Users report Claude shared links indexed by Google Search
Viral posts claim public Claude chats and artifacts were indexed by search engines. Examples allegedly include API keys, resumes, legal notes, dashboards, project plans, and medical summaries.

TL;DR
- Claude shared-chat URLs reportedly appeared in Google results: Om Patel's report said searches under
site:claude.ai/sharesurfaced public links with credentials, resumes, legal notes, internal projects, and personal chats. - Published artifacts entered the same blast radius: Patel's artifact update said
site:claude.ai/public/artifactsexposed apps, docs, dashboards, project plans, and a clinical-trial summary. - Anthropic's own docs describe sharing as a public snapshot, because the Claude shared-chat guide says anyone with the link can view messages sent before sharing, including artifacts, matching the shared-chat report.
- The technical gap is search discoverability, not account access: Patel's source screenshot showed indexed Claude share URLs, while Google Search Central says robots.txt is not a way to keep pages out of Google.
The artifact search in Patel's screenshot surfaced Claude-hosted pages with titles that read like presentations, claim analyses, academic calendars, and press releases. The artifact help page says Free, Pro, and Max users can publish artifacts as public links that anyone can view and interact with. Google's own robots.txt guide gives the blunt search-side version: robots.txt manages crawling traffic, not search exclusion.
Search-indexed share links
Patel's first post described a privacy incident around Claude's Share button: a public link meant for "anyone with the link" had become discoverable through search. He claimed people were finding:
- API keys, credentials, and crypto wallets
- Resumes with real names, addresses, and phone numbers
- A lawyer working through a possible ethics violation
- An engineer's internal company project details
- What appeared to be Social Security numbers
- Deeply personal chats
The screenshot attached to Patel's source post showed a Google query for site:claude.ai/share and multiple claude.ai/share/ results labeled "Claude," with Google saying no page information was available.
Public reporting framed the incident as indexed shared pages, not a bypass of private Claude accounts. The CyberSec Guru's incident writeup said it had found no evidence that unshared private conversations were exposed.
Public artifacts
Patel's second post moved from chat transcripts to Claude Artifacts, the app, document, dashboard, and tool format creators use when Claude generates an interactive output. He claimed people had already found:
- Internal company dashboards
- Full project plans with client details
- A clinical-trial summary with medical data
- Financial models, cap tables, and revenue spreadsheets
- Payroll and salary breakdowns with employee names
- Customer lists and CRM exports
- Internal wikis, roadmaps, and unreleased product plans
- Legal docs, contracts, tax documents, and bank-statement breakdowns
- Dashboards containing pasted API keys or environment variables
The same artifact search appears in Patel's artifact source post, with Google results under site:claude.ai/public/artifacts. That screenshot matters more for creators than the chat transcript screenshot, because artifacts are often the polished thing people publish after Claude turns messy source material into a usable page.
Anthropic's artifact publishing docs say "publishing" on Free, Pro, and Max makes an artifact publicly available, while Team and Enterprise "sharing" is limited to authenticated users inside the organization. The docs say anyone with the public link can view and interact with a published artifact.
Robots.txt and noindex
Claude's current robots.txt disallows /share/*, but it does not include a matching /public/artifacts disallow. That split matches the screenshots: the shared-link screenshot shows Google listing Claude share URLs without page details, while the artifact screenshot shows artifact page titles.
Google's robots.txt introduction says the file tells crawlers which URLs they can access, but "is not a mechanism for keeping a web page out of Google." Google's no page information help page says Google can still index a URL it cannot crawl and show it in results without a snippet.
Patel blamed the absence of a noindex tag in the original report. Google says a noindex rule drops a page from Search results after Googlebot crawls the tag or header.
Shared-chat snapshots
Anthropic's shared-chat help page gives the product boundary:
- Chats are private by default.
- Sharing creates a snapshot viewable by anyone with the link.
- The snapshot includes messages sent before sharing, including artifacts.
- Messages sent after sharing remain private unless the chat is unshared and shared again.
- Uploaded files themselves are not included in the shared snapshot.
- Raw MCP tool-call data remains hidden, while final chat output and conversation text remain visible.
- Team and Enterprise users can share chats only with members of the same organization, not publicly.
That boundary still leaves a common creator failure mode: pasted source material can survive in the conversation text or artifact output even when the original file is not attached. Patel's examples in the shared-chat report and the artifact update were mostly about content users had placed into Claude, then published through Claude's own sharing surfaces.
Unshare and unpublish controls
Anthropic documents two different revocation paths.
- Shared chats: the shared-chat guide says Free, Pro, and Max users can open Settings > Privacy > Shared chats, review titles, dates, and links, then click Unshare; Patel's original post pointed users to the same menu.
- Published artifacts: the artifact guide says Unpublish revokes access, but the same artifact cannot be published again afterward.
- Persistent-storage artifacts: the artifact guide says unpublishing permanently deletes associated storage data when an artifact used persistent storage.
- Team and Enterprise artifacts: the artifact guide says shared artifacts stay inside the organization and can be unshared from the Artifact shared modal.
Patel's artifact post said third-party sites had begun scraping and organizing published artifacts into browsable collections in the artifact update. Search removal and link revocation do different jobs once a public page has been copied elsewhere.