An artifact is a finished piece of work that Claude creates and shows next to your conversation.

Anthropic’s Help Center describes an artifact as anything Claude makes for you that you would put in front of someone else. Examples it gives include a design, a deck, a document, a dashboard, or a small interactive tool. Instead of being buried in the flow of the chat, the work opens in its own area beside the conversation, so you can see it clearly, keep working on it, and come back to it later.

Artifacts matter because many things people ask Claude for are not just answers. They are things to use. A report to send to your manager, a one-page website, a chart for a presentation, a quiz for your class, or a calculator for your shop. Artifacts give these outputs a proper home instead of leaving them as text scrolling up the screen.

flowchart LR
    A["You describe what you need in the chat"] --> B["Claude makes it"]
    B --> C["It opens as an artifact beside the chat"]
    C --> D["You review, ask for changes, or share it"]
        

A few simple words appear again and again in this topic. Here they are in plain English.

Not everything Claude says becomes an artifact. The difference is whether you will use the result as a thing.

When you ask a quick question, Claude answers in the chat, and that is all you need. You read it and move on. But when you ask for something you will keep, edit, or share, an artifact is usually the better home. As a rule of thumb, if the result is something you would reread, polish, reuse, or hand to another person, it fits an artifact. If it is a quick explanation you will read once, a normal reply is enough.

Artifacts come in many shapes, from plain documents to small working apps.

You do not press a special button. You simply describe what you want.

Artifacts are made to be improved, and every improvement is remembered.

An artifact is not a one-shot result. You refine it by talking to Claude, and each meaningful change is saved as a new version. A version selector lets you switch between versions, so you can look at an earlier result or go back if a change was not an improvement. This takes away the fear of experimenting, because nothing is lost.

You can also have more than one artifact in a single conversation. Controls let you switch between them and choose which one you want Claude to update. This is helpful when a project involves, for example, a report and a matching slide deck.

Artifacts start private. Nothing is shared until you decide to share it.

This is an important point for peace of mind. According to Anthropic’s article on sharing artifacts, artifacts start private to you, and nothing is shared until you share it. When you are ready, you open the artifact and use its Share control. What you can choose depends on your plan and on where the artifact was made.

flowchart TD
    A["New artifact"] --> B["Private to you by default"]
    B --> C{"Do you want to share it?"}
    C -- "No" --> D["Stays private"]
    C -- "Yes" --> E["Open the Share control"]
    E --> F["Choose who can open it and what they can do"]
    F --> G["Viewers see the version you chose"]
        

Some artifacts can use Claude inside them, turning them into small apps.

Anthropic describes AI-powered artifacts as artifacts that embed Claude’s abilities. Instead of being a fixed page, such an artifact can respond to the person using it. Those people can ask questions, generate content, get coaching, play games, and solve problems, with Claude adapting to what they enter.

How usage works

According to the Help Center, when you share an AI-powered artifact, others can use it right away, with no API keys needed and no cost to you as the creator. People using it sign in with their own Claude account and interact with their own instance of the artifact. On Team and Enterprise plans, colleagues in your organization can use shared AI-powered artifacts without adding costs for the creator.

Artifacts can sometimes use your connected apps, but only with your approval.

Some artifacts are more useful when they can reach information in your other apps. According to the Help Center, the first time an artifact needs a connected app, Claude shows which apps and tools it will use and asks you to approve them. You can turn individual tools off, and your choice carries over to later uses of that artifact. The article also notes that connector tools that need approval for each action are not available to artifacts.

Engineers can turn their coding sessions into shareable pages.

Claude Code, the product for software engineers, can also create artifacts. Anthropic describes them as web pages hosted on claude.ai that start private and that the person can later choose to share with teammates. They are useful when a visual explanation is clearer than terminal text, such as a dashboard summarizing test results or a page explaining a set of code changes.

The sharing rules for these can be stricter than for chat artifacts. For example, the documentation says that by default these artifacts are visible only to their creator, that they can be shared with others in an organization, and that they cannot be shared publicly. Each time the artifact is published again, it becomes a new version, and people you share with are viewers by default, seeing each version you publish.

Let us follow an imagined shop owner named Meena as she builds her first artifact.

In under an hour Meena had a working tool made for her exact needs, without writing code. She kept control throughout: she tested, checked, and chose who could see it.

Clear requests lead to better first drafts.

Artifacts are available widely, with some differences in sharing and controls.

The Help Center states that accessing artifacts in the sidebar and using Claude-powered artifacts are supported on the Free, Pro, Max, Team, and Enterprise plans. Making an artifact does not carry a separate fee. It uses your normal allowance, so a large or elaborate artifact uses more of it than a small one.

Differences appear mainly in sharing. The sharing options and who can open an artifact vary by plan, and on organization plans administrators have extra controls. If a sharing option you expect is missing, the likely cause is your plan or your organization’s settings.

A finished-looking artifact still needs a careful human review.

An artifact can look polished, which makes it easy to trust too quickly. A neat calculator can still contain a wrong formula. A well-designed report can still include an invented fact. Anything you plan to rely on or send to others should be checked first, in the same way you would check a draft written by a colleague.

A few habits cause most of the trouble with artifacts.

Artifacts add real value, with a few things to keep in mind.

Artifacts sit alongside other Claude features, and it helps to know how they differ.

flowchart LR
    A["Projects: organize your work"] --> D["Chats inside the project"]
    D --> E["Artifacts: finished outputs"]
    B["Connectors: reach other apps"] --> E
    C["Memory: knows you over time"] --> D
        

Quick answers to the questions beginners ask most often.

It is a finished piece of work that Claude makes for you, such as a document, design, deck, dashboard, or small tool, shown beside the chat so you can edit, keep, and share it.

A normal reply is an answer you read in the conversation. An artifact is a standalone result that you can open, refine, save versions of, and share.

Describe what you want in the chat. For substantial, reusable results Claude will usually create an artifact, and you can ask for one directly if it does not.

Yes. According to the Help Center, artifact access is supported on Free, Pro, Max, Team, and Enterprise plans, although sharing options differ by plan.

Artifacts rely on code execution and file creation being enabled in your settings under Capabilities. On organization plans, this is managed in the organization’s settings.

Yes. Ask Claude for changes in plain language, and use the version selector to move between versions if you want to go back.

Only you, until you share it. Sharing options depend on your plan, and on organization plans artifacts stay inside the organization by default.

An artifact with Claude built in, so people using it can ask questions or generate content. When shared, others use it with their own Claude account, with no cost to the creator.

Sometimes, through connectors, but Claude asks you to approve the apps and tools first, and you can turn individual tools off. Tools that need approval for each action are not available to artifacts.

Tell Claude what you see and ask it to fix the problem. It will try, but success is not guaranteed and some errors need extra troubleshooting.

No. Artifacts can look very polished and still contain mistakes. Check important facts and test calculations before relying on or sharing them.

In the Claude Help Center, in the articles “What are artifacts and how do I use them?” and the one on publishing and sharing artifacts.

An artifact is a finished piece of work that Claude creates and shows beside your conversation, such as a document, design, slide deck, dashboard, or small interactive tool. You make one simply by describing what you want, refine it by talking to Claude, and move between saved versions as you improve it. Artifacts start private, and you choose whether and how to share them, with options that depend on your plan and your organization’s rules. Some artifacts can contain Claude itself, letting others use them as small AI apps with their own accounts, and some can use connected apps with your approval. Because polished results can still contain errors, the sensible habit is to test, verify, and share only as widely as needed.