Claude Dashboards Tutorial: Build Live Dashboards from Plain-Language Questions

Introduction

Most data questions die in a queue. Somebody files a ticket, an analyst writes SQL, and the answer arrives days later as a static screenshot that is already out of date. Claude Dashboards removes the ticket step for quick, exploratory questions. You connect a data source once, ask a question in plain language, and Claude builds a dashboard that stays current as the underlying data changes. This tutorial walks through that whole loop: connecting a source, asking a well-formed question, verifying the numbers before you share them, and handing a dashboard off to a BI tool when the question outgrows Claude.

Quick answer

Connect a data warehouse or Salesforce to Claude, ask a plain-language question in any chat or from the Output menu, then verify each chart by clicking its number to read the query and checking its refresh time.

Evidence

The feature is young, so the evidence here is about scope, not maturity. Anthropic launched Claude Dashboards on October 8, 2026, in beta on paid plans, alongside Claude Motion and the general availability of Docs, Slides, and Design. The launch post states the core mechanism in one sentence: “Claude Dashboards connects to a data platform like BigQuery, Databricks or Snowflake, or a CRM tool like Salesforce, and is in beta on paid plans.” That sentence sets the whole tutorial: the input is a connected source, the output is a dashboard, and the availability gate is a paid plan.

The transparency claim is the part worth testing in practice. Anthropic writes: “Each chart shows the query behind it, and each animation is built as code you can edit.” For dashboards, that means every number on the page can be traced back to the SQL Claude wrote for it. The same post adds the intended use case, and it is a limit as much as a promise: “Dashboards work well for quick, exploratory questions, like how this week’s signups compare with last month’s.” A dashboard that answers a one-off question in minutes is the win. A governed, audited company metric is still a job for your BI tool, and Anthropic says so itself by building a send-to path for deeper analysis.

The numbers that frame the launch are concrete. Five data platforms are named at launch: Amazon Redshift, BigQuery, ClickHouse, Databricks, and Snowflake. Eight destinations are named for sending a dashboard onward: Amplitude, Grafana, Hex, Mixpanel, Omni, Perplexity, PostHog, and Sigma. And the scale of the surrounding artifact format is large: Anthropic says users had made more than 45 million docs, decks, and designs in Claude since the artifact templates merged into every conversation. Those 45 million artifacts are adoption evidence for the format, not a quality claim about any single dashboard, and this tutorial treats them that way. Full details are in Anthropic’s Dashboards and Motion announcement and the Claude Help Center guide to getting started with Dashboards, both checked on November 4, 2026.

Prerequisites

• A paid Claude plan: Pro, Max, Team, or Enterprise. The Free plan does not include Dashboards.
• A data source you are allowed to query: a warehouse such as BigQuery or Snowflake, or an app already connected to Claude, such as Salesforce.
• Read access to the tables or objects you plan to ask about. If you only have access to part of a warehouse, say so in your question so Claude does not guess at tables it cannot see.
• On an Enterprise plan, an administrator must first turn Dashboards on under Organization settings, then Artifacts. It is off by default.
• A question worth a dashboard, not a single number. Dashboards pay off when the answer has at least two views, such as a trend plus a breakdown.

Step-by-step guide

Step 1: Connect your data source before you ask anything

Dashboards read from connections, not from pasted data. Before creating a dashboard, connect the data you want to ask about in Claude. For a warehouse, that means the platform connection for Redshift, BigQuery, ClickHouse, Databricks, or Snowflake. For a CRM question, it means the app connector you already use in Claude, such as Salesforce. Confirm the connection works by asking Claude a trivial question first, such as listing the tables or objects it can see. If Claude cannot see the source, fix the connection now, not after you have written a long dashboard prompt.

This step matters because of a sharing rule that surprises first-time users. A dashboard built on a connected source reads data through the viewer’s own connection, so a teammate who opens your dashboard needs their own access to the same source. Test sharing with one teammate early, before you build a dashboard the whole team depends on.

Step 2: Ask for the dashboard in plain language

Open any chat and ask for the dashboard directly. Anthropic’s own example prompt is: “Build a dashboard of this quarter’s revenue by region.” You can also select Output in the message box and choose Dashboards, or start from the Dashboards template in the Artifacts tab. All three routes produce the same kind of artifact: a page of charts built from one question.

Write the question the way you would ask an analyst, not the way you would label a chart. The Help Center gives this exact guidance: say what you want to know, not just what to chart. “How do this month’s signups by channel compare with last month’s?” gets a better dashboard than “a signups chart,” because the first version tells Claude the comparison, the grouping, and the time window. If the data lives in a specific table or app, name it in the question.

A good first dashboard has three parts: one headline trend, one breakdown, and one table. For example: “Build a dashboard of weekly signups for the last six months, split by plan, plus a table of the top ten signup sources this month.” That shape is easy to verify chart by chart, and it matches the exploratory use case Anthropic describes.

Step 3: Verify every chart before you trust the page

Verification is the step that separates a dashboard from a guess. Claude writes a SQL query for each chart and runs it against your data source, and each chart shows its query along with when its data was last refreshed. Work through the page chart by chart:

1. Click a number and read the query behind it. Check the table, the date filter, and the grouping. A query that groups by the wrong column produces a confident, wrong chart.
2. Check the refresh time on each chart. A dashboard that stays current is only current if its charts have actually refreshed since the data changed.
3. Cross-check one headline number against a source you trust, such as the warehouse console or the CRM report for the same period. If the headline matches, the breakdowns are more likely sound. If it does not, stop and fix the query before sharing.
4. Ask Claude to explain any query you cannot read. The explanation is a check on your understanding, not a substitute for reading the query itself.

Do this verification before the dashboard leaves your screen. Once a dashboard is shared, every viewer treats its numbers as yours.

Step 4: Refine the dashboard in conversation

A first dashboard is a draft. Refine it by asking follow-up questions in the same chat: narrow the date range, split a chart by a second dimension, drop a chart that adds noise, or rename a chart so its title states the question it answers. Keep each refinement to one change at a time so you can tell which request broke a chart. When a refinement produces a number that contradicts an earlier version, click into both queries and find the difference, usually a date filter or a grouping, before deciding which version is right.

Name charts after the question they answer. A chart titled “Signups by channel, this month vs last month” survives a forwarded screenshot. A chart titled “Chart 2” does not.

Step 5: Share it, and know what sharing does and does not do

Dashboards start private to you. To share one, open it and click Share. Sharing rules follow the artifact rules in your organization: an administrator may allow sharing outside the organization or restrict it to people with a link, and a dashboard that reads from a connected app uses the viewer’s own connection, not yours. That last rule is a feature for access control and a trap for demos: a teammate without access to the source will not see your numbers, they will see a dashboard that cannot load its data. Confirm the viewer can open it before you present it in a meeting.

Step 6: Send the dashboard onward when the question gets deep

When a question needs deeper analysis, send the dashboard from Claude straight to Amplitude, Grafana, Hex, Mixpanel, Omni, Perplexity, PostHog, or Sigma and pick up the analysis there. This is the correct exit for questions that need governed metrics, cohort definitions, or joins across sources Claude cannot see. Treat the Claude dashboard as the exploration that found the question, and the BI tool as the place where the answer becomes official. Databricks Genie, Looker, monday.com, and Tableau are listed as coming soon, so check the current destination list before promising a handoff to one of them.

Troubleshooting

The dashboard option does not appear

This is almost always a plan or admin problem, not a bug. Dashboards are in beta on paid plans only, so a Free plan account will not see the option. On Team plans, confirm your workspace has it enabled. On Enterprise plans, Dashboards and Motion are off by default, and an administrator must turn them on under Organization settings, then Artifacts. If a teammate sees the option and you do not, compare plans first, then admin settings.

A chart shows a number you know is wrong

Click the number and read the query. The usual causes are a wrong date filter, a grouping on the wrong column, or a table that is not the one you meant. Rephrase the question with the table name and the exact date range, and ask Claude to rebuild only that chart. If the rebuilt chart still disagrees with the warehouse console, the problem may be the connection’s permissions: Claude may be reading a view or a filtered object rather than the raw table. Check which object the query targets before assuming the data itself is wrong.

The dashboard loads, but a teammate sees no data

A dashboard built on a connected source reads through the viewer’s own connection. Your teammate needs their own access to the same warehouse or app, connected to their own Claude account. If they have access and still see nothing, check whether their connection covers the same tables your query uses. For a dashboard that must be viewable by people without source access, rebuild it on data that does not require a live connection, or move the analysis to a BI tool and share that instead.

A chart looks stale after the data changed

Each chart shows when its data was last refreshed, so check that timestamp first. If the refresh time predates the data change, the chart has not picked up the new data yet. Ask Claude to refresh the dashboard, and recheck the timestamp. If the timestamp updates but the number does not move, the query may be filtering to a closed period, such as last month, that genuinely has not changed. Read the query’s date filter before concluding the refresh failed.

The dashboard counts against your usage faster than expected

Anthropic states that Claude Dashboards counts toward your plan’s usage limits, like the rest of your work with Claude. A dashboard with many charts runs many queries, and each refinement runs them again. Keep dashboards to the charts the question needs, refine in batches rather than one chart at a time, and reserve heavy exploration for the BI tool once the question is settled.

FAQ

See the front matter FAQ above for the most common questions. Two more points come up often in practice. First, dashboards are artifacts, so they live in the Artifacts tab and can be shared like other artifacts, subject to your admin’s sharing rules. Second, dashboards are not a BI replacement: Anthropic positions them for quick, exploratory questions, and the send-to path exists precisely because deeper analysis belongs in a dedicated tool.

Conclusion

Claude Dashboards earns its place by shortening the path from question to chart, and it keeps that place only if you verify what it builds. Connect one source, ask one well-formed question, read the query behind every chart, check the refresh time, and share only after a teammate can open the result. When the question outgrows exploration, send the dashboard to your BI tool and let the governed metric live there. For the wider artifact workflow, see our tutorial on generating design slides and docs with Claude Code artifacts, and for the connector side of the same setup, see how to connect an MCP server to Claude Desktop. If your team runs its automation in n8n rather than in chat, the complete n8n guide covers the same connect, ask, and verify loop for workflows.

Frequently asked questions