4.1.4. Copilot for Report Authoring

💡 First Principle: Copilot in Power BI turns natural-language requests into report content by reading your semantic model's tables, columns, measures, names, and descriptions. It accelerates the first draft, but it can only be as good as the model it reads, and it runs on paid capacity under admin control.

The PL-300 names three authoring scenarios:

• Suggest content for a new report page: in the Copilot pane, ask Copilot to suggest content. It proposes page outlines (for example, "Sales performance by region," "Customer retention overview") based on the model, each describing the visuals it would create. You then choose one to generate.
• Create a new report page: ask Copilot to create a page from a prompt ("Create a page showing revenue trends by product category and region for the last two years"). Copilot builds a page of visuals using model fields and measures; you then refine formatting, swap visuals, and verify accuracy.
• Create a narrative visual with Copilot: add the Narrative with Copilot visual from the Visualizations pane and choose a prompt (give an executive summary, answer likely questions from leadership, create a bulleted list of insights) or write your own. It summarizes the visuals on the current page or across the report, can reference specific pages, and updates as filters change. It's the modern way to add written commentary that stays in sync with data.

Prerequisites and governance (frequently tested):

• Copilot usage must be billed to a paid capacity of at least F2 or P1. Either the report's workspace sits on a Microsoft Fabric (F SKU) or Power BI Premium (P SKU) capacity, or the user is assigned to a Fabric Copilot capacity, which lets them use Copilot in Power BI Desktop and in Pro or Premium Per User workspaces while billing the usage to that designated capacity.
• A Fabric administrator must enable Copilot in the tenant settings (and settings for data processing across geographies may apply). For a Fabric Copilot capacity, a capacity admin assigns the users.
• Without a Fabric Copilot capacity assignment, Desktop authors select a Copilot-enabled capacity workspace to use for Copilot even though they're authoring locally.

Quality depends on the model: clear table and column names, measure descriptions, synonyms, hidden technical fields, correct data categories, and star-schema design all make Copilot's output more accurate. Microsoft also provides Prep data for AI features (AI data schema, verified answers, AI instructions) that let model authors guide Copilot. Always review generated visuals and narratives; Copilot can choose a reasonable but wrong field.

⚠️ Exam Trap: When a stem says Copilot options are unavailable for a report in a Pro workspace, the fix is providing paid capacity (moving the workspace to an F or P capacity, or assigning the user to a Fabric Copilot capacity) and confirming the tenant setting is enabled, not reinstalling Desktop or granting the user Build permission.

Reflection Question: Two models contain the same data, but Copilot builds much better pages from one of them. List three model characteristics that likely explain the difference.