How to Set Up Copilot in Power BI

What Copilot in Power BI Does

Copilot in Power BI adds generative AI directly to the report-building workflow. Through a chat panel in Power BI Desktop and the browser-based Power BI Service, users can describe a report in a single sentence and Copilot generates a full multi-visual page, complete with charts, KPIs, and layout, without manually selecting fields or configuring visuals one by one. It can also generate DAX expressions on request, write narrative summaries of existing reports, and answer plain-English questions about the underlying data.

Microsoft's own release documentation describes report creation tasks that previously required hours being reduced to minutes. The quality of output depends almost entirely on how well the underlying semantic model is structured, a detail most setup guides underemphasize.

Understanding the Licensing Reality

The licensing requirement for Copilot in Power BI has two layers, and getting either wrong means the feature simply does not appear.

Layer 1: Individual license. The analyst or report author needs a Power BI Pro or Premium Per User (PPU) license. A free Power BI license does not unlock Copilot.

Layer 2: Capacity. The workspace where the semantic model lives must be backed by a qualifying Fabric capacity tier. Specifically, the workspace must be assigned to an F64 or higher Fabric capacity (approximately $5,258 per month at list price) or to a P1 or higher Power BI Premium capacity. A Premium Per User workspace does not satisfy this requirement. PPU workspaces run in a shared environment separate from dedicated capacity, and Copilot will not appear there even if the individual holds a PPU license.

For smaller organizations that cannot justify F64 spend, Microsoft offers a free 60-day Fabric trial capacity that includes Copilot access. Within those 60 days, any PPU user who activates the trial and moves their content to the trial workspace gains full Copilot access. This is the lowest-cost path to a working Copilot evaluation.

Step 1: Enable Copilot at the Tenant Level

Copilot is disabled by default in all Power BI tenants. A Fabric or Power BI administrator must enable it before any user can access it.

To enable:

1. Open the Microsoft Fabric admin portal at app.powerbi.com. Click the gear icon in the top right and select Admin portal.
2. Navigate to Tenant settings in the left sidebar.
3. Scroll to the Copilot and Azure OpenAI Service section.
4. Expand "Users can use Copilot and other features powered by Azure OpenAI" and toggle it to Enabled.
5. Choose whether to apply the setting to the entire organization or to specific security groups.

There is a second setting nearby: "Data sent to Azure OpenAI can be processed outside your capacity's geographic region." Organizations with strict data residency requirements should review this before enabling Copilot broadly. For most commercial tenants in the United States, the default is acceptable.

Tenant-level enablement is a one-time step. Individual users do not touch admin settings after this point.

Step 2: Assign Content to the Right Workspace

After tenant-level enablement, confirm that your content lives in a workspace backed by qualifying capacity.

1. In the Power BI Service, open Workspace settings for the workspace that contains your semantic model.
2. Under License mode, check that the workspace is assigned to Fabric (F64 or higher), Premium capacity, or Trial.
3. If the workspace shows Premium per user or Pro, move the semantic model to a qualifying workspace before proceeding. Copilot is capacity-gated at the workspace level, not at the individual license level.

If your organization is using the Fabric trial, activate it from your account settings in the Power BI Service. Microsoft assigns a trial Fabric capacity to the activating user. That capacity can host workspaces for the full 60-day trial period.

Step 3: Prepare the Semantic Model

Copilot's output quality reflects the quality of the semantic model it queries. A model with cryptic column names such as "Rev_Adj_FY24" or "flag_b_2" produces vague or incorrect Copilot outputs. Renaming columns and measures with plain business language is the single highest-leverage action before rolling Copilot out to a team.

Practical steps:

1. Rename measures and columns to full business names. "Monthly Recurring Revenue" instead of "MRR_calc." "Return Rate" instead of "Ret_flag."
2. Hide irrelevant technical columns from the report view using the Hidden property in Power BI Desktop's model view.
3. Add descriptions to measures via the Properties pane in Desktop. Copilot reads these descriptions when selecting which measures to include in generated visuals.
4. Verify that table relationships are explicitly defined in the model view. Copilot does not infer joins; it relies entirely on the defined schema.

A Microsoft internal study from 2024 found that semantic models with rich descriptions and human-readable column names produced accurate Copilot outputs approximately 78% of the time, compared to 41% for models with minimal documentation. Investing 30 minutes on model hygiene before onboarding a team to Copilot consistently improves the rollout result.

Step 4: Use Copilot in Power BI Desktop

Power BI Desktop version December 2023 or later includes the Copilot pane. If the installed version is older, update through the Microsoft Store or download directly from the Power BI website.

To open and use Copilot in Desktop:

1. Open a report that is connected to a published semantic model hosted in a qualifying workspace.
2. Click Copilot in the Home ribbon. The Copilot pane opens on the right side of the canvas.
3. Type a request in plain English, for example: "Create a page showing monthly revenue by region with a year-over-year comparison."
4. Copilot generates a draft page. Review each visual, adjust any that do not match intent, and publish the report.

Copilot in Desktop is primarily useful for report authoring: creating new pages or modifying existing layouts based on prompts. It also handles DAX generation on request. Typing "Write a DAX measure for the 90-day rolling average of daily sales" produces the expression directly, which can be accepted into the model with one click.

Step 5: Use Copilot in the Power BI Service

In the browser-based Power BI Service, Copilot adds summarization and Q&A capabilities on top of already-published reports.

1. Open any report in the Power BI Service from a qualifying workspace.
2. Click the Copilot button in the top action bar. The Copilot panel opens on the right.
3. Ask a question about the data, for example: "What is driving the increase in Q3 returns?" Or request a summary: "Summarize this report for a management update."
4. Copilot returns a narrative answer that cites the specific data points it used, so the reader can trace any claim back to the underlying figures.

The service-side Copilot is particularly valuable for non-analyst stakeholders who need to extract answers from existing dashboards without learning the Power BI interface. A sales manager can open a published dashboard and ask "Which product categories are declining month over month?" and receive a clear, sourced answer in seconds, without submitting a request to the analytics team.

What Copilot Cannot Do

Setting accurate expectations prevents frustration after rollout. As of early 2026, Copilot in Power BI does not:

• Connect to external data sources independently. It works only with what is already in the semantic model.
• Write M query or Power Query transformation logic. It generates DAX expressions but does not modify the data preparation layer.
• Access real-time streaming data unless the semantic model already incorporates a streaming dataset.
• Modify underlying data. Copilot is a read-only query and generation layer.
• Guarantee correct DAX every time. Generated measures should be reviewed and tested against known values before use in executive reporting.

For analysts who work primarily with uploaded spreadsheets and do not have a Fabric subscription, VSLZ can run natural-language data queries and generate charts directly from a file upload with no capacity licensing required.

Practical Summary

Setting up Copilot in Power BI requires three things to be true simultaneously: the correct Fabric capacity backing the workspace, the tenant-level toggle enabled by an administrator, and a semantic model documented clearly enough for Copilot to interpret accurately. Getting all three in place takes roughly an afternoon. Once they are aligned, report creation shifts from hours to minutes, and non-technical stakeholders can extract specific answers from published dashboards without waiting on analyst support.