Claude Data: Live Sharing, Refresh Behavior and Permission Checks

Claude Data became generally available on November 10, replacing the Claude Dashboards name and adding more ways to present company data. Before sharing a dashboard, check whose connection runs its queries and when its numbers refresh. Those two details determine whether colleagues see the right data at the right time.

The product announcement introduces certified queries and live sharing as beta features. They are useful for teams that want one agreed definition of a metric rather than several attractive dashboards with different answers.

Separate the metric from the presentation

Consider an illustrative weekly activation report for a developer product. Sales wants to count accounts that signed up, while engineering defines activation as completing a successful API request. Both can produce a rising chart. Only one answers whether new users reached the product’s useful outcome.

Before asking Claude for a dashboard, write a definition: one activated account is a distinct customer account with its first successful production API request during the reporting week. Specify the timezone, exclude internal accounts and decide how retries are handled. These are business decisions, not formatting choices Claude should quietly make.

A useful prompt is: “Show weekly activated customer accounts using the approved activation definition. Show the query, reporting timezone and data refresh time. Do not substitute signups if activation events are unavailable.” This is an example to adapt, not a tested query for your database.

What a certified query establishes

The Claude Data guide says charts expose their SQL and refresh time. Editors can certify queries, and editing a certified query removes its certification. Definitions can also come from dbt, Looker or Cube through existing connectors.

Certification is a governance decision, not a proof that the metric is correct. For the activation example, ask the owner of the metric to review a small sample of qualifying accounts. Compare them with the source events, including an internal test account and an account that only signed up. The chart should exclude both.

Keep the reviewer’s name and the reason for the definition somewhere the team can find. A badge is useful only if people know what review it represents. If a product change creates a new activation event, revisit the definition rather than assuming an old approval covers the new behavior.

Live sharing changes whose access is used

Without live sharing, viewers use their own data connections. With it, the shared dashboard can show results using the owner’s access, including saved results. Viewers receive the approved query results rather than permission to run arbitrary queries as the owner.

For a weekly activation report, that means choosing the output before choosing the audience. A company-wide aggregate may be appropriate for the whole team. A drilldown containing customer emails, contract details or employee activity may need a much narrower audience even if everyone would find it interesting.

Review the actual result columns, not merely the dashboard title. “Activation by account” can reveal considerably more than “Weekly activations.” Use the smallest audience that needs the result and avoid granting access to a detailed table just to share a chart.

Refresh intervals are not scheduled reports

Claude Data refreshes when a dashboard is opened and its data is due for refresh. It does not run on a schedule. A configured interval therefore does not establish that a report was regenerated at an exact time.

If the team meets at 9 a.m., make one person responsible for opening and checking the report before the meeting. Look at the displayed freshness information, then verify that the source has received the latest relevant events. A refreshed query over a delayed pipeline can still produce incomplete numbers.

For an unattended alert or a report that must arrive at a fixed time, use an explicitly scheduled workflow and verify its delivery. An interactive dashboard and a scheduled operational dependency require different checks.

Check Enterprise defaults before November 19

The Artifacts admin guide says Claude Data defaults on for Enterprise on November 19, 2026; the Data help guide notes that an owner’s previously saved setting is respected. Certified queries and live sharing have separate controls. Both default on for Team and off for Enterprise.

Turning live sharing off makes viewers connect their own data source. Turning a template off is different from disabling Artifacts entirely: the latter blocks access to the organization’s artifacts, including existing shared links, until enabled again.

Review those settings deliberately instead of switching off everything when one dashboard raises a concern. The useful rollout starts with a trusted metric, a limited audience and a visible freshness check. Expanding access should follow evidence that those three pieces work together.

I’m Paras. I work at the intersection of AI, engineering and distribution.

I help technical founders turn product details and practical lessons into clear stories that developers and buyers can use.

Book a 30-minute call.