Live on Product Hunt Intraface launched today. Less steering, more vacations. See the launch
Start for free

What is an AI artifact?

An AI artifact is a self-contained piece of work an AI assistant makes for you and shows beside the conversation, not inside it: a document, a web page, a small app or a diagram. You iterate on it in place, and you can publish, share or export it on its own, so it outlives the chat that made it.

Part of Glossary: the terms behind MCP, coding agents and shared pages, defined plainly.

How it works

You ask in the chat, the assistant builds the artifact in its own panel, you edit it there, and a link or export carries it out of the chat.the conversationthe thing it madeyouassistantClaude, ChatGPT, Geminithe artifactdoc, page, applink or exportviewers, copies, a board1. asks for something2. builds it beside chat3. edits in place4. publish, export
  1. You ask for something substantial: a page, a report, a quiz, a tiny app. The assistant treats it as an artifact, not a reply, and opens it in its own panel.

  2. The artifact is self-contained. You read it, preview it or run it there while the chat continues alongside.

  3. You iterate: ask for a change in chat, highlight a passage and describe the edit, or type into it directly. Earlier versions stay reachable.

  4. When it is done you publish a link, export a file, or copy the content elsewhere. The artifact can be opened without the chat that produced it.

  5. To keep editing it with other people, it needs a home outside the chat: a copy each, a shared document, or a shared board that agents can edit.

    A link usually grants viewing. Editing by others is a separate setting, and often a separate copy.

Source: Claude Help Center: What are Artifacts and how do I use them?

Example: the same idea in three assistants

Anthropic defines an artifact as anything Claude makes for you that you would put in front of someone: a design, a deck, a document, a dashboard, a small interactive tool. It opens in its own window beside the conversation. Template artifacts (docs, slides, designs) can be typed into directly, and a doc lets you highlight text and pick Edit with Claude to aim a change at one paragraph. Exports go to Word, PDF, Markdown or Google Docs for documents, PowerPoint or Google Slides for decks, and standalone HTML for designs.

ChatGPT's canvas is the same shape for writing and code: a side-by-side workspace you open by asking ChatGPT to use canvas, with a shortcut menu for reading level, a final polish, bug fixes and ports to other languages. Some of those shortcuts edit the project in place.

Gemini's Canvas lets you create or edit a doc, app, slides or code, preview an app with a console for errors, browse saved versions, export text to Google Docs and Python to Colab, and share a public link that anyone can view and copy.

Why it matters

A chat is a transcript; an artifact is a product. The moment a model's output is something you would hand to someone else, you want to look at it whole, edit one part without regenerating the rest, keep its history, and send it on. Every major assistant has converged on the same answer because the alternative, pasting long replies into other tools, loses the thread between the thing and the conversation that shaped it.

The second reason is iteration. Highlight a sentence and ask for a change, and only that sentence changes. Without an artifact, each request risks a full rewrite.

AI artifact versus a chat reply

A reply is read once in context. An artifact is referred back to, edited and shared. That changes what good looks like: the artifact should stand alone, carry its own title, and make sense to someone who never saw the chat. Assistants decide between the two for you, and you can override by asking for one.

After the chat: how artifacts get shared and kept editable

Sharing is solved. Claude publishes a public link that anyone holding it can open, or shares with named people at Can view, Commenter or Can edit access. Gemini shares a g.co link. Both offer a copy to the viewer.

Co-editing the same original is where the chat tools stop. A copy each means versions drift; a Google Doc export keeps the text but drops the app. If several people's coding agents need to keep working on one page, a shared board that each agent pulls from and publishes to is one way to give the artifact a home with a version history.

Common mistakes

  • Treating the chat as the home of the artifact. Chats get long, branched and deleted. Publish or export the artifact when it matters.
  • Assuming a link means editing. A public link is almost always view-only. Check the access level before telling a teammate to make changes.
  • Forgetting what the artifact stores. Interactive Claude artifacts can keep shared storage that every viewer sees, and connected-app data uses each viewer's own connections. Read the article's warnings before putting anything sensitive in one.

FAQ

What counts as an AI artifact?
Anthropic's rule of thumb for Claude: content that is substantial, typically over 15 lines, self-contained, likely to be edited or reused, and worth referring back to. A short answer stays in the chat; a document, a single-page site, a diagram or a small interactive tool becomes an artifact.
Can other people edit an AI artifact after I share it?
Usually they can view it, and sometimes copy it. Claude's Share dialog has Can view, Commenter and Can edit levels; a published Claude artifact can be copied by signed-in users; a shared Gemini Canvas has a Copy Canvas button that makes the viewer their own editable copy. Editing the same original together needs a shared workspace.
Is an AI artifact the same as a canvas?
Same idea, different names. Anthropic calls them Artifacts; OpenAI and Google both call their version Canvas. All three open a panel beside the chat holding one document or one piece of code that you refine in place.

Stop sending HTMLover Slack.

Create an account in a minute, connect your coding agent, and make your first board.