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

This guide covers the mechanics — removing a Claude artifact, deleting the conversation, and what happens to a link you already published — and then the part nobody wants to hear: deletion after the fact is the weakest control you have. The durable fix is content that expires whether or not you remember it exists.

What a Claude artifact actually is

A Claude artifact is the rendered side-panel output Claude produces when you ask for something document-shaped: HTML, a React component, a chart, a markdown doc. It lives attached to the conversation turn that created it. Claude live artifacts — the interactive ones you can click through and iterate on — are the same object with a runtime attached.

Two things matter for deletion. First, an artifact is a child of a conversation, so its lifecycle is the conversation's lifecycle. Second, publishing or sharing an artifact creates a separate object: a hosted copy with its own URL, which is not governed by the chat you deleted it from.

That second point is where most of the frustration behind this search lives.

How to delete a Claude artifact, step by step

Vendor UIs move. The four operations below are stable; the exact menu label may differ slightly from what you see.

1. Remove a single artifact from a conversation

1. Open the conversation and select the artifact in the side panel.
2. Use the artifact's own menu (the overflow control in its header) and choose the delete or remove option.
3. If no delete control exists for that artifact type, delete the message that produced it — the artifact goes with its turn.

2. Unpublish or unshare a published artifact

1. In the artifact's share/publish panel, look at the sharing state. If it says anything other than private, a public copy exists.
2. Turn sharing off, or delete the published version explicitly. Do this before deleting the conversation.
3. Confirm by opening the shared URL in a private window. If it still renders, you unshared the wrong copy.

3. Delete the whole conversation

1. From the conversation list, open the item's menu and choose delete.
2. Deleting the conversation removes the artifacts inside it from your view.

4. Delete everything, including account-level history

Account settings has the bulk options: clear all conversations, or delete the account. If a client or auditor is asking about a specific document, do this last and document what you did first — bulk deletion destroys your own evidence of what was shared.

Why deletion in a chat tool is not data removal

Deleting an artifact removes your access path to it. That is not the same as removing the content from the world, and the gap has four predictable holes.

• Copies. Anyone who opened the link could save the page, screenshot it, or paste the HTML somewhere else. Deletion cannot reach a copy.
• Caches and CDNs. A published page may sit in an edge cache or a corporate proxy for a while after the origin stops serving it.
• Retention on the vendor side. "Deleted" usually means removed from your interface, with backend deletion following on the provider's own retention schedule. Read the terms; do not assume same-second erasure.
• Anyone who already read it. The most important one, and no product feature fixes it. Deletion at 11:52 does nothing about a page read at 9:14.

Deletion is a cleanup tool, not a containment tool. It reduces future exposure. It does not undo past exposure.

Designed-in expiry beats reactive deletion

Reactive deletion depends on a human remembering, at the right moment, that a document exists. Designed-in expiry does not depend on anyone. You decide the lifespan when you create the link, and the platform enforces it whether or not you ever think about that file again.

On HTMLvault this is two separate settings, and one ends access while the other ends existence:

• Link expiry ends access. The URL stops resolving on a date, after a duration, or once a view cap is reached. Anything 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. The window governs how long HTMLvault stores the underlying HTML and its analytics. Configurable from auto-delete through two years on Pro and above; Free retains 90 days. See the data retention window for how the auto-delete end of the scale behaves.

Expiry alone leaves a dead link with live content behind it. Retention alone leaves an open door until the window closes. Set both, and the deliverable removes itself: access closes on schedule, then the content leaves storage, and there is nothing left for anyone to have to delete later.

Worked example: the proposal that should never have needed deleting

Same document, same 11:47 p.m. energy, different setup. The rep asks Claude for a pricing page, then publishes it through HTMLvault's MCP server instead of the chat tool's own share control:

scan_html          → 1 flag: person (contact name in the footer)
create_link
  title            "Acme — Q3 pricing"
  expires_in       "14d"
  max_views        25
  retention        "30d"
  password         on
  channel          "email"

What that buys, without anyone remembering anything:

• The regex scanner flags the PII categories it knows — names, emails, phone numbers, financial strings, API keys — before the link exists. Zero tokens, no AI call. Teams and Enterprise can layer an AI scan on top using their own Anthropic, OpenAI, or Google key.
• Day 15, or view 26, the link is closed. No cleanup ticket.
• Day 30, the HTML leaves storage. There is no artifact to hunt down in a chat history six months later.
• Meanwhile the analytics answer the question deletion never could: who opened it, how far they scrolled, how long they stayed, and whether it got forwarded — repeat views from a new city is a forward.

Decision table: if the content contains X, set retention to Y

Defaults beat judgment calls at 11 p.m. Pick a row, set it once per content type, and let the platform enforce it.

• Named customer data, pricing, or a lead list → expiry 7–14 days, retention 30 days, password on. This is the row most sales and RevOps content belongs in.
• PII the scanner flagged and you chose to keep (a contact block, an address) → expiry 24–72 hours, retention auto-delete. Short-lived by design; see what PII privacy means.
• Anything containing a credential or token → do not publish it. Rotate the key, then republish clean. Scan HTML for API keys first, every time.
• Internal dashboards and reports → expiry 30 days, retention 90 days, so quarter-over-quarter comparisons survive.
• Public marketing pages and campaign landing pages → expiry never, retention 2 years. These are meant to be found; the analytics history is the asset.
• Training material → expiry never, retention 1–2 years, refreshed on a review cycle rather than deleted ad hoc.

Limits and caveats

• Nothing recalls a page someone already read. Expiry and retention shrink the window; they do not close it retroactively.
• Auto-delete deletes the analytics too. When content leaves storage, its view history goes with it. Export anything you need for reporting before the window closes.
• Expiry is not access control. A link that is public for fourteen days is public for fourteen days. Add a password, or read public links vs controlled access.
• The regex scanner catches patterns, not meaning. It reliably finds structured strings — SSNs, keys, emails, phone numbers. A sentence that is sensitive because of context is a human judgment call, or a job for the BYOK AI layer.
• Retention windows are per link, but defaults are the real control. Set them at the team level so nobody has to choose correctly under deadline pressure. Teams plans add custom roles and audit logs so compliance can verify the policy is actually being followed.
• You still delete things sometimes. delete_link exists for the genuine mistake. It should be the exception, not the workflow.

Why this matters

The reason "how do I delete a Claude artifact" is such a tense search is that it is almost always asked too late. The person typing it already knows the answer they want does not exist.

The rep who sets a fourteen-day expiry and a thirty-day retention window at creation never types that search. Neither does the marketer whose campaign preview closed itself the day the campaign shipped, or the RevOps analyst whose lead list stopped existing before the next quarterly review. They are not more disciplined than everyone else — they just moved the cleanup decision from the panicked end of the process to the two seconds where it costs nothing. That is a capability, not a habit, and it is the difference between hoping nobody forwarded the file and knowing the file is gone.