Claude Artifacts (2026): What They Are, Who Can Open Them, and the Sharing Rules That Bite a Business

The short answer

A Claude artifact is anything Claude makes that you would put in front of someone else: a document, a deck, a design, a dashboard, or a small working app. It opens beside your conversation, you can edit it, and you can share it with a link. Artifacts work on every plan, including Free. As of September 2026 the rule that matters most is the one almost nobody reads: in the new Claude experience, everyone who opens your artifact needs a Claude account, and an AI-powered artifact runs on the viewer's usage, not yours.

That one rule decides what artifacts are good for. Inside a team where everyone has a seat, they are the fastest internal-tool builder most companies will ever get. Pointed at customers, they hit a wall. Everything below is from Anthropic's own help center, read September 23, 2026.

What an artifact actually is

Anthropic's definition is refreshingly plain: an artifact is "anything Claude makes for you that you'd put in front of someone." Claude decides to make one when the output is significant and self-contained (typically over 15 lines), something you will want to edit or reuse, and something that stands on its own without the chat around it.

In practice that means six kinds of thing:

• Documents in Markdown or plain text: briefs, SOPs, proposals.
• Code snippets you will copy somewhere else.
• Single-page HTML sites: a landing page draft, a one-page report.
• SVG images and diagrams or flowcharts.
• Interactive React components: calculators, trackers, dashboards, quizzes.

On top of that sit three branded tools that make a specific kind of artifact: Claude Design for visuals, prototypes and landing pages built from your design system, Claude Slides for presentations that export to PowerPoint or PDF, and Claude Docs for living documents your team writes together. Those three are in beta and on paid plans only. A plain artifact is not.

Which plans get what

Two setup details trip people up. Artifacts require "Code execution and file creation" to be switched on under Settings, then Capabilities (or Organization settings on Team and Enterprise); Anthropic no longer supports artifacts without it. And artifacts count toward your plan's usage limits like any other work, which is covered in my Claude usage limits breakdown.

How sharing actually works: the account rule

This is the section the product pages skip, and it is the one that decides whether an artifact can do the job you have in mind.

Artifacts shared from the Share dialog need a Claude account to open, "even if they have the link." That covers everything made in the new Claude experience, plus everything from Claude Design, Slides, Docs and Claude Code. There is one exception left over from the previous chat experience: an artifact published from chat on a Free, Pro or Max plan opens without an account for basic use, and only asks the visitor to sign up when they touch an AI feature. As the new experience rolls out, that exception covers less and less of what people make.

Then the usage rule. When you share an AI-powered artifact, one that asks Claude questions as people use it, Anthropic says there are "no costs to you" and no API key required, because "usage counts against each user's own Claude subscription limits, not yours."

Put the two together and you get the practical rule: an artifact is a tool for people who already have Claude. Inside a Team plan, that is perfect. Every seat can open it, every seat pays for its own usage out of an allowance the company already bought, and nobody wires up a server. Pointed at customers, it inverts: each customer needs an account, and each one spends their own plan to use your tool. If you are picturing a client portal or a public calculator, you are picturing a product, and an artifact is a prototype of it.

What the limits and storage rules do not tell you

Persistent storage is what turns an artifact from a page into an app: a tracker that remembers, a shared leaderboard, a team log. The published spec is 20 MB per artifact, text only, with no images, files or binary data. Storage only works once the artifact is published, so it will not save anything while you are still testing it. It can be personal, where each user sees only their own entries, or shared, where everyone sees everything.

Three consequences are not on the spec sheet:

• Unpublishing is a delete button. It permanently removes all stored data, personal and shared, and you cannot publish that same artifact again. A tracker a team has used for three months is one click from gone, so export first.
• Sharing a chat artifact shares the chat's files. Viewers also get the attachments and files from the conversation that made it. Build the client-facing version in a clean conversation, not the one where you pasted the contract.
• Connected apps are per person. An artifact that reads Slack or Google Calendar makes every user authenticate on their own, and admins can switch the feature on or off but cannot choose which connections an artifact uses.

What a small team should actually build with them

The pattern that fits those rules is boring and very effective: small internal tools that one person used to rebuild in a spreadsheet every week.

• A quote or pricing calculator your sales team shares inside the organization.
• A one-page status dashboard generated from a pasted export, rebuilt Monday morning.
• An onboarding checklist with personal storage, so each new hire tracks their own progress.
• A proposal or brief template as a Markdown artifact, edited in place with "Edit with Claude."

What not to build: anything a customer must use, anything that needs more than 20 MB of text, and anything whose data you cannot afford to lose to one unpublish. Those are real software, and the honest move is to prototype in an artifact and then build it properly.

The owner's version

If your team is on Free, artifacts already work, but the three tools most people actually want (Design, Slides, Docs) and storage do not. Pro is $17 a month on annual billing ($20 monthly) per person; a Team plan starts at $25 per seat monthly for teams of 2 to 150, and it is the plan where sharing inside the organization and governing it actually make sense. The full plan comparison is on my Claude AI pricing page.

If you want your team building these internal tools in the first week instead of the first quarter, with sharing set up so client files stay inside the organization, that is part of what my Claude for Small Business install covers.