Copilot
The AI assistant that answers questions, writes DAX and builds report pages from your semantic model.
Copilot is only as good as the model it reads. The same question gets a sharp answer from a well-named, described model and nonsense from one full of fct_sales and Measure 7.
Part of the AI layer of the ecosystem. Reviewed 13 November 2026 by Piyush Jain.
What Copilot is
Copilot in Power BI can summarise a report, answer questions in plain language, suggest and explain DAX, and draft report pages. It works out what your data means from the semantic model: table, column and measure names, descriptions, relationships and any AI instructions you add.
It needs paid Fabric capacity (F2 or above) or a Power BI Premium P SKU, in a supported region, and an admin has to allow it in the tenant settings. It does not run on trial capacity.
When to use it
Use it when
- Your model has business names and descriptions, and one definition per measure.
- You want business users to ask questions without building reports.
Think twice when
- The model has cryptic names and duplicate measures. Fix those first, or Copilot will confidently give wrong answers.
- You have not decided who may use AI tools on which data. Set that policy before switching it on.
The mistakes I see most
From opening other people's models. Each one is fixable.
- 1
Switching it on before preparing the model
The first impression decides whether people trust it. A readiness pass, with names, descriptions and synonyms, costs far less than winning back users who got wrong numbers.
- 2
Three measures that sound the same
Revenue, Net Revenue and Sales Amount, all slightly different, give Copilot a coin toss. Keep one, describe it, and hide the rest.
- 3
Treating it as a Q&A replacement only
Q&A is being retired and Copilot covers far more. Plan for it writing DAX and building pages too, and review that output like any other change.
Questions people ask
- What capacity does Copilot in Power BI need?
- Paid Fabric capacity at F2 or above, or a Power BI Premium P SKU, in a supported region. Trial capacities are not supported, and an admin must enable Copilot in the tenant settings.
- Why does Copilot give wrong answers on my data?
- Usually because the model is hard to read: technical names, no descriptions, or several measures that look alike. Copilot guesses meaning from that metadata, so improving it improves the answers.
- Does Copilot replace Q&A in Power BI?
- Yes. Microsoft is retiring Q&A and points users to Copilot for natural-language questions.
Sources: Microsoft Learn: Copilot in Fabric overview; Microsoft Learn: Enable Copilot in Power BI. Opinions are my own.