What Claude Artifacts Actually Are, and When to Use One

You ask Claude for something, and instead of the answer appearing in the chat, a panel slides open on the right with the thing sitting inside it. A document. A table. A little calculator that actually works.

If you've seen that happen and quietly wondered what it was, you're in good company. Artifacts are one of the most useful things in Claude and one of the least explained.

So let's fix that.

An artifact is something you own, instead of a message you scroll past

Anthropic describes artifacts as "substantial, standalone content" that appears in "a dedicated window separate from the main conversation". They exist so you can "turn ideas into shareable apps, tools, or content".

That's the official version. The practical version is simpler.

A chat reply is a message. You read it, you scroll, and three days later you can't find it. An artifact is a thing. It has its own window, its own version history and its own life outside the conversation that produced it.

That distinction is the whole point. It's the difference between asking Claude a question and walking away with an asset.

Anthropic reported more than half a billion artifacts created since launch, back in June 2025. It's not a side experiment. It's one of the main ways people get work out of Claude.

What makes Claude open one

Claude decides on its own, using a few rules of thumb. According to Anthropic's documentation, it creates an artifact when the content is:

• Significant and self-contained, typically over 15 lines
• Something you'd want to edit, iterate on, or reuse outside the conversation
• A complex piece of content that stands on its own without needing extra conversation context
• Something you might refer back to or use later

You don't have to ask for one, though you can. If Claude gives you something in the chat that you'd rather have as a proper document, say so. "Put that in an artifact" works fine.

What can go in an artifact

Anthropic lists these as the common types:

That last row is the one worth pausing on. Interactive means it does something. A pricing calculator your team can use. A quiz that scores itself. A dashboard that sorts and filters.

You don't need to know what React is to ask for one. You describe what you want it to do, and Claude builds it.

Artifacts don't save themselves

This one catches people out, and it's worth knowing before you lose something good.

Artifacts don't automatically appear in your Artifacts sidebar. Anthropic's documentation is direct about it: to add an artifact to your Artifacts section, open it and click Publish.

So if you build something useful and just close the tab, it stays buried in that conversation. Publish it, and it lives in your collection where you can find it again.

Publishing is also what switches on persistent storage, which is the bit that lets an artifact remember data between visits. Unpublish it later and that stored data is deleted.

Worth knowing: publishing an artifact to your own collection and sharing it with other people are two separate actions. More on sharing in a moment.

While you're there, two other things are worth knowing:

• You can switch between earlier versions using the version selector, so a bad edit isn't permanent
• For Markdown documents, Edit with Claude lets you highlight a section and say what you want changed, instead of re-describing the whole document

Sharing one, and who pays for it

This is where artifacts get interesting for business.

You can share an artifact with a link. No hosting, no deployment, no developer. Anthropic announced in July 2025 that people can build, host and share interactive AI-powered apps directly in the Claude app.

And the billing works in your favour. When someone uses an AI-powered artifact you've shared, they sign in with their own Claude account, and their usage counts against their subscription, not yours.

Think about what that means for a client-facing tool. You build a calculator once. A hundred people use it. You're not paying for their usage.

Artifact sidebar access and Claude-powered artifacts are supported on Free, Pro, Max, Team and Enterprise plans.

Persistent storage is narrower. It works on Pro, Max, Team and Enterprise, on Claude web and desktop, only once an artifact is published, with a 20 MB storage limit per artifact and text-only data. No images, no files, no binary.

You'll also need code execution and file creation enabled. That sits under Settings, then Capabilities on individual plans, or in Organization settings on Team and Enterprise.

Before you share anything with business information in it

A real thing happened in July 2026 that's worth knowing about.

Over the weekend of 25 and 26 July, Reddit users discovered that shared Claude conversations and artifacts were publicly searchable on Google. TechCrunch reported that Futurism found chats containing medical reports, internal company documents and employee reviews with personal information.

Anthropic's explanation, through spokeswoman Amie Rotherham, was that share links only turn up in search results when they've been posted somewhere a search engine can crawl, like a forum or a social media post. A link sent privately stays out of search. Her words: "These shareable links are not guessable or discoverable unless people choose to share them themselves." And: "When someone shares a conversation, they are making that content publicly accessible, and like other public web content, it may be archived by third-party services."

TechCrunch pushed back on that, pointing out that Claude's own interface warns only that "Anyone with the link can view," which reads like a feature for sending something to a colleague rather than publishing it to the internet. By Monday afternoon a test search by TechCrunch no longer returned results.

I'm not raising this to make you nervous. Creating an artifact isn't the risk. Sharing one is the decision that can put content on the open web.

A shared link is a public decision. Treat it the way you'd treat anything you post on the internet. Client names, pricing you haven't published, anything commercially sensitive, keep it out of an artifact you intend to share.

You can review what you've already shared under Settings, then Privacy, then Shared Chats. Worth five minutes.

When an artifact is the wrong tool

An artifact is the wrong choice when:

• You just need an answer. A short reply belongs in the chat. Forcing it into a window adds a step and gives you nothing.
• The content will change constantly. If something needs updating weekly by several people, it wants a proper system, not an artifact.
• It needs to live inside your existing tools. An artifact lives in Claude. If the output needs to sit in your CRM or your project management tool, build it there.
• The information is confidential and needs controlled access. Sharing an artifact publicly is a link, not a permissions system. On Team and Enterprise you can share inside your organisation instead, but a public link is open to anyone who has it.

Knowing when not to use something is usually worth more than knowing how to use it.

What decides whether your artifact is any good

Now for the part that matters more than any of the mechanics above.

An artifact is only as good as what Claude knows when it builds one.

Ask Claude to build you a pricing calculator with no context, and you'll get a generic pricing calculator. Perfectly functional. Completely detached from how you actually price, who you actually sell to and what your margins actually look like.

Give Claude your real business context first, your offers, your audience, your numbers, your positioning, and the same request produces something you could hand to a client that afternoon.

It's the same as briefing a contractor. A good one with a vague brief gives you competent work that misses. A good one with a proper brief gives you exactly what you needed.

That's why my method is Extract, Organise, Automate. AI works best when your business knowledge is current, clear and captured in one place.

The artifact is the output. The context is the reason it's any good.

Where to start this week

Pick one thing you rebuild by hand every month. A quote template. A capacity planner. A checklist you rewrite for every client.

Open Claude. Describe what it needs to do and who uses it. Let it build the artifact.

Then click Publish, so you still have it next month.

That's the whole loop. One useful thing, saved where you can find it, instead of another conversation you'll never scroll back through.

If you want the wider picture on what else Claude can do for your business, I've written a longer piece on ten Claude features worth setting up properly.

And if you'd rather work out where AI actually fits in your business before you build anything, that's what the AI Business Power Hour is for. One hour, your business, a clear answer on where to start.

Deb Szabo is an AI Business Strategist and Claude Specialist based in Pokolbin, Hunter Valley. She brings 30 years of marketing experience to installing AI Brains and practical AI systems for Australian business owners.