7.3 Copilot in Power BI Reports

Quick Answer: Copilot in Power BI uses generative AI to create narrative visuals (text summaries), suggest report layouts, and generate entire pages from natural-language prompts. It requires a paid Fabric capacity (F2 or higher) or Power BI Premium, and it responds to filter context so summaries stay dynamic.

Copilot Capabilities for Report Authors

1. Create a Narrative (Smart) Visual

Insert a narrative visual (Insert > Narrative, with the Copilot icon). Copilot analyzes the semantic model and current filter context, generates a text summary of key insights, and updates dynamically as slicers and filters change. Authors can steer it with prompts such as "Summarize the key trends in revenue by region," "Highlight the top and bottom performing products," or "Explain the year-over-year change in profitability."

2. Create a New Report Page

From the Copilot pane, prompt "Create a report page that shows..." and Copilot selects appropriate visuals for the data types, arranges them, applies filters and formatting, and adds titles. Example prompts: "Sales performance by region with trend over time," or "An executive summary with key KPIs, revenue by product, and monthly trends."

3. Suggest Content and Summarize the Model

Copilot can suggest which measures and dimensions to use and which visuals fit, and it can describe the semantic model, table purposes, relationships, available measures, and suggested analyses, which is helpful when you inherit an unfamiliar model.

Narrative Visual Configuration

After the first draft, you can edit the prompt to focus on specific topics, reference specific measures or dimensions, set the tone, include or exclude data points, and pin insights that should always appear. The narrative is dynamic: it regenerates as slicers change, reflects cross-filter selections, and updates with page- and report-level filters.

Requirements and Licensing (Updated)

Important update: Copilot used to require a large F64 (or P1) capacity. Since April 2025, Microsoft made Fabric Copilot available on all paid SKUs starting at F2, dramatically lowering the entry cost. Treat any study material that still states an F64 minimum as out of date, the current minimum is F2 (paid), and trial capacities remain unsupported.

Limitations

Copilot may misinterpret complex relationships, so review generated narratives before sharing. It does not suggest custom visuals, very large models can respond more slowly, and results are best in English. Use Copilot as a starting point and refine, rather than publishing its output unchecked.

Best Practices

1. Review generated content for factual accuracy, Copilot can misread data context.
2. Be specific in prompts: "Show revenue by region for 2026" beats "show sales."
3. Use it as a draft, then customize visuals, formatting, and layout.
4. Combine with manual design, let Copilot scaffold a page, then refine it.
5. Test under different filters, confirm narratives stay accurate when the report is sliced.

For governance, remember Copilot honors the model's security: row-level security and workspace permissions still apply, so users only see data they are allowed to see. The tenant admin can also scope Copilot to specific security groups during rollout.

On the Exam

The PL-300 tests what Copilot can and cannot do in reports, creating narrative visuals and their dynamic behavior, using Copilot to suggest or create pages, the licensing/capacity requirement (paid Fabric F2+ or Premium, not trial), and recognizing when Copilot is the appropriate solution.

Where Copilot Fits in the Authoring Flow

Copilot is best understood as an accelerator for the report author, not a replacement for modeling judgment. It shines at first drafts, scaffolding a new page from a prompt, summarizing a region's performance in prose, or explaining an unfamiliar semantic model you have inherited, and it stays current because narrative visuals re-evaluate with the filter context. The PL-300 expects you to know what Copilot can do (narrative summaries, page generation, content and model suggestions) and, just as importantly, what it cannot do reliably (interpret intricate relationships flawlessly, suggest custom visuals, or be trusted without review).

Treat its output as a starting point you refine.

The Capacity Story Has Changed, Know the Current Facts

The single most common stale fact about Copilot is its capacity requirement. At launch it demanded a large F64 (or Premium P1) capacity, which priced out smaller teams. Since April 2025, Microsoft enabled Fabric Copilot across all paid Fabric SKUs starting at F2, a dramatic reduction in the entry cost, while still excluding trial capacities and still requiring tenant-admin enablement. For the exam and for real planning, the correct answer to "minimum capacity for Copilot" is now a paid capacity of F2 or higher (or Power BI Premium), and any material citing F64 as the floor is out of date.

Governance and Trust

Because Copilot reads the same semantic model your users do, it inherits the model's governance, row-level security and workspace permissions still apply, so Copilot cannot reveal data a user is not entitled to see. Admins can also roll Copilot out to specific security groups. The practical author responsibilities are to write specific prompts (naming measures, dimensions, and time frames), to review generated narratives for accuracy before sharing, and to test across filter contexts so a summary that is correct for "All Regions" stays correct when sliced to one region.

These habits keep AI-assisted reports both fast to build and safe to publish.