How to Set Up Power BI Copilot for AI Reports

Power BI Copilot adds a chat panel to Power BI reports and a standalone full-screen experience where you can ask questions in plain English, generate DAX formulas, create report pages from a single prompt, and produce written summaries of your data. This guide covers the requirements, admin setup, and semantic model preparation steps needed to get Copilot working reliably.

What Power BI Copilot Does

Copilot in Power BI is not a single feature but a collection of AI-assisted capabilities built into the service. The most commonly used version is the Copilot pane, a chat panel that opens on the right side of any Power BI report. Business users can ask questions about the report content, request a summary, or dig into specific metrics without needing to know DAX or how the underlying model is structured.

Report authors get additional capabilities: they can describe a report page in plain text and Copilot will build it, ask Copilot to write a DAX measure from a description, or add a narrative visual that generates written commentary updated dynamically as filters change.

A standalone full-screen Copilot experience, currently in preview, lets you search across all reports and semantic models you have access to and ask cross-report questions. An app-scoped version, also in preview, limits that same experience to a specific Power BI app's content.

Requirements: What You Actually Need

The licensing requirement for Power BI Copilot confuses many teams. Several third-party guides incorrectly state that Copilot requires Fabric F64 capacity at roughly $5,000 per month. That figure applies to some advanced Fabric workloads but it is not the threshold for Power BI Copilot.

According to Microsoft's official documentation, Copilot requires either:

• Paid Fabric capacity at F2 or higher (F2 capacity runs approximately $262 per month on a pay-as-you-go basis in US regions)
• Power BI Premium capacity at P1 or higher

Standard Power BI Pro licenses ($10/user/month) and Power BI Premium Per User (PPU) at $20/user/month are not sufficient on their own. Copilot requires organizational capacity, not per-user licensing. Trial capacities and free SKUs are not supported.

Your Fabric capacity must also be in a supported region. If your capacity is outside the United States or France, a Fabric admin must explicitly enable the setting that allows data to be processed outside your tenant's geographic boundary before Copilot will function.

Step 1: Enable Copilot in the Fabric Admin Portal

Copilot is enabled by default for most tenants. Your organization's Fabric administrator may have turned it off, so the first step is confirming the setting is active.

1. Go to the Microsoft Fabric admin portal at app.fabric.microsoft.com, then navigate to Admin and select Tenant settings
2. Search for "Copilot" in the settings search bar
3. Verify that "Users can use Copilot and other features powered by Azure OpenAI" is set to Enabled
4. If enabling for a subset of users only, add the relevant security group before saving

For the standalone Copilot experience (the full-screen preview version), a second tenant setting must also be enabled separately: "Users can access a standalone, cross-item Power BI Copilot experience." This toggle is independent of the main Copilot setting.

Once enabled, changes can take up to 24 hours to propagate across all users and workspaces.

Step 2: Link Your Workspace to a Copilot-Enabled Capacity

Copilot availability is workspace-scoped. Even after enabling the tenant setting, each workspace must be assigned to a capacity tier that supports Copilot.

1. Open the workspace in the Power BI service
2. Go to Workspace settings, then select the Premium or License tab
3. Set the license mode to your Fabric F2+ capacity or Power BI Premium P1+ capacity
4. Save the setting; propagation takes up to 15 minutes

Workspaces on shared capacity (the free tier) do not support Copilot even if the tenant setting is on. If the Copilot button does not appear in a report ribbon after enabling the tenant setting, the workspace capacity assignment is usually the cause.

Step 3: Prepare Your Semantic Model for AI

This is the step most setup guides skip entirely, and it has the largest impact on whether Copilot returns useful answers or generic nonsense.

Power BI Copilot generates responses by interpreting the structure of your semantic model, including table names, column names, measure descriptions, and context you provide explicitly. When a model uses cryptic column names like acct_cd_rev_adj or has no descriptions set on measures, Copilot produces generic or incorrect output.

Microsoft's own documentation states: "Without this prep, Copilot can struggle to interpret data correctly, leading to generic, inaccurate, or even misleading outputs."

To prepare your semantic model in Power BI Desktop:

1. Click the Prep data for AI button on the Home ribbon. This opens a unified preparation panel.
2. Use the AI data schema section to write plain-English descriptions of each table and its key columns. A description like "This table contains one row per sales order, with the order date, customer ID, and net revenue after returns" gives Copilot context that column names alone cannot convey.
3. Add synonyms for any column or measure that users refer to by different names. If your measure is called Net Revenue ARR but users ask "what is annual recurring revenue," adding that phrase as a synonym helps Copilot map the question to the correct measure.
4. Set up verified answers for your most common questions. Verified answers pin a specific visual to a specific trigger phrase, so when a user asks "show me sales by region," Copilot returns the verified visualization rather than generating a new one from scratch.
5. After completing preparation, mark the model as approved. In the Power BI service, open the semantic model settings, expand the "Approved for Copilot" section, and select the checkbox. This removes a friction warning that appears in the standalone Copilot experience for models that have not been explicitly reviewed.

After saving changes, close and reopen the Copilot pane to load the updated configuration. Changes can take a few minutes to take effect.

Step 4: Use the Copilot Pane in Reports

Once setup is complete, access the Copilot pane by opening any Power BI report and clicking the Copilot button in the report ribbon.

Prompt patterns that work well on well-prepared models:

• "Summarize this report" returns a written paragraph covering the key findings across all visuals on the current page
• "What drove the increase in revenue in Q1?" queries the semantic model and returns a written answer with specific figures
• "Create a new page showing monthly orders by product category as a bar chart" builds an entire report page with the requested visuals

When Copilot answers a question from your semantic model data (rather than from its general training), a "How Copilot arrived at this" link appears below the response. Clicking it shows the DAX query or reasoning path used, which lets you audit the accuracy of the answer before sharing it.

On well-prepared models, responses are consistent and auditable. On models without descriptions or synonyms, Copilot often returns a correct-looking answer that references the wrong measure.

Limitations to Know Before Deploying

Copilot supports up to 10,000 characters per prompt across all Copilot surfaces. Prompts submitted in languages other than English may return relevant results, but multilingual use is not officially supported.

The standalone full-screen Copilot experience is not yet available in several regions including India West, Indonesia Central, Korea South, Malaysia West, New Zealand North, Qatar Central, Taiwan North, UAE Central, France South, Germany North, and Norway West.

Copilot cannot access data outside your organization's Power BI environment. It does not browse the web or retrieve external context.

Copilot responses are cached when the same prompt is submitted on an unchanged model within a 24-hour window. The "clear chat" button does not bypass this cache. To force a fresh response, rephrase your prompt or refresh the semantic model.

When you scale up your Fabric capacity, it can take up to 24 hours for Copilot to recognize the change and become available.

Practical Summary

Power BI Copilot works reliably when three conditions are met: the workspace is assigned to F2+ or P1+ capacity, the tenant setting is enabled, and the semantic model has been prepared with table and column descriptions, synonyms, and verified answers for common queries.

Most teams that find Copilot underwhelming have completed steps one and two but skipped step three. The semantic model preparation is not optional; it is what separates a Copilot that answers correctly from one that generates plausible-sounding but inaccurate output.

If your organization is not yet on a Power BI Premium or Fabric plan and you want to run plain-English analysis on spreadsheet or CSV data without infrastructure setup, VSLZ AI lets you upload a file and get the same kind of natural-language querying from a single prompt with no configuration needed.