Copilot in Power BI Reports
Use Copilot to create narrative visuals, generate report pages, suggest content, and summarise the underlying semantic model — the newest PL-300 exam topic.
AI meets your reports
Imagine having an analyst sitting next to you. You ask “summarise this month’s sales” and they write a paragraph explaining the highlights, trends, and outliers. You ask “create a page showing campaign performance” and they draft the layout with appropriate charts.
Copilot in Power BI does exactly this. It can write text summaries of your data, suggest which visuals to add, create entire report pages, and explain what’s in your semantic model. It’s a new feature — and the April 2026 PL-300 exam tests four specific Copilot skills.
Narrative visuals with Copilot
A narrative visual is a text block that automatically describes your data in natural language.
How to create:
- Insert a Narrative visual from the Visualizations pane
- Click Copilot in the visual
- Copilot generates a summary based on the data visible on the current page — or you can direct it to summarise specific visuals or the entire report.
Riley at Coastal Fresh (🛒) adds a narrative visual to her sales dashboard. Copilot generates:
“Total revenue for the selected period is $1.2M, a 12% increase from the prior period. The North region leads with $420K (35% of total). Dairy saw the strongest growth at 23% YoY, while Bakery declined 5%.”
Customising the narrative:
- Use prompts to guide the summary: “Focus on regional comparisons” or “Highlight products with declining sales”
- The narrative can be refreshed when filters change — click the refresh icon to regenerate
- You can adjust the output by reprompting rather than editing the text directly
Exam tip: narrative visuals and semantic models
Copilot generates better narratives when the semantic model is well-documented:
- Measure descriptions help Copilot explain what metrics mean
- Column descriptions provide context
- Synonyms (defined in Q&A settings) help Copilot understand business terminology
If Copilot’s narrative seems generic or inaccurate, improving the semantic model metadata is the fix.
Creating report pages with Copilot
Copilot can generate an entire report page based on a natural language prompt.
How to use:
- Click Copilot in the ribbon (or Copilot pane)
- Type: “Create a page showing sales trends by region over time”
- Copilot suggests a layout with appropriate visuals, fields, and formatting
- Review and adjust — Copilot’s output is a starting point, not a finished product
Nadia at Prism Agency (📊) asks Copilot: “Create a page comparing Google Ads and Meta Ads campaign performance.” Copilot generates a page with a column chart comparing platforms, a line chart for trends, and cards for key metrics.
Suggesting content for pages
Instead of creating the whole page, you can ask Copilot to suggest what should go on a page you’re building.
How to use:
- Create a blank page or a page with some visuals
- Ask Copilot: “What visuals should I add to this page?”
- Copilot analyses the semantic model and suggests relevant charts, measures, and fields
This is useful when you know the topic but aren’t sure which visuals will tell the story best.
Summarising the semantic model
Copilot can explain what’s in a semantic model — tables, relationships, measures, and their business meaning.
How to use:
- Open the Copilot pane
- Ask: “Summarise this semantic model” or “What tables and measures are available?”
- Copilot describes the model structure, key measures, and relationships
Dr. Ethan at Bayview Medical (🏥) inherits a complex model from a departed colleague. Instead of manually exploring 15 tables and 40 measures, he asks Copilot to summarise the model — getting a quick overview of what’s available and how tables relate.
| Copilot Feature | What It Does | Best For |
|---|---|---|
| Narrative visual | Generates a text summary of the data on the page | Executive summaries, automated insight callouts |
| Create report page | Builds an entire page with visuals from a prompt | Quick first drafts, exploring data visually |
| Suggest content | Recommends visuals and fields for a page | Inspiration when you're unsure what to show |
| Summarise model | Describes tables, relationships, and measures | Understanding inherited or complex models |
Requirements and limitations
| Requirement | Detail |
|---|---|
| Licence | Fabric capacity F64+ or Power BI Premium P1+ (PPU alone is not sufficient) |
| Semantic model quality | Better metadata → better Copilot results |
| Language | English produces the best results; other languages supported but less reliable |
| Not a replacement | Always review Copilot output for accuracy — it can misinterpret data relationships |
Best practice: prepare your model for Copilot
Kenji at Apex Manufacturing (🏭) improves Copilot results by:
- Adding descriptions to all measures and key columns
- Setting up synonyms in Q&A settings (e.g., “output” = “UnitsProduced”)
- Organising columns into display folders
- Using clear, business-friendly names (not technical codes)
This isn’t just good practice for Copilot — it makes the model easier for all report authors.
Knowledge check
Riley adds a narrative visual to her dashboard. The generated text is vague and doesn't mention her key measures by name. What should she do to improve it?
Dr. Ethan inherits a Power BI model with 15 tables and 40 measures. He wants a quick overview of what's available. What should he ask Copilot?
Next up: Report Pages and Paginated Reports — configure pages and choose when to use paginated reports.