Artifact
In chat products, an artifact is a deliverable the AI produces outside the message stream—code, documents, diagrams, mini-apps—users edit, export, and version independently.
Artifacts are not diffusion noise or glitches; they are first-class outputs with their own panel, lifecycle, and sync back to the conversation.
What it means
The product renders structured output (HTML, React, Markdown doc, Mermaid chart) in a dedicated surface with edit, copy, download, and history—not only inline chat text.
Why designers should care
Artifact UX drives trust for serious work: version dropdown, diff from last edit, export to Figma/code repo, and clear “artifact updated” cues in the thread.
Example
ChatGPT creates a PRD artifact in a side panel; the user edits headings in-place, asks the model to “add risks section,” and exports Markdown to Notion without losing chat context.
Common mistakes
- • Treating long chat replies as artifacts with no separate edit or export surface.
- • No version history when agent edits overwrite a working draft.
- • Confusing users by using “artifact” for low-stakes inline code blocks only.