Claude Can Now Show You: Diagrams, Tools, and Visualizations

▶

Beyond Words: When a Conversation Needs to Show, Not Just Tell

For most of its existence, Claude has communicated exclusively in text. This is a powerful medium — text can describe, explain, reason, argue, and instruct with remarkable precision. But there is a substantial class of information that is better shown than described. A decision tree is clearer as a visual flowchart than as nested bullet points. A dataset’s patterns are more legible in a chart than in a paragraph describing the trends. An interactive tool for comparing options is more useful than a static comparison table.

The artifact system Claude introduced changes the fundamental medium of the conversation. Rather than describing what a chart would show, Claude can generate the chart. Rather than explaining how a decision framework works, Claude can build an interactive tool that lets you apply it to your actual situation. The conversation expands from text exchange to a mixed-media environment where visual and interactive outputs appear alongside explanatory text.

This is not image generation in the sense of AI-generated photographs or artwork. Claude’s artifacts are functional outputs — code that runs, diagrams that organize information visually, interactive tools that respond to user input. The artifact panel functions as a mini-application runtime embedded in the conversation: code Claude writes executes there, and the results are visible in real time.

Types of Artifacts Claude Can Generate

The range of artifact types covers several distinct use cases, each suited to different kinds of informational needs.

Diagrams and flowcharts are perhaps the most immediately intuitive category. Any process, relationship, or structure that benefits from spatial organization — decision trees, org charts, system architectures, causal diagrams, process flows — can be rendered visually rather than described textually. Claude generates these as SVG or as rendered code, not as rasterized images, which means they remain crisp at any size and can be edited programmatically.

Data visualizations transform tabular or numerical information into charts that reveal patterns. Bar charts, line graphs, scatter plots, histograms, and more complex visualization types (treemaps, network graphs, heatmaps) can all be generated when you provide data or ask Claude to work with a dataset. For exploratory data analysis, being able to say “visualize this data as a scatter plot colored by category” and see the result immediately — rather than having to write visualization code first — compresses the feedback loop dramatically.

Interactive tools are the most powerful category and the one that most clearly distinguishes Claude’s artifact system from static image generation. An interactive tool is a small, self-contained application that runs in the artifact panel and responds to user input. Examples include: a decision-support tool that takes your specific inputs and walks you through a framework; a calculator built around a specific formula or methodology; a comparison matrix where you can adjust weights and see how options rank; or a scenario planner where you can adjust variables and see projected outcomes update in real time.

The difference between receiving a comparison table as text and receiving an interactive comparison tool is significant. The text table is fixed — if you want to change the criteria or adjust how you weight different factors, you need to ask Claude to regenerate it. The interactive tool lets you explore the space yourself, which produces a different kind of understanding.

The Artifact Interface: Sandbox, Shareable Links, and Iteration

Claude’s artifacts run in a sandboxed environment — the code executes in an isolated container within your browser, with no access to your local filesystem or other browser tabs. This makes them safe to generate and run without worrying about malicious code execution.

The sandboxing also means artifacts are self-contained: everything the artifact needs to run is included in the code Claude generates. This makes artifacts straightforwardly shareable — Claude can generate a shareable link to any artifact, and the recipient can view and interact with it in their own browser without any installation or configuration. For tools you want to share with colleagues or clients, this is much more frictionless than sharing code that they would need to run themselves.

Iteration works naturally within the conversation. You can tell Claude “make the chart a line chart instead of a bar chart,” “add a third option to the comparison tool,” or “change the color scheme to be more accessible” — and Claude updates the artifact accordingly, keeping the context of previous versions in the conversation. This iterative refinement loop, where you can see the result and immediately ask for changes, is significantly faster than the traditional workflow of editing code and re-running it yourself.

How to Prompt for Visual Artifacts

Generating useful artifacts requires slightly different prompting instincts than generating useful text responses. A few principles make a significant difference.

Be explicit about the artifact type you want. Claude will sometimes generate a visual artifact proactively when it recognizes the content would benefit from it, but specifying “create a diagram,” “build an interactive tool,” or “visualize this as a chart” makes the intent clear and avoids a text response when you wanted something visual.

Specify the format within the type. “A diagram” is less useful guidance than “a flowchart showing the decision points in this process” or “a network diagram showing which services depend on which.” For charts, specify the chart type when you have a preference: bar chart, line chart, scatter plot, and pie chart all communicate different kinds of relationships between data.

Provide the data explicitly when you have it. If you want Claude to visualize your data, paste it into the conversation rather than describing it — Claude can generate a more accurate and useful chart when it can see the actual numbers rather than working from your summary.

Ask for interactivity when exploration is the goal. If you want to explore a dataset or test different scenarios, request an interactive tool explicitly: “build an interactive version where I can filter by category” or “make this a tool where I can adjust the assumptions and see how the output changes.”

Real Use Cases: From Student Planning to City Parks

The demonstration Anthropic showed in the announcement illustrated two instructive use cases that highlight the range of artifact applications.

The student major selection use case showed a decision-support artifact that helped a student think through their college major choice by mapping their interests and career goals against program options. This is a classic case where an interactive tool — one that the student could adjust as they reflected on their priorities — is more valuable than a static recommendation. The tool maintained the complexity of the decision (multiple relevant factors, personal weighting) in a form that invited exploration rather than collapsing it to a single answer.

The city park approval use case showed a more civic and professional application: visualizing the approval process for a public project in a way that made the bureaucratic stages, decision points, and stakeholder dependencies legible at a glance. Process diagrams of this kind are genuinely useful for anyone navigating complex multi-step procedures — regulatory submissions, grant applications, product launches — where understanding the full sequence and the dependencies between stages is important for planning.

Both use cases share a common thread: the artifact made complex, multi-factor information navigable in a way that text alone could not. That is the reliable indicator of when to reach for artifact generation: when the information you are working with has a structure — decision tree, process flow, data distribution, comparative ranking — that benefits from visual representation.

Here's my data:
[paste your data]

Create a bar chart comparing [X variable] across [Y dimension]. 
I want to highlight that [the comparison/pattern you care about].
Make it interactive so I can hover over each bar to see the exact values.

Create a flowchart for [process name]. The steps are:
1. [Step 1, including who is responsible and what decision points exist]
2. [Step 2...]
[continue for all steps]

Show decision points as diamonds and actions as rectangles. 
Include arrows indicating which path leads where when a decision goes yes vs no.

Build an interactive scoring tool for evaluating [decision type].
The criteria I care about are: [list 4-6 criteria].
Let me adjust the weight of each criterion with a slider.
Show a total score for each option that updates as I adjust weights.
I want to be able to enter [2-4 options] and compare them.