Claude artifacts are standalone blocks of output (code, documents, visualisations, mini-apps) that Claude places in a separate window next to the chat, so you can view, edit and share them straight away. They turn Claude from a tool that only answers into a tool that delivers working output. If you are wondering what artifacts in Claude actually are: they are the separate blocks of output lifted out of the chat once they are substantial or self-contained.
The feature has existed since June 2024 but has developed considerably in 2025 and 2026. An artifact can now be static output or a mini-application that you share with others via Share, with persistent storage and even its own Claude API calls. New artifacts are private by default until you share them. This opens up a wide range of uses, from prototypes to small internal tools.
What is a Claude artifact?
An artifact in Claude is a block of output displayed in its own window next to the chat. When Claude generates something larger and more self-contained than an ordinary answer (a document of several pages, a piece of code, a diagram, an interactive visualisation), a second window opens to the right of the chat containing that output. There you can view, edit, copy, download or share the artifact. Artifacts in Claude appear automatically when the output meets those criteria; you do not have to call the feature explicitly. The setting Create and edit files (in the old view, Cloud code execution and file creation) must be switched on, though: on Free, Pro, Max and Ultra under Settings > Capabilities, and on Team and Enterprise under Organization settings > Capabilities.
Artifacts appear on their own, but working with them deliberately is something else. Briefing an artifact is a fixed step in the LearnLLM Claude Course, so you get output you can use straight away rather than text you still have to edit afterwards.
A short explanation or a few lines of code stay in the chat; only when the output is substantial and self-contained (typically more than fifteen lines) does an artifact window appear. The output must also be something you are likely to want to edit or reuse later, and readable on its own without the chat context.
If you mainly create designs rather than documents, Claude Design for visual work is the environment in which a canvas and design tools sit next to the chat.
Artifacts, including Claude Design, Claude Slides and Claude Docs, are available on all plans, including the free plan. For the basics (creating, editing and sharing artifacts via Share), you do not need a paid plan. Advanced capabilities such as persistent storage between sessions and connections to external services via the Model Context Protocol are reserved for Pro, Max, Ultra, Team and Enterprise.
When does Claude create an artifact?
An artifact appears automatically when Claude determines that the output fits better in a separate window than in the chat itself. That decision is based on three criteria that Anthropic has described explicitly.
The first criterion is size. Output of more than about fifteen lines goes into an artifact. The second is self-containment: an artifact must be understandable without the surrounding conversation. The third is reusability: things you are likely to want to open, copy or edit again later belong in an artifact.
Interactive output such as HTML, React components and SVG has an extra condition: it only works in an artifact, because it cannot be rendered in the chat column. For such output, the artifact window is not a preference but a prerequisite.
You do not have to call the feature explicitly. When you ask for a document, a dashboard, a widget or a script, Claude assesses whether it belongs in an artifact. You can, however, prompt one by asking explicitly for “an interactive version”, “a working app” or “a dashboard”; Claude will then very likely go straight into artifact mode.
What types of artifacts can you create?
The types of Claude artifacts fall into four main categories, each displayed differently in the artifact window.
Code artifacts are the most familiar type. Python, JavaScript, SQL, shell scripts: a larger block of code that Claude generates ends up in a code artifact, with syntax highlighting and the option to copy or download the file directly. If you use code as a tool without being a developer (think of analyses, data processing, automation scripts), this is a quick way to get usable scripts.
Document artifacts are Markdown or plain-text documents that are longer than an ordinary chat response. Reports, articles, specs, briefings: anything suited to a structured text of several pages ends up here. From the artifact, you can download the document or copy it into another program.
HTML and React artifacts run live in the window: an HTML page opens directly as a web page, and a React component works with clicks and forms and keeps its own state. This is the basis of what Anthropic calls “mini-apps”: small interactive applications that work within Claude itself. Interactive visualisations also belong in this category, such as charts that accept input, calculators that work through scenarios or dashboards that display data from an upload.
SVG and diagram artifacts are vector graphics and Mermaid diagrams. Logos, flowcharts, organisation charts, graphs: Claude generates them as text that renders directly into a visual. If you quickly need a simple illustration, that saves you a trip to a separate design tool.
How do you use Claude artifacts in practice?
Working with Claude artifacts is a back-and-forth process. You describe what you need, Claude produces a first version in the artifact window, you assess the result and ask for changes. In practice, three principles make the difference between an artifact that adds value and one that goes unused.
The first is to start with the problem, not the solution. “Build a project cost calculator” works less well than “I need to give clients a quick idea of what a project will cost based on hours, rate and margin”. With the second phrasing, Claude can choose a suitable structure: which inputs are useful, which calculations make sense and which output actually helps you. With the first, you get something generic that you still have to adapt afterwards.
The second principle is to work in small iterations. An artifact does not come out right in one go; it gets there through a series of requests such as “change the colour to dark blue”, “add a field for VAT”, “make it responsive”. If you ask for large changes all at once, Claude is more likely to break parts that were working. Small steps are safer and faster.
The third principle is to use versions. You can go back to any message in the chat and edit the conversation to take a new direction, without losing the existing version of the artifact. That encourages experimentation: try a radically different approach on a separate branch and choose afterwards which variant works better. This is one of the places where Claude artifacts stand apart from a purely text-based exchange.
How do you share an artifact?
You share a Claude artifact with the Share button in the artifact window. New artifacts are private by default, and only after sharing can others view and use them. On Pro, Max and Ultra, when you share, you choose between just you and anyone with the link. Inviting specific people by email address is available in beta on Pro, Max, Ultra, Team and Enterprise.
For each invited person, you set the role, from view only to making changes. An artifact you share with anyone with the link opens in most cases without a Claude account. The parts that use Claude or connected apps then do not work for people who are not signed in. Docs can only be viewed by people signed in to Claude.
On the Team or Enterprise plan, artifacts stay within the organisation by default. You share them with invited colleagues or with everyone in the organisation, and a link for people outside only works if an owner has switched on External sharing. Use of Claude in such an artifact counts towards the plan limits of each colleague who uses it, at no extra cost to the creator. For internal tools such as a calculation, a tracker or a form, that saves a great deal in management and cost allocation.
The Publish button with a public link now only exists for legacy artifacts, created in a chat before 16 September 2026. On Free, Pro, Max and Ultra, anyone with that link can view and use such an artifact without a Claude account; features that use Claude then count towards the viewer’s limits. On Team and Enterprise, the button for legacy artifacts is called Share & copy link, and viewers must be members of the organisation or, if External sharing is switched on, have a Claude account.
You can always take a published legacy artifact offline again with the “unpublish” button. After that, the old link stops working for everyone. Unpublishing permanently deletes the artifact’s storage, and it can no longer be republished; to share it again, you create a new artifact. Anyone who had made their own copy earlier keeps that copy.
What are AI-powered artifacts?
An AI-powered artifact is an artifact that can itself send a request to Claude while in use and process the answer in its own logic. So you are not building a static application but an interactive one that gives a different result per user and per input. Think of an artifact that assesses a text, generates a prompt or gives feedback on the end user’s input.
The architecture has been kept deliberately simple. The creator of the artifact does not have to manage an API key, set up hosting or put cost monitoring in place. Usage counts towards the user who actually uses the artifact, not the creator. If you build an artifact used by a hundred people, each of those hundred people is on their own Claude plan; you pay nothing extra.
The uses vary widely. Think of educational tools in which a subject teacher builds a learning resource that students can practise with and get instant feedback from. There are also content generation tools in which a template plus instructions together produce a working generator, for example for marketing copy or summaries. Other examples are small assistants for specific domains, such as a readability checker that assesses texts against CEFR level B1 or a tool that checks whether a text matches the company’s house style. Finally, there are internal aids for recurring work, for example an interview preparer that generates suitable questions based on a job profile.
The limit of AI-powered artifacts lies in what can reasonably run in a browser environment. For heavy backend tasks, integrations with several systems or dependable data storage, you hit the limits of the sandbox model. For prototypes and small production aids, it works very well.
What is the difference between Artifacts, Projects, Cowork and Skills?
Claude artifacts sit close to Projects, Cowork and Skills, but solve a different part of the problem. The four features complement each other rather than compete.
A Claude project is a workspace in which you record context for a series of chats. Skills are reusable sets of instructions that Claude calls on automatically for specific tasks. Cowork is the agent that carries out tasks for you; since 16 September 2026, Anthropic has been merging Cowork and the chat into one Claude, starting with Pro and Max. Artifacts are about how Claude presents certain results, regardless of where or with what you create them.
That means artifacts can exist alongside the other three. You can generate artifacts within a project. You can use skills that produce artifacts as output. You can create artifacts in Cowork sessions; since 6 October 2026, new Cowork tasks on Pro, Max and Ultra have run in the cloud, so sessions and files sit with your Claude account. Within Claude Code, there is a variant on every plan that includes Claude Code: a session there becomes a live, shareable page that updates as the agent works, explicitly intended as a record of work, not as an application.
In practice, the dividing line is one question: do you want to record context, encode a way of working, have work carried out, or get a working result you can share? For the last of these, artifacts are the right answer; for the first three, you choose another feature.
What are the limitations of Claude artifacts?
Working with Claude artifacts comes with three kinds of limitations you need to know about before you use them for production work.
The first limitation is technical. Artifacts run in a sandbox environment within Claude. Complex backend logic, your own database connections or integrations with external systems fall outside what is possible in an artifact. MCP connections partly lift that boundary on paid plans, but an artifact is not the right place for heavy applications; those need real development with their own hosting.
The second limitation is reliability in interactive artifacts. Live artifacts that pull data from external sources are sensitive to network problems, usage limits and changes to the sources. For demos and prototypes that is acceptable; for processes in which every action must be traceable, it is not. From a compliance perspective, a tool without guarantees of execution or an audit trail is unsuitable for regulated work, however much value the tool otherwise offers. For unregulated work, that is a minor point; for financial reporting, client administration or compliance reports, it is a blocker.
The third limitation is the editing workflow. An artifact is not a permanent project: changes are made through conversations with Claude, not through direct manual edits in a structured editor. If you are used to an IDE or a visual editor, that sometimes feels clumsy. Specific changes you would make in an editor with a few clicks sometimes take a rather long-winded instruction via the chat.
Who are Claude artifacts suitable for?
Claude artifacts are valuable for two groups of knowledge workers. The first group is professionals who regularly produce structured output that involves formatting, structure and visual presentation, such as consultants who write reports, marketers who create landing page concepts or analysts who visualise calculations.
The second group is people who want a working prototype quickly to test an idea. A product manager who wants to present a feature concept, a team leader who wants to test an internal tool, a trainer who wants to build a learning resource: in all these cases, an artifact delivers within one conversation what would otherwise require a development project with its own hosting. That makes it far easier to test ideas, and that is where the productivity gain lies.
If you only use Claude occasionally, artifacts are not a separate feature you need to learn; they appear on their own when your request calls for them. The gain lies in knowing what the feature can do, so you can use artifacts actively instead of waiting to see whether they happen to turn up. Artifacts are immediately usable for an individual user, but teams that want to use them in a structured way should agree who may share and manage an artifact.
Structuring this within a broader AI approach requires a fixed place for artifacts alongside projects, skills and Cowork, so that it is clear per task which feature you use. If you want to move from one-off experiments to repeatable, shareable artifacts in your daily work, you will find the practical approach in the Claude Course, with attention to when an artifact adds value and when an ordinary chat is faster.
Frequently asked questions about Claude artifacts
Once you have made Claude artifacts, the questions are often about finding and reusing them.
Where can you find Claude artifacts you made earlier?
You can find Claude artifacts you made earlier under the Artifacts tab in Claude’s sidebar. According to Anthropic, everything you create is stored there, so you can reopen it from any conversation. Artifacts are private by default, so only you see them there until you share them.
Can you export a Claude artifact to Word or PowerPoint?
Yes, you can export a Claude artifact to the format that matches its type. Docs go to Word, PDF, Markdown and Google Docs, presentations to PowerPoint, PDF or Google Slides. You export designs as a .zip, PDF, PowerPoint or a standalone HTML page. For older artifacts from the chat, created before 16 September 2026, you copy or download the content with the buttons at the top of the artifact.
Do Claude artifacts also work in the mobile app?
Yes, in the Claude app for iOS and Android you can request a document or presentation, view the result under Artifacts and edit it. According to Anthropic, you need Claude on the web or Claude Desktop to change the sharing settings. If you want to share an artifact with a colleague from your phone, you will need to do that later on your computer.
How much data can a Claude artifact store?
A Claude artifact can store up to 20 MB of data, and that storage only accepts text, not files or binary data. According to Anthropic, storing data and connecting apps requires a Pro plan or higher. For a small tracker or a form with text input, that space is enough; for an archive of documents, it is not.
