Microsoft has placed Copilot in yet another corner of Power BI. This time, it's inside the web-based semantic modeling experience, currently in preview. The feature promises to let analysts and developers make changes to the data model using natural language, without having to open Power BI Desktop or write a single line of DAX by hand.
It's worth understanding what that actually means before getting your hopes up.
What Copilot in web modeling is
Copilot in web modeling is an AI-powered assistant integrated into the Power BI service's semantic model experience. In plain terms: you open your semantic model in the Power BI Service, switch to editing mode, and click "Copilot" in the ribbon. From there, you start talking to it.
You can ask things like "create a year-over-year growth measure for the revenue column" or "rename all columns in the fct_sales table to follow business naming conventions" and Copilot executes, or at least tries.
The process works like this:
- Open the semantic model in Model view in the Power BI Service.
- Switch from Viewing mode to Editing mode.
- Select Copilot from the ribbon.
- Type what you want to do.
Before acting for the first time, Copilot asks for explicit permission to review and modify the model. This approval covers the entire session. And it creates an automatic restore checkpoint at the start of each session, making it easy to roll back anything that didn't go as planned.
What it can do
The feature's capabilities fall into three categories:
Analysis and recommendations
Copilot can review your model's structure and flag issues. Inconsistent column names, poorly defined relationships, undescribed tables. Think of it as a reviewer who read the entire model and handed you a prioritized list of improvements.
Model editing
This is the most practical part. Using natural language, you can create tables, columns, measures, and relationships. You can also generate DAX measures (totals, growth metrics, aggregations) and create RLS rules to restrict data access by user profile.
Discoverability and organization
Copilot suggests descriptions for tables, columns, and measures, proposes display folder structures, and recommends which technical fields should be hidden from report authors. Anyone who has inherited a poorly documented semantic model knows how much this matters.
What it doesn't do (yet)
- The feature is in preview, meaning it can change, break, or underperform in specific situations.
- Using Copilot to configure prepare data for AI settings within the model is not yet supported.
- Changes respect existing permissions. If you don't have write access to the semantic model, Copilot won't be able to edit anything on your behalf.
- It works best on well-structured models. A messy model with cryptic names will generate poor responses. AI doesn't perform miracles.
Who can use it
Here's the part that will frustrate a lot of people: Copilot in web modeling requires Microsoft Fabric F64 capacity or higher, which includes Power BI Premium P1 licenses or equivalent. It does not work with a standard Power BI Pro license.
F64 is a shared capacity SKU. Think of it as renting a slice of Microsoft's processing infrastructure. It costs significantly more than an individual Pro license, so this feature, for now, is for organizations with a more robust infrastructure in place.
On top of that, the Fabric admin settings must allow the use of Copilot and Azure OpenAI-powered features. Without that enabled at the tenant level, nothing works.
Why this matters for Power BI developers
Historically, any structural change to a semantic model required Power BI Desktop. Open the .pbix file, make the change, publish it back. That's not the end of the world, but it adds unnecessary friction when the model is already published and someone just needs a quick adjustment.
The ability to make edits directly in the browser, with AI assistance for generating DAX or creating RLS rules, reduces that cycle. For analysts who work on shared models and don't have easy access to Desktop, that's a real, practical improvement.
It's also a signal of the direction Microsoft is heading. Power BI is progressively becoming a Fabric-first tool, and web modeling is part of that transition. Anyone still working on 100% Desktop-based workflows will need to adapt.
Prepare your model before using Copilot
One detail the Microsoft documentation emphasizes: Copilot performs significantly better on well-prepared models. That means:
- Clear descriptions on tables and columns.
- Names that make sense to the business, not internal technical abbreviations.
- Correctly defined relationships.
- Key measures already created and documented.
A model without that foundation will generate generic recommendations and questionable DAX measures. Copilot uses the model's context to understand what you have. If that context is poor, the output will be too.
Wrapping up
Copilot in web modeling for Power BI is a genuinely useful feature for organizations that have the infrastructure to run it. The idea of editing semantic models in natural language, directly in the browser, makes sense and solves a real problem.
The issue is the licensing requirement. F64 puts the feature out of reach for most small and mid-sized businesses that rely on Power BI Pro or Premium Per User. Those who have access will benefit. Those who don't will be watching from the sidelines for a while.
For now, the smartest move is to get your semantic models properly structured. When access eventually comes, you'll be ready to make the most of it.