How to Delete Claude Artifacts (and What Deletion Actually Removes)

If you are searching for how to claude delete artifacts right now, something probably went out and you want it back. Work through the steps below in order. Unpublishing comes first because it is the only step that stops new readers, and you want the conversation still open while you find its sharing control.

How to delete a Claude artifact (unpublish first)

Claude's interface changes often, so each step names what the control does rather than where it sits. For background on artifacts and published artifacts, see what a Claude artifact is and how to share one.

1. Unpublish the published link. Open the artifact and find its share or publish control. If it shows any state other than private, turn sharing off or delete the published version. If you also shared the conversation itself as a link, revoke that too. Do this before anything else, while the conversation still exists.
2. Delete the artifact. If the artifact has its own menu with a delete or remove option, or your account lists artifacts in a separate space that offers one, use it. If there is no per-artifact delete, the conversation is the unit of deletion and step 3 handles it.
3. Delete the conversation. Open the conversation's menu, from the chat list or the conversation itself, and choose delete. The conversation and every artifact in it leave your history.
4. Verify from outside. Open the old published URL in a private browser window. If it still renders, a published copy survived; look for a list of shared or published items in your account settings, or send the URL to Anthropic support.

Clearing all history and deleting the account sit in account settings. If a client or auditor may ask about this document, write down what was shared, when, and at which URL before you bulk-delete, because bulk deletion also erases your own record of what went out.

Deleting takes the conversation out of your history. How long Anthropic keeps it on its back end after that is governed by Anthropic's retention policy, not by the button you pressed. Anthropic's Privacy Center documents the current periods, which can differ between consumer plans and commercial agreements (Team, Enterprise, API) and can depend on your privacy settings. Check the current version there rather than relying on a number you remember or read secondhand.

What deletion doesn't reach

Deleting an artifact removes your access path to it. Every copy that left the conversation before you clicked delete is now controlled by someone else, and no setting in Claude reaches it.

A downloaded file or a screenshot is its own document from then on, with no link back to you and no expiry. Cached copies, the versions a browser or corporate proxy stored on the way through, age out on their own schedule rather than yours. Where shared data goes after it leaves the system of record traces how far those copies tend to travel.

On a work plan, your organization's admin settings and its agreement with Anthropic decide what is retained after a user deletes something. The delete button doesn't override them, so ask your admin what your plan keeps before you tell a client it's gone.

If what you published was a live artifact, Claude live artifacts: what breaks, what leaks covers what else a live one can expose.

Avoid the cleanup next time

Deletion depends on someone remembering a document at the right moment. Expiry set at creation doesn't depend on anyone. If Claude publishes through HTMLvault's MCP server (the Model Context Protocol connection that lets Claude call HTMLvault's tools directly) instead of its own share control, the link carries two settings from the start:

• Link expiry ends access. The URL stops resolving at the time you set, anywhere from one hour to never on Pro and above; Free links expire at 30 days. Details in how to create expiring HTML links securely.
• Data retention ends existence. When the window closes, HTMLvault deletes the HTML and its analytics. Pro and above choose anywhere from 7 days to 2 years; Free retains 90 days. See the data retention window.

Abbreviated, the publish step for a pricing proposal like Chip's looks like this:

scan_html      → 1 flag: person (contact name in the footer)
create_link
  title        "Q3 pricing proposal"
  expiry       14 days
  retention    30 days
  password     on

The scan runs before the link exists. It is regex-based, costs zero tokens, and flags structured patterns such as names, emails, phone numbers, financial strings, and API keys; Teams and Enterprise can add an AI scan on top using their own Anthropic, OpenAI, or Google key. Anything containing a credential or token should not be published at all. Rotate the key, then republish clean. Scan HTML for API keys first, every time.

The password limits who can open the link; the expiry limits how long anyone can. On day 15 the link stops resolving. On day 30 the HTML and its view history leave storage, so export any analytics you need before then. While the link is live, its analytics answer the question deletion never could: who opened it, how far they scrolled, and whether a repeat view from a new city points to a forward. To pick windows per content type, see link expiry settings that match the deal. If you need a link gone early, delete_link ends it on demand.

If you shared something you now regret, three things are still yours to pull back, and each takes minutes. Unpublish, and new views stop as soon as the change takes effect; confirm it in a private window rather than trusting the toggle. If the page held an API key or token, rotate it, which closes that exposure completely no matter who saw it. Then delete the conversation, which clears your history at once while Anthropic's back-end copy follows the schedule in its Privacy Center. The 9:14 read, the download, and the screenshot are already out of reach, so spend the next five minutes writing down what was shared and when, before someone asks.