archived, copied, or reshared. Do not include confidential information.

That is less elegant than a friendly Share button. It is also considerably harder to misunderstand.

How to check every Claude chat you have shared

Free, Pro, and Max users can review their public chat snapshots from Claude's privacy settings:

  1. 1Open Claude.
  2. 2Select your profile or initials.
  3. 3Open Settings.
  4. 4Select Privacy.
  5. 5Find Shared chats.
  6. 6Select Manage.
  7. 7Review every title, sharing date, and link.
  8. 8Select Unshare beside anything that no longer needs to be public.

Unsharing changes the snapshot from public to private and disables the direct link. When Anthropic has no public snapshots recorded for the account, the Shared chats window displays No shared content found.

Do not rely on the conversation title. A harmless-looking chat can contain a pasted email, a customer name, an internal URL, a résumé, an API response, or a document summary further down the thread.

Check published Artifacts separately

Published Artifacts use a separate management system, so clearing the Shared chats list does not finish the audit.

Open Claude's Artifacts section and review anything you previously published. Unpublish any Artifact that no longer needs public access.

Anthropic warns that unpublishing is permanent for that version. The same Artifact cannot simply be published again later, and any persistent storage attached to it is deleted. You would need to create a new Artifact if you wanted another public version. Slightly inconvenient, but considerably better than leaving company material on the open web.

For Team and Enterprise accounts, internal Artifact sharing has an additional catch. Anthropic says members who receive access may also gain access to attachments and files in the conversation that created the Artifact. Organizations should review that source conversation before sharing, not only the finished output.

What to do if something sensitive was exposed

Start with the source page. Search cleanup comes afterward.

  1. 1Unshare the chat or unpublish the Artifact. Confirm the original URL no longer provides access.
  2. 2Rotate credentials immediately. Replace any password, API key, access code, wallet secret, or token that appeared in the conversation.
  3. 3Escalate business exposure. Notify the appropriate security, privacy, legal, or management contact if customer, employee, health, financial, or confidential company data was involved.
  4. 4Search for the exact URL and distinctive text. Check more than one search engine.
  5. 5Request a search update after the page is gone. Google's Refresh Outdated Content tool can update results for a page that no longer exists or has removed important content. It cannot remove a public page that remains live.
  6. 6Assume copies may exist. Removing the original link prevents normal future access. It cannot guarantee that nobody archived, downloaded, or copied the content while it was public.

Changing an exposed API key matters more than removing its Google snippet. The search result is embarrassing. The working credential is the actual incident.

Businesses using consumer Claude accounts should also make one policy explicit:

A public AI share link must never contain credentials, customer information, unreleased work, legal advice, health records, internal financial data, or confidential source documents.

Sensitive work belongs in an approved business workspace with authenticated access and clear sharing controls.

Choosely's view

Claude chats remain private by default. This incident did not expose every private conversation or allow strangers to search inside ordinary Claude accounts.

Anthropic still handled the sharing risk poorly.

The company built a feature that converted conversations into public web pages, relied on search controls that had already failed once, and left too much room between the technical meaning of public and the way ordinary users understand Share. Its explanation that users had posted the links somewhere crawlable may be technically relevant. It is not a complete answer to a recurring product-design failure.

Anthropic deserves credit for giving users one place to review and revoke shared chats. That control is useful, and the five-minute audit is worth doing.

The safer mental model remains:

Share means publish.

Review every shared chat. Check published Artifacts separately. Revoke anything old, sensitive, or unnecessary.

Discovering your private business plan through a branded Google search would be a fairly ordinary way to ruin the morning.

Verified sources

The Change Brief

Get the week’s AI changes in one clear read

Pricing moves, tool launches, free-tier changes and practical stack updates, filtered for people who actually use these tools.

Stay ahead of AI without following it all day. We’ll send you what matters each week.

Continue reading

Related reads