How to use the Data plugin in ChatGPT Work and Codex

Use the Data plugin to investigate a business question in ChatGPT Work and Codex. It can use connected company data and business context to analyze what changed, explain findings, and create interactive dashboards and reports. You refine, validate, and publish by asking follow-up questions in the same conversation.

The Data plugin must be available in your ChatGPT Work or Codex account or workspace before that works. To analyze a connected service, the relevant data-source plugin and any included app must also be available. Queries use the connected account's existing permissions, including applicable table, row, and column restrictions.

Make Data available

If you are a workspace administrator:

Go to the Admin Console and select the intended ChatGPT workspace.
Open Plugins and select Data.
Review its installation policy. Available lets members install it themselves. Installed installs it by default.
Enable and configure the relevant data-source plugins and manage access to their included apps.

In Enterprise and Edu, configure plugin installation policy and each app's role access for the roles assigned to the members or groups who need access. In Business, manage plugin installation policy and app availability across the workspace. The help page points that role setup to its role-based access controls article.

Data will prompt you to install necessary third-party plugins from within the conversation, so you do not need other setup up front as long as the relevant plugins are enabled in your workspace. Plugin installation and access to the underlying apps are separate. A plugin can be installed while a required app still needs setup, authorization, or workspace access. Some apps require an administrator to configure an app template before members can use it. Existing guidance is in the ChatGPT app templates article, including the Snowflake and Databricks app-template guides.

Add a warehouse, definitions, and Sites

For best results, the page strongly recommends at least:

A data warehouse plugin, like Databricks or Snowflake
A data semantic layer, which contains authoritative definitions and queries to keep the model on track
ChatGPT Sites, which allows you to publish and share dashboards and reports with your team
Optionally, a BI tool, which can provide more useful context and a familiar visualization service

Data can work with approved sources such as Amazon Redshift, ClickHouse, Databricks, Google BigQuery, MongoDB, and Snowflake. It can also bring files and documents from Google Drive and SharePoint into the analysis when those connections are available. It can work with connected business intelligence tools, including Omni, Oracle BI, Power BI, Sigma, Tableau, and ThoughtSpot. Available dashboard actions depend on the connected tool, its supported capabilities, and your access.

Your organization's business definitions, sometimes called a semantic layer, help Data improve accuracy. These may include metric definitions, custom calculations, and relationships between datasets from semantic layers, documentation, or trusted dashboards. Tell Data which definition or source to use when your team has a preferred one.

Install Data and name the question

Open Plugins in the ChatGPT desktop app, or the plugin directory in ChatGPT on the web.
Find and select Data.
If it is not already installed, select Install plugin.
Complete any required account-connection or setup steps for the data sources you want to use.
Start a conversation with @Data and describe your business question.

Include the source, metric, time period, and comparison when you know them. For example:

"@Data Compare revenue this quarter with last quarter and explain the largest changes."
"@Data Investigate why customer retention declined last month. Show the data and definitions you used."
"@Data Build a dashboard showing sales pipeline by region, and call out the biggest changes."

The Data plugin can be triggered implicitly, so it is not always necessary to add @Data. Start with the explicit mention, including specific skills, to build familiarity, then rely more on implicit invocations. If you are unsure, ask ChatGPT Work or Codex if the Data plugin was used.

As you work, Data will search for available plugins in the workspace and offer to install them if they would help the query. That includes OpenAI plugins and plugins on an internal company marketplace.

Review the result, then make a dashboard

After an initial answer, ask follow-up questions to compare groups, examine a change, or review the evidence. Check the source, time period, filters, and metric definition before relying on a result. If a result differs from an existing report, ask Data to compare those details.

Ask Data to turn an analysis into an interactive dashboard. Describe the metrics, breakdowns, and comparisons you want to see. You can ask it to adjust the charts or layout and provide brand guidelines to tailor the dashboard's appearance. You can edit, share, and refresh dashboards as needed. Ask Data to explain which information supports a chart and to refresh the analysis when you need updated results.

Publish with the copy in mind

If you have OpenAI Sites installed, select Publish in your locally generated artifact, enter the prompt, and choose who you would like to share it with. The data used in the analysis is copied into the published site, so be mindful of data permissions when choosing who to share it with.

Once published, you can keep the dashboard up to date by setting up a cloud automation, which the Data plugin can also help with. Sharing through Slack or email and taking actions in connected tools depends on the tools' supported actions, your permissions, and applicable approval requirements. Review the destination and content before approving an action. You can also ask ChatGPT Work or Codex to recommend next steps or prepare findings for others.

Reuse a template or a context skill

Templates are for repeatable analyses, dashboards, and reports.

Say "@Template Creator please make a template from ". This creates a skill, for example "$product-feature-deepdive".
Later, prompt "Use $product-feature-deepdive on the new image carousel feature".
Share the skill as a zip, or package it in a plugin and ask your administrator to share it. An administrator can auto-install that plugin for the workspace. More native sharing features are coming soon.

Context skills hold instructions that modify the Data plugin's default behavior, from analysis best practices to a design system. They can also include a lightweight semantic layer for key definitions.

Say "@Data I'd like to make a new context skill".
Go through the guided flow and populate the context you want used later.
Review and accept the skill, and set the audience to yourself, the team, or the whole company.
It is saved as a local skill and used automatically when needed for later data work.
Share it as a zip, or package it in a plugin and ask an administrator to share it or auto-install it.

When something is missing

If Data or a required source is unavailable, check that you are using the intended account and workspace. Review the plugin's availability and setup requirements. In a managed workspace, ask an administrator to check Data's installation policy and access to the underlying apps.

If a connection works but information is missing, confirm that the intended provider account is connected and can access the information directly in the provider's service. A successful connection does not create additional source permissions. Include a specific dataset, report, or metric in your request.

If a dashboard or action cannot be completed, check whether the connected tool supports the requested action, the account can perform it, and workspace settings allow it. ChatGPT may also be waiting for approval. If an analysis looks incorrect, review the underlying sources and definitions before sharing or taking action.

What to do next

If you administer the workspace, open the Admin Console, set Data to Available or Installed, and also set up a data warehouse plugin such as Databricks or Snowflake, a data semantic layer, and ChatGPT Sites. If you are a member, install Data from Plugins or the plugin directory, finish the account connections you are allowed to use, and start with an explicit @Data mention, including specific skills. Include the source, metric, time period, and comparison when you know them. Before you select Publish, read the help article below and decide who should receive a copy of the analysis data.

Sources

Using the Data plugin in ChatGPT Work and Codex