Creating with Artifacts: Turn Conversations into Real Deliverables

You’ve probably noticed Claude sometimes opens a separate panel beside the conversation instead of putting everything directly into the chat. That separate piece of work is called an Artifact.

Artifacts began as a practical way to work on substantial documents, code, and visual content without burying them inside a long conversation. They have since developed into something much broader. Claude can now use Artifacts to create documents, websites, diagrams, interactive tools, AI-powered applications, and other reusable outputs that can be revised, shared, published, or embedded elsewhere.

For a business user, the important distinction is simple: a chat response is usually something you read and respond to. An Artifact is something you are building.

This guide explains when to use one, how to create and refine it, what you can do with it afterwards, and where Artifacts are most useful in practical business work. Pasted text

What an Artifact Is

An Artifact is a substantial, self-contained piece of content Claude creates in a dedicated workspace alongside the conversation.

Anthropic describes Artifacts as content that is significant enough to stand on its own, is likely to be edited or reused, and does not depend on the rest of the conversation to make sense. Typical examples include documents, code, single-page websites, SVG graphics, diagrams, and interactive React components.

That makes an Artifact different from an ordinary answer.

If you ask:

What should an onboarding checklist include?

Claude may simply answer in the conversation.

If you ask:

Create an editable onboarding checklist for new marketing clients, organised by week and ready for my team to use.

that output has a clearer reason to exist as a standalone deliverable.

The difference is not only visual. Once something becomes an Artifact, you can continue working on the deliverable itself rather than repeatedly generating disconnected versions of it inside the conversation.

When Claude Creates an Artifact Instead of Replying in Chat

Claude generally uses Artifacts for content that is substantial enough to be worked on independently.

That could include:

• a report
• a long-form document
• a web page
• a diagram
• a piece of code
• an interactive tool
• a dashboard
• a calculator

Anthropic says Artifacts are typically used for self-contained content of meaningful size, often more than around 15 lines, that you are likely to edit, reuse, or refer back to later. Short explanations and ordinary conversational answers usually remain in the chat.

You do not need to rely on Claude to make that choice for you, though. If the result is meant to become a deliverable, say so.

For example:

Create this as an Artifact I can continue editing.

Or:

Build this as an interactive Artifact rather than explaining how I could build it.

That makes the intended result much clearer.

Common Types of Artifacts

Artifacts can now cover a much wider range of outputs than simple documents.

Type	Useful for
Written documents	Reports, proposals, guides, briefs, policies, long-form content
Code	Scripts, prototypes, utilities, small applications
Websites	Landing-page concepts, one-page resources, interactive pages
Diagrams	Workflows, process maps, architecture diagrams, decision trees
Interactive tools	Calculators, quizzes, assessments, planning tools
Data visualisations	Charts, dashboards, interactive summaries
AI-powered tools	Coaches, content helpers, guided questionnaires, assistants
Live Cowork Artifacts	Persistent dashboards connected to current business data

The format matters because it determines how people will use the result. A business report may need to be read and edited, a calculator needs to respond to input, and a dashboard needs to make information easier to interpret.

An Artifact should therefore be chosen around the job the finished output needs to perform, not simply because it looks more impressive than a chat response.

How to Ask for an Artifact Deliberately

You do not need special technical language to create an Artifact. The important thing is to describe both the deliverable and what you expect to do with it.

Compare these two requests.

Basic request:

Write a report on our Q3 marketing performance.

Claude could reasonably answer that directly in chat.

A more specific version would be:

Create an editable document Artifact for our Q3 marketing performance report. Include sections for channel performance, major changes from Q2, key wins, problems to investigate, and recommendations for Q4.

The second prompt tells Claude that you are not simply asking for information. You are asking it to create something you intend to keep working with.

The same principle applies to more interactive outputs.

Instead of:

Explain how to calculate customer acquisition cost.

you could ask:

Create a simple interactive Artifact where I can enter marketing spend and number of new customers and it calculates customer acquisition cost, with a short explanation underneath.

Now the deliverable itself is different.

Start With the Outcome, Not the Technology

One easy mistake with Artifacts is becoming more interested in what Claude can build than in whether the result solves a real problem. Start with the outcome.

Ask:

• Who will use this?
• What do they need to do with it?
• What information should it contain?
• Does it need to be editable?
• Does it need to respond to user input?
• Will it be used internally or shared publicly?
• Does it need live or stored data?

Those answers help determine what kind of Artifact makes sense.

For example, if you want to help customers estimate project costs, an interactive calculator might be useful. If you simply need to send a client a written project estimate, building an application is unnecessary.

Artifacts expand what you can create with Claude, but the simplest useful format is often still the right one.

Iterating on an Artifact

One of the main advantages of Artifacts is that you can continue improving the same piece of work. Instead of starting a new draft every time something changes, tell Claude what needs to be revised.

For example:

Make the introduction shorter but leave the rest unchanged.

Add a comparison table after the pricing section.

Remove the third step and renumber the process.

Make the calculator show monthly and annual totals.

Change the layout so the results appear above the input form.

Claude updates the Artifact while the surrounding conversation remains available for context.

For Markdown documents, Claude also supports more targeted editing. You can highlight a specific section inside the Artifact, choose Edit with Claude, and describe what should change there. Claude can then apply the edit directly to that selected area.

Claude also keeps versions of an Artifact, allowing you to move between previous iterations rather than treating every change as irreversible. That makes the workflow feel closer to collaborating on a working document or prototype than repeatedly requesting fresh answers from a chatbot.

Working With More Than One Artifact

A conversation does not have to contain only one Artifact. Claude can create and work with several Artifacts within the same conversation, and you can switch between them when needed.

That is useful when a larger project has several distinct deliverables.

For example, a launch-planning conversation might contain:

• a campaign brief
• a launch checklist
• an interactive budget calculator
• a landing-page prototype

Keeping these as separate Artifacts can be cleaner than trying to combine everything into one large document. The conversation provides the common context, while each Artifact has a clear purpose of its own.

Saving and Finding Artifacts

Artifacts are no longer limited to being buried inside the conversation where they were created. Claude now includes an Artifacts section in the sidebar where you can view and manage Artifacts you have added there.

There is one important detail: an Artifact created inside a conversation does not automatically appear in that sidebar collection. Publishing the Artifact adds it to the Artifacts section, where it becomes easier to find again later.

You can also copy or download Artifact content for use outside Claude. Depending on the type, that might mean downloading the underlying file or copying the code behind an interactive Artifact.

For important business deliverables, it is still sensible to keep the final version in whatever system your business normally uses for documents, files, or version control rather than treating a Claude conversation as your only record.

Publishing an Artifact

For Free, Pro, and Max users, Claude can publish an Artifact through a public link. Once published, other people can view and interact with it. Basic published Artifacts can even be used by people who do not have a Claude account. AI-powered functionality requires the relevant Claude authentication and uses the visitor’s own Claude usage rather than the creator paying for API access.

This creates several useful business possibilities.

You could publish:

• a calculator
• a quiz
• a self-assessment
• an interactive guide
• a public reference tool
• a lightweight prototype

Published Artifacts can also be embedded on another website using automatically generated embed code. Claude lets the creator specify which domains are allowed to embed the Artifact.

For Team and Enterprise accounts, sharing works differently. Artifacts can be shared internally with members of the organisation rather than published openly to the public. That distinction matters if the Artifact contains internal business information.

Be Careful What You Share

Publishing and sharing make Artifacts more useful, but they also introduce a reason to review what is attached to them.

Anthropic notes that when a Team or Enterprise Artifact is shared, viewers may also gain access to attachments and files from the conversation that created it. That means you should not automatically share an Artifact simply because the finished output itself looks harmless. Check the surrounding conversation and source files as well.

This is especially important if the work involved:

• customer data
• internal documents
• financial information
• contracts
• strategy material
• confidential client files

The convenience of sharing should not replace ordinary information-handling judgment.

AI-Powered Artifacts

One of the biggest changes to Artifacts is that they can now include Claude itself. Instead of building only a static page or calculator, you can create an Artifact that responds intelligently to what the user enters.

For example, you might create:

• a brainstorming assistant
• a guided content planner
• a personalised learning tool
• an interactive coach
• a question-and-answer tool
• a business self-assessment

Claude writes the application, and the AI capability runs through Anthropic’s infrastructure. People using the Artifact do not need the creator to provide an API key. AI usage counts against each user’s own Claude usage limits.

This makes AI-powered Artifacts much more accessible to non-developers. You can create a small AI-enabled tool through conversation without setting up your own API integration, authentication system, or hosting environment.

That does not mean every business idea should become an AI app. It simply means the boundary between “ask Claude for something” and “use Claude to build something other people can interact with” has become much thinner.

Connecting Artifacts to Other Tools

Artifacts can also connect to external services through MCP, or the Model Context Protocol, on supported paid plans.

That allows an interactive Artifact to work with services such as:

• calendars
• project-management tools
• communication platforms
• other connected business systems

For example, an internal planning Artifact could potentially work with information from a connected calendar or task-management system rather than depending entirely on manually entered data.

Each user still has to authenticate the external service independently, and Claude asks for permission when the Artifact first needs access.

This is a significant step beyond the original idea of Artifacts as standalone documents or prototypes. They can increasingly become interfaces into the systems where work is already happening.

Persistent Storage

Supported Artifacts can also store information between sessions. Persistent storage is currently available for published Artifacts on Pro, Max, Team, and Enterprise plans on Claude web and desktop.

Storage can be configured in two ways.

Personal storage keeps each user’s information private to them. That could work for something like a personal journal, progress tracker, or planning tool.

Shared storage makes information visible across users of the same Artifact. That might support something like a leaderboard or collaborative tracker.

This means an Artifact no longer has to forget everything once the user closes it. It can behave more like a lightweight application with continuity between sessions.

There are still limitations, including storage limits and restrictions on the types of data that can be stored, so Artifacts should not be treated as replacements for full business databases or purpose-built software.

Live Artifacts in Cowork

Cowork introduces another variation: live Artifacts. These are persistent, interactive HTML dashboards that Claude can create inside Cowork and update using current information from connected applications.

For example, instead of asking for a one-time project-status summary, you could have Claude create a dashboard that reflects information coming from the systems the team already uses.

The difference is important. A normal Artifact is something you create and then interact with. A live Artifact can become an ongoing view of changing work.

Live Artifacts currently have different availability and sharing rules from ordinary published Artifacts, including being limited to Claude Desktop and supported paid plans. For most business owners starting with Artifacts, this is an advanced use case rather than the first place to begin.

Artifacts vs. Custom Visuals

Claude can also create custom visuals directly inside Chat and Cowork. These may look similar to Artifacts, but they serve a slightly different purpose.

Custom visuals are designed to help explain or explore something in the moment. You can copy or download them, and if one becomes worth keeping, you can save it as an Artifact.

A useful distinction is:

Custom visual: useful for understanding something now.

Artifact: useful when you want to keep building, sharing, or using it later.

For example, if Claude creates a quick visual explaining a process during a conversation, there may be no reason to save it. If you decide that visual should become an interactive resource for your team, turning it into an Artifact makes more sense.

Business Use Cases

The practical value of Artifacts becomes clearer when you stop thinking of them as a Claude feature and start thinking about the deliverables your business already needs.

A Client Proposal

Instead of drafting a proposal in chat and copying it into another tool, ask Claude to create the proposal as a structured document Artifact from the beginning.

You can refine sections individually, compare versions, and then move the final material into the format your business normally uses.

An Interactive Calculator

A consultant might create a pricing estimator. A marketer might build a return-on-ad-spend calculator. A service business could build a basic project-budget tool.

The useful part is not that Claude can write the code. It is that you can describe the business logic in ordinary language, test the result, and refine the tool through conversation.

A Customer Self-Assessment

An Artifact could ask a visitor several questions and return a result based on their answers.

That could work as:

• a lead-generation tool
• an educational resource
• a readiness assessment
• a recommendation tool

If AI functionality is useful, the Artifact can go further and generate a personalised explanation rather than returning one fixed result.

An Internal Dashboard

Instead of repeatedly asking Claude to summarise the same information, a live Artifact in Cowork could provide an ongoing dashboard connected to current business data.

This is a more advanced workflow, but it shows how Artifacts are moving from “output” toward “working interface.”

A Landing-Page Prototype

Before building a page in your website platform, you can ask Claude to create a working prototype. That gives you something visual to react to.

You can change the hierarchy, wording, sections, or interaction and then use the final prototype as a reference for the actual build.

A Reusable Internal Tool

Some small business tools do not justify commissioning full custom software.

An Artifact might be enough for:

• a decision helper
• a checklist
• a planning tool
• a questionnaire
• an internal calculator
• a lightweight tracker

This is where Artifacts can become especially useful for non-technical users. The goal is not to become a software developer. It is to recognise when a small custom tool can replace a repetitive manual process.

What Artifacts Are Not

Artifacts are powerful, but they are not automatically the right format for every task.

You probably do not need one for:

• a quick factual answer
• a short rewrite
• a simple explanation
• brainstorming you will not reuse
• an ordinary conversational question

Creating a separate application or document for every response adds complexity rather than reducing it.

They also should not automatically be treated as production-grade software. A prototype built through Claude may be completely suitable for a lightweight internal tool, but something handling sensitive customer information, payments, authentication, regulated workflows, or critical business processes deserves more careful technical review.

The fact that a tool works does not necessarily mean it is ready for every environment.

Practical Takeaway

Use an Artifact when the output is something you want to build, revise, keep, share, publish, or use, rather than simply read once.

For a simple deliverable, that may be a document or diagram. For something more interactive, it could be a calculator, dashboard, prototype, or AI-powered tool.

The workflow is straightforward:

1. Define what the finished deliverable needs to do.
2. Ask Claude to create it as an Artifact.
3. Test and refine it through the conversation.
4. Decide whether it should remain private, be downloaded, published, embedded, or shared internally.
5. Use more advanced features such as AI, MCP connections, or persistent storage only when the use case genuinely requires them.

The real value of Artifacts is not that Claude can produce something in a separate panel. It is that the conversation can become the place where an idea turns into something usable.

FAQ

Do I need to ask for an Artifact by name?

No. Claude may create one automatically when the content is substantial and self-contained. If you know you want an editable or interactive deliverable, asking for an Artifact directly makes your intention clearer.

Can I edit an Artifact directly?

Yes, depending on the type. Markdown Artifacts support targeted editing through Edit with Claude, and you can also ask Claude to revise the Artifact through the conversation.

Can one conversation contain more than one Artifact?

Yes. Claude can create and manage multiple Artifacts within the same conversation.

Can I download an Artifact?

Yes. Claude provides options to copy or download Artifact content for use elsewhere.

Can I publish an Artifact publicly?

Free, Pro, and Max users can publish supported Artifacts through public links. Team and Enterprise accounts use organisation-only sharing instead.

Can I put an Artifact on my website?

Published Artifacts can be embedded on approved website domains using code Claude provides.

Does someone need a Claude account to use a published Artifact?

Not necessarily. People without Claude accounts can view and interact with basic published Artifacts. AI-powered functionality requires Claude authentication because the AI usage runs against the user’s own Claude allowance.

Can an Artifact remember information between visits?

Supported published Artifacts can use persistent storage on eligible paid plans. Storage may be personal to each user or shared between users, depending on how the Artifact is built.

Can Artifacts connect to other business tools?

On supported plans, Artifacts can use MCP connections to work with external services. Users must authenticate the relevant connection and approve access.

Are Artifacts the same as the custom visuals Claude sometimes creates in a conversation?

No. Custom visuals are primarily generated to help explain or explore something in the moment. If you want to keep and develop one, you can save it as an Artifact.

What are live Artifacts?

Live Artifacts are interactive dashboards created through Cowork that can refresh using current information from connected applications. They currently have different availability and sharing rules from standard Artifacts.

Can Artifacts replace custom software?

Sometimes for lightweight tools and prototypes, but not universally. A simple calculator or internal workflow tool may be perfectly suitable as an Artifact. Software handling sensitive data, critical business processes, payments, security, or regulatory requirements may need a more conventional development and review process.