The suggestions are built from measure names and descriptions, the things readers slice by, and where the data comes from. Nothing about your data values.
The model export doesn’t say where tables are loaded from. This second query reads the Power Query behind each table so the sources can be named for you.
Connected with a PBIP folder above? Its report pages are read from it too, so there’s nothing to add here. Otherwise, for a report saved as a Power BI Project (.pbip): the page names and slicers feed the tips, and measures on more visuals rank higher as key measures. If the folder includes the semantic model, data sources are read from it too. Files are read on this page; nothing is uploaded.
The few numbers that matter most to readers. The summary, business value, questions and notes are written around them, and the Copilot prompt lists only these. A few are suggested to start you off: tick the ones that belong and untick the rest.
A few answers only you know. Each one improves the drafts in Step 4; leave any of them blank.
Each part says why it helps. Tick the ones you want, then edit the draft or pick another suggestion. Drafts update from Steps 1 to 3 until you edit them.
Copilot can smooth the summary, business value and questions. The prompt holds only the plain-language facts above (names, meanings, sources), no DAX. It asks Copilot not to invent anything and to list what it couldn’t tell. You choose which parts of the reply to use.
Paste it into a text box on an About page, a report description, or a SharePoint or Teams page. Read it through once: suggestions are a starting point, and AI can make mistakes.
For an HTML content visual from AppSource. The text above stays the source of truth: edit it there, then copy the measure again. A measure is harder to review and edit than a text box, so use it when you want the About page styled or the last refresh shown live.
The DAX-only copy is for New measure in Power BI Desktop. The measure replaces any measure with the same name in that table.