← All postsWhat Are Claude Artifacts and What Can You Build With Them?

What Are Claude Artifacts and What Can You Build With Them?

Carlos GarciaCarlos GarciaPublished

Ask Claude for something substantial and it does not always answer in the chat. Sometimes a panel opens beside the conversation with a document, a dashboard or a small working app in it, and the chat becomes the place where you ask for changes rather than the place the work lives.

That panel is an artifact, and it has quietly become one of the most consequential things about using Claude. It is also the feature most people describe inaccurately, partly because it changed substantially in 2026 and most write-ups have not caught up.

This guide works from Anthropic's own documentation and product pages. Where those pages disagree with each other — and on one point they do — this guide says so instead of picking whichever answer reads better.

What Are Claude Artifacts?

Anthropic's own definition is refreshingly plain: an artifact is "anything Claude makes for you that you'd put in front of someone." It opens beside the conversation, you can edit it, come back to it later and share it.

The product page expands the list without changing the idea: "a design, an animation, a deck, a document, a dashboard, or a small interactive tool." The common thread is that an artifact is a thing rather than a message. A chat reply is something you read once; an artifact is something you keep, revise and hand to somebody.

Mechanically, that distinction is what Claude is deciding when it chooses whether to open the panel. Anthropic documents four criteria for content becoming an artifact: it is "significant and self-contained, typically over 15 lines"; you are "likely to want to edit, iterate on, or reuse outside the conversation"; it "stands on its own without needing extra context from the conversation"; and it is "content you're likely to refer back to or use later."

You do not have to wait for Claude to decide. Asking for an artifact directly works, and in Claude Code the documentation says explicitly that Claude "may publish an artifact on its own when the output suits a page, or you can ask for one directly."

A dashboard nobody acts on is just a prettier spreadsheet. Get a free audit and find out which of your analytics and search metrics actually move the business.

Free audit

Where do you stand in AI search?

A manual SEO + AI search audit of your site: how ChatGPT, Gemini, Perplexity and Claude read and cite it today, and what to fix first. Report within 48 hours, no credit card.

Get my free audit

Or have it all done: AI SEO services, $999 once

What Can Actually Become an Artifact?

There are two answers, because Anthropic ships two different things under the same word.

Artifacts Claude Makes in a Conversation

For ordinary chat artifacts, Anthropic says Claude "can make an artifact out of almost anything," and then enumerates:

  • documents
  • code snippets
  • single-page websites
  • images
  • diagrams and flowcharts
  • dashboards
  • small interactive tools

Templates You Start From Deliberately

Separately, Anthropic now offers named templates you choose up front rather than discover. The admin documentation lists them as Slides, Design, Design systems, Docs, Data, Motion in beta, Sheets in beta and Whiteboards in beta. The product page also names Claude Dashboards, described as being in beta on paid plans.

This is the part that has moved fastest, and it is why older explanations of artifacts feel thin. The feature started as a code-and-document panel. It is now closer to a small set of editors — a deck editor, a document editor, a design canvas — that happen to be driven by a conversation.

What Artifacts Are Not

Two things commonly listed as artifact types are not, on Anthropic's own pages, artifact types at all. SVG appears only as an upload format with a size limit, and React is mentioned as a library an artifact page may load from an approved CDN, not as a kind of artifact. If you want a diagram, you ask for a diagram and Anthropic's guidance suggests SVG as the sensible internal format.

How Do You Create, Edit and Share One?

Creating one is mostly a matter of asking for something substantial enough to qualify, or picking a template.

Editing has three documented routes. You can ask Claude in the conversation. You can edit directly on the canvas in a template, including from a phone. And inside a Doc, you can highlight text and use "Edit with Claude" to work on that passage alone. When something breaks, there is a "Try fixing with Claude" button, which Anthropic describes with admirable honesty: "success isn't guaranteed."

Version history is where expectations and reality part company. For chat and template artifacts, Anthropic's pages describe no version history. What they describe instead is conversation branching: "Edit an earlier message to create a different version of the chat, with its own artifacts." That gives you alternatives, not a timeline.

Claude Code artifacts do have versions. There, Claude edits the underlying file and republishes to the same URL, "each publish becomes a version," and the Share control lets you choose which version viewers see. A stale publish is refused rather than silently overwriting someone else's newer one.

Sharing starts closed. "Artifacts start private to you," and the product page puts it as "only you, until you share one." On Free, Pro and Max you can then "share with anyone who has the link." On Team and Enterprise, links stay inside the organization unless an administrator turns external sharing on, and that setting is off by default.

One practical detail that surprises people: viewers usually do not need an account. Anthropic's admin guide says "Most artifacts shared this way open without a Claude account or sign-in." Docs are the exception, opening "only for people signed in to Claude, who can only view them."

Shipping faster is not the same as shipping the right thing. Get a free audit and see where your analytics and search data are pointing you wrong.

Artifacts That Run Claude Themselves

This is the capability that changes what artifacts are for, and it is widely misunderstood.

An artifact can call Claude. Anthropic's wording: "You can build artifacts that call Claude directly, turning them into small apps." So the thing you share is not necessarily a static page — it can be a tool that reasons when somebody uses it.

The billing arrangement is the interesting part, and it is the opposite of what most people assume. "People using it sign in with their Claude account and interact with their own instance." There are "no API keys required, and there's no cost to you," because "usage counts against each person's own plan limits rather than yours."

Read that again if you build things for other people. You can hand out a working AI tool without metering, without keys and without paying for strangers' usage. Inside an organization, Anthropic notes that colleagues "can use what you share without adding cost for you."

Permission is explicit rather than assumed: "New artifacts ask you for permission the first time they want to use Claude." And there is a documented boundary — connector tools that require approval for each individual action are not available to artifacts, which rules out a category of automation some people will reach for first.

Which Plans Include Claude Artifacts?

The headline answer is that artifacts are broadly available: "Artifacts are available on Free, Pro, Max, Ultra, Team, and Enterprise plans, and in Claude Code on every plan that includes Claude Code."

Free is genuinely included, not nominally. Anthropic's plan table gives Free users creating artifacts in a chat and starting from a Docs, Slides or Design template, and the product page states that "Claude Design, Claude Slides, and Claude Docs are included within every plan, including the Free plan."

What Free does not get, per that same table, is the Data template, Sheets, Whiteboards, connected apps and stored data. Motion is limited to Team and Enterprise, and Dashboards is described as beta on paid plans.

There is one genuine conflict in the documentation worth knowing before you plan around it. The Claude Code artifacts page states that artifacts there "are available on Pro, Max, Team, and Enterprise plans" and does not list Free. The most coherent reading is that availability is per surface — chat artifacts on every plan, Claude Code artifacts from Pro upward — but Anthropic does not reconcile the two statements, so verify on the pricing page before committing to a workflow on a free account.

What Are the Limitations Worth Knowing?

Several, and the documented ones are more specific than the vague warnings you usually see.

The September 2026 Legacy Cutover

This is the fact most coverage of artifacts is missing. Anthropic defines legacy artifacts as "artifacts made in a chat before September 16, 2026," and the consequence is blunt: "They keep working, and you can still publish and share them, but you can't make new ones."

Legacy artifacts also carry worse limits — storage capped at 20 MB and text only — and one genuinely destructive behavior: unpublishing a legacy artifact "permanently deletes all its stored data." If you have an old published artifact holding anything you care about, unpublishing it is not a reversible tidy-up.

A related cutover: artifacts made in Claude Cowork before August 19, 2026 are live artifacts, which "keep working, but you can't edit them in place." And the standalone site at claude.ai/design "closes on December 14, 2026," with Anthropic directing people to move across.

Storage and File Limits

Current artifacts get, by default, "100 MB of data and 25,000 documents," whichever is reached first. Uploaded files are capped at 20 MB each, with a lower 2 MB ceiling for SVG, up to 5,000 files and 1 GB in total.

What a Claude Code Artifact Cannot Do

The Claude Code constraints are the tightest and the most concrete. An artifact there is "a single HTML page" with no backend, which means it cannot authenticate viewers itself. The source must be .html, .htm or .md. The rendered page is capped at 16 MiB. External requests are limited to an approved CDN list, relative links do not resolve, and the page cannot start a download on its own.

It also will not run everywhere. Artifacts are unavailable through Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry, API-key sessions cannot publish at all, and publishing is off by default in the Agent SDK, the GitHub Action and non-interactive mode.

Limits Anthropic Does Not State

Worth saying plainly, because invented numbers circulate. Anthropic's pages give no cap on how many artifacts a person or organization may have, and no rate limit or call count for AI-powered artifacts. The nearest documented throttle is narrow and specific: automatic replies to comments stop after 60 sent comments or thread activations per artifact per hour.

Measuring the right things beats building faster. Get a free audit and let us check what your analytics are really telling you.

Claude Artifacts vs ChatGPT Canvas and Gemini Canvas

All three vendors now put generated work in a panel beside the chat, and at the level of "a document you can revise" they are interchangeable. The differences are in what the panel is allowed to become.

For writing and editing prose, the three are close enough that the better tool is whichever assistant you prefer talking to. A side panel with versioned text is table stakes in 2026, not a differentiator.

For building something interactive, Claude Artifacts wins clearly, and the AI-powered-artifact billing model is why. The ability to publish a tool that calls Claude on the viewer's own plan, with no keys and no cost to the builder, has no equivalent in the other two. If you want to hand a working AI tool to a team or a client, this is the one.

For organizational control, Claude Artifacts also wins, because the controls actually exist and are documented. Administrators enable cloud code execution and file creation first, then Artifacts, with per-template defaults, external sharing off by default, and the ability to scope capabilities through custom roles on Enterprise. Some artifacts cannot leave the organization at all even with sharing enabled — those that connect to other websites, and those carrying an uploaded file that failed or skipped checking.

Where Claude loses is the thing nobody advertises: the legacy cutover. Anybody who built chat artifacts before September 16, 2026 has a set of things they can share but not recreate, and that is a sharper discontinuity than either competitor has imposed.

Build the dashboard, by all means — but build it on numbers that matter. Get a free audit and start from the metrics that count.

Final Thoughts

An artifact is Claude's answer to a simple problem: some outputs are not messages. A deck, a dashboard, a working tool and a contract all want to live somewhere you can return to, revise and send on, and a scrolling transcript is a poor home for any of them.

The two details worth carrying away are the ones least often reported. Artifacts that call Claude bill the person using them, not the person who built them, which makes sharing a real tool practical rather than expensive. And the September 16, 2026 line means old chat artifacts are frozen in a way that matters if you are still relying on them.

If you want the same idea as the competition implements it, our explainer on ChatGPT Canvas covers the panel model from the OpenAI side, including where it is stronger and where it stops.

Beyond that, the honest advice is to stop treating the panel as a novelty. If you ask Claude for things you intend to keep, you are already working in artifacts, and the only question left is whether you are using the version with the better limits.

Want this done for you?

SEO Stuff gets businesses cited by ChatGPT, Gemini, Perplexity and Claude, and ranking in Google. Start with the free audit, a call, or the package.

See where you stand

A free, manual SEO + AI search audit of your site, with a report within 48 hours. No credit card, no sales call.

Get my free audit

Talk it through

A free 15-minute call with the founder about whether AI search is a meaningful opportunity for your business.

Free strategy call

Have it done

The Done-For-You Package: audit, 10 pages of content, 3 DR50+ placements, dashboard. $999 once, delivered in 21 business days.

See the package