Some apps, documents and other artifacts created in Claude and shared through public links are appearing in Google search results, turning a convenience feature into a useful warning about the way AI products define privacy.
The distinction matters: Claude chats are private by default, and the reported examples were not private accounts being breached. Anthropic’s own sharing documentation says a user must create a shareable snapshot. Anyone with that link can then view the messages and artifacts included at the time of sharing.
Axios reported on July 27 that Google had indexed a range of publicly shared Claude artifacts, including business plans and infrastructure diagrams. It found artifacts rather than the underlying chatbot conversations in its searches. Anthropic said it does not give search engines a directory or sitemap of those pages and that a link becomes discoverable only after someone shares it somewhere a crawler can see, such as a forum or social network.
The design gap inside “anyone with the link”
Technically, the system behaved like the public web. Practically, many users treat an unguessable link as a private handoff. That mental model was reinforced by years of collaboration products in which “anyone with the link” often meant a document would be accessible but unlikely to surface in search.
Generative AI raises the stakes because a single shared page can combine prompts, strategic thinking, generated code and finished artifacts. The content may look polished enough to distribute before anyone stops to classify the information inside it. A public-link action therefore needs more than a generic visibility label; it needs to communicate the real downstream possibilities, including indexing, archiving and redistribution.
What companies should do now
Organizations using consumer AI tools should treat public share links as publication, not collaboration. Teams should review existing shared-link logs, unshare material that no longer needs to be public and move sensitive work into controlled enterprise environments. Policies should also define which content may never be placed in a public snapshot, even if the underlying file attachment or tool data is excluded.
Product teams have a design question of their own: should public AI pages carry a clear search-indexing warning, a no-index default or a separate option for deliberate publication? None of those choices removes the user’s responsibility, but each would better align interface language with how people actually understand sharing.
The broader lesson is straightforward. In AI products, privacy is no longer only a data-storage promise. It is also the design of the moment when private work becomes public.
About Julie: Julie Wohlberg is a journalist, communications strategist and entrepreneur with more than 25 years of experience building brands and launching companies. A former Tribune news and feature writer and early executive at Fotolia, she writes about media, brand, consumer behavior and the technologies reshaping how companies grow. She is the Founder + Principal at Inflection Point, a founder-led, senior growth solutions partner for startups and growth stage businesses.