Claude's artifacts teardown
Updated June 14, 2026
When Claude generates code, a document, or an interactive tool, the reply includes an artifact card first. Opening it expands into a split pane: conversation on the left, deliverable on the right.
Users iterate via follow-up prompts while the artifact stays open, switch between Preview and Code, and optionally publish a public link while the chat stays private.
Artifacts announced in the thread

What works
- The artifact card names the deliverable and labels it “Interactive artifact” so users know something runnable is attached.
- Claude still explains features in prose above the card, the thread documents intent; the card is the entry point.
- Opening the card expands to the split pane without losing the chat history on the left.
What we would push on
- Card and bullet list repeat the same information. Consider collapsing the list once the artifact is open.
Takeaway
Treat the in-thread card as a preview and handoff, not the final workspace.
Pattern: Chat Artifacts
Preview mode in the split pane

What works
- Conversation and deliverable stay visible together. Users never hunt for “where did that code go?”
- Preview tab shows the running UI; Copy and Publish sit in the artifact chrome, separate from chat actions.
- The composer remains at the bottom so follow-ups (“make the buttons bigger”) happen in context.
What we would push on
- On smaller viewports the split pane may feel cramped. Mobile likely needs a full-screen artifact mode.
Takeaway
Split pane is Claude’s core artifact pattern: thread as reasoning trail, right pane as the product of the session.
Pattern: Chat Artifacts
Code mode in the same pane

What works
- Preview and Code are two modes of one pane, not separate destinations. Users flip tabs on the calculator artifact above.
- Syntax highlighting and component structure are readable without exporting to an IDE first.
- Copy in the artifact header is scoped to the code block, not the whole chat transcript.
What we would push on
- No inline edit in the Code tab in this flow, mechanical fixes still require a follow-up prompt or external editor.
Takeaway
Pair Preview and Code as adjacent tabs. Developers verify in Preview, inspect and copy in Code.
Pattern: Smart Code Blocks
Pattern: Chat Artifacts
Interactive outputs you can use

What works
- Artifacts can be full interactive apps: sliders update the chart and summary cards in real time.
- Educational prompts (compound interest) become explorable tools instead of walls of markdown.
- The artifact title matches the lesson topic so returning users recognize what they built.
What we would push on
- Full-width artifact mode hides the chat entirely. Users lose conversational context until they dismiss or resize.
Takeaway
Artifacts are workspace objects, not paste-only blocks. Steal the interactive preview when your outputs need hands-on verification.
Pattern: Chat Artifacts
Pattern: Generative UI
Publish outside the chat

What works
- Publish is explicit, modal copy states the artifact becomes internet-accessible and may appear in search.
- Chat privacy is called out separately so users know only the artifact link is shared, not the full conversation.
- Usage policy reminder appears before Publish & copy link, appropriate friction for public sharing.
What we would push on
- Publish sits next to Copy with similar visual weight. Accidental public sharing is a real risk for enterprise users.
Takeaway
Sharing is a distinct mode from iterating. Separate publish from copy and spell out what becomes public.
Pattern: Chat Artifacts
How artifacts fit with output
Artifacts are the second act: card in the reply → split pane → Preview and Code tabs → interactive examples → optional Publish.
In-thread refinement (Try again, version navigation, edit-to-branch) still applies to the conversation while the artifact is open. The output teardown covers that half of the loop.
The pattern to steal: treat the thread as context and the artifact as the deliverable users copy, run, or ship.
Steal this
- In-thread artifact card as handoff before the workspace opens
- Split pane so chat context and deliverable stay visible together
- Preview/Code toggle for interactive and shippable outputs
- Publish flow that separates public artifact from private chat
Skip this
- Burying long code outputs only in the message stream
- Side pane for answers that fit in a single paragraph
- Publish buttons with the same weight as Copy
How others output, artifacts & refinement
Same job, different product bets, and what each tradeoff reveals.
Short answers and Try again stay in the thread before work promotes to an artifact.
Read teardownChatGPT keeps images and refinement inside the thread, no persistent side-pane workspace.
Read teardownPerplexity optimizes cited answers and source panels, not artifact workspaces.
Read teardownOriginal gallery pages: Artifacts