What Are Claude Artifacts? claude.ai vs Claude Code vs Cowork, Explained (2026)
By Marcin, founder of Markloop · Updated July 2026
Definition: a Claude artifact is a self-contained piece of content - most often a single interactive HTML page - that Claude creates and displays outside the chat or terminal, at its own URL or panel, so you can view it, interact with it, and come back to it as Claude updates it. In 2026 artifacts exist in three places: claude.ai chat (since 2024), Claude Code (since June 2026), and Claude Cowork (live artifacts, local to your device). Claude Code artifacts and many chat artifacts are ordinary self-contained HTML - which is their superpower: the file works anywhere.
The three kinds of artifacts
| claude.ai chat artifacts | Claude Code artifacts | Cowork live artifacts | |
|---|---|---|---|
| Where they're made | A chat on claude.ai | A coding session (CLI or desktop app) | A Cowork session |
| What they are | Interactive panel next to the chat: mini-apps, documents, visualizations | A published web page at a private claude.ai URL, updating as the session works | Local pages that refresh with current data from your connected apps each time you open them |
| Who can see them | You; Free/Pro/Max can publish a public link; Team/Enterprise share in-org only | You; on Team/Enterprise, people in your org you share with. No public option. | Only you - live artifacts are local to your device; sharing is on Anthropic's roadmap (as of July 2026) |
| Feedback / comments | None | None - "viewable, not co-edited" | None |
The common thread: an artifact is Claude switching from telling you something in prose to showing you something as an interface - a dashboard from your data, a PR walkthrough with annotated diffs, a spec with diagrams, a set of design options side by side.
What artifacts are genuinely great at
- Rich output where text falls short. Anthropic's own docs recommend artifacts for "output that is easier to look at and interact with than to read line by line" - dashboards, diffs, comparisons, timelines.
- Living pages. A Claude Code artifact republishes to the same URL as the session progresses - anyone with the page open sees it update in place.
- Self-containment. A strict Content Security Policy forces inline CSS/JS and embedded images. That constraint is a gift: the underlying HTML file renders identically anywhere, forever.
This is one face of a bigger 2026 shift - Anthropic engineers now write plans and specs as HTML instead of Markdown because people engage with them better. We cover that thesis (and what it means for review) in HTML is the new Markdown.
The limits people hit
- The org wall. Claude Code artifacts cannot be viewed by anyone outside your claude.ai organization - no public link exists, and Pro/Max artifacts are private to you alone. The docs' own workaround is to export the HTML file. (Full guide: how to share an artifact outside your org.)
- No feedback channel. Viewers can look, not comment. The moment a teammate or client has an opinion, the conversation leaves the artifact - Slack, email, screenshots.
- No backend, one page. An artifact "is a capture of work, not an application": no form storage, no API calls at view time, no multiple routes, 16 MiB rendered-size cap.
- Retention. Org admins can auto-delete artifacts on a schedule, so an artifact URL is not a durable deliverable.
When the artifact needs a second life
An artifact usually starts as something you look at - and becomes something someone must respond to: a spec a client should approve, a report an expert should correct, a plan with open questions. That's the point where the artifact format (self-contained HTML) matters more than the artifact URL: take the file, put it where people can mark it up.
That's what Markloop does - it's a web app for reviewing agent-made HTML documents. Claude Code pushes the file over MCP (or you upload it), reviewers comment on the exact element and answer your embedded questions in the browser - invited by email or via a share link with comments turned on (no login, no Claude account, always free), and your agent pulls everything back as structured context and publishes the next version.
FAQ
Are Claude artifacts free?
Artifacts in claude.ai chat exist on all plans. Claude Code artifacts require a Pro, Max, Team, or Enterprise plan and a claude.ai login in the session.
Are Claude artifacts public?
Chat artifacts on Free/Pro/Max can be published to a public link. Claude Code artifacts are never public - private to you on Pro/Max, org-only on Team/Enterprise.
Can people comment on a Claude artifact?
No. Artifacts have no commenting or feedback mechanism on any plan - they are "viewable, not co-edited". To collect comments on one, export the HTML file and use a review tool such as Markloop.
How do I get the HTML file of an artifact?
In Claude Code, the artifact starts as an HTML or Markdown file in your project directory - if it's Markdown, or you're not sure, just ask: "Save the artifact as a standalone HTML file." In claude.ai chat, ask Claude to provide the artifact as a downloadable HTML file.
What's the difference between skills and artifacts in Claude?
They sit on opposite ends of the work: a skill is a reusable instruction package that changes how Claude does a task; an artifact is the output - the page, document, or mini-app Claude produces. A skill can produce artifacts; an artifact is never a skill.
What does it mean to publish a Claude artifact?
In claude.ai chat (Free/Pro/Max), publishing creates a public link to your artifact that anyone can open. In Claude Code, "publish" means pushing the page to its private claude.ai URL - it never makes it public.
What's the difference between an artifact and Canvas (ChatGPT)?
Same idea, different ecosystems: both are side-by-side workspaces for content the model produces. Artifacts lean on self-contained HTML pages; Canvas is OpenAI's equivalent for documents and code in ChatGPT.
Got an artifact that needs opinions, not just views?
Markloop turns the same HTML file into a commentable, versioned document - and sends the feedback back to Claude Code over MCP.
Start free - 14-day trial, no card required →Reviewers are always free.
Related: How to share a Claude Code artifact outside your organization · Artifact sharing options compared