Copilot in Power BI

How It Works, Capabilities and Requirements.

While tools such as Microsoft 365 Copilot Chat help users work with general organizational resources, Copilot in Power BI was designed specifically for working with data models and business reports.

Where exactly can you find Copilot in the Power BI environment?
How are user questions processed, what context does AI use, and what limitations and requirements should you be aware of before implementation?

This article provides a detailed expansion of the Power BI section in our comprehensive guide to Microsoft Copilot:
„Microsoft Copilot for Business: Which Copilot Should You Choose?”

Copilot and Power BI Touchpoints – Where Can You Find It?

Copilot in Power BI is not a single, monolithic tool. Depending on your role in the organization (developer or analyst vs. business user) and the task you are performing, you can use Copilot across three main environments: Power BI Desktop, Power BI Service, and the Microsoft 365 ecosystem.

Copilot interaction environments based on the user’s role and task.

Power BI Desktop

A local environment designed for analysts and report creators. Here, Copilot primarily supports work on reports, DAX code, and model documentation:

Based on a text description, Copilot can suggest and create a layout of visualizations on the report canvas.

The assistant makes it easier to generate and explain DAX queries, and can help simplify or refactor existing code, add comments, and organize logic in DAX Query View.

It also supports the creation of business descriptions for measures. These descriptions are stored in the model metadata and help document its logic; well-prepared metadata can also improve the quality of Copilot’s responses.

Power BI Service

This is the most comprehensive environment for working with Copilot. Depending on the task, users can access several distinct Copilot experiences:

Standalone Copilot / Copilot Hub (preview)

Available as a full-screen experience from the Power BI navigation.

It allows users to search for reports and semantic models they have access to, add them as context, and ask questions without having to open a specific report first.

Copilot in a Power BI app (preview)

Available within a Power BI app and limited to the content shared through that app.

It answers questions in the context of the reports and semantic models made available to the user within the app and helps identify the report that contains the information needed to answer a business question.

Copilot in an individual report

Available in the Copilot pane on the right side of an open report – for both report consumers and report authors, depending on their permissions.

It can summarize report content, answer questions about the semantic model, and create visualizations based on natural-language questions. If the answer is not available in the existing visuals, Copilot can use measures and columns from the semantic model to build a new visualization – including for metrics that are not currently visible on the report page.

Model editing view in Power BI Service (preview)

In web modeling, Copilot analyzes the structure of the semantic model and can identify potential issues with naming or model design.

It can create and update tables, columns, measures, and relationships, generate or correct DAX, create RLS roles, and suggest descriptions and display folders for fields.

Narrative visual and summaries

Copilot can create narrative summaries of an entire report, selected pages, or specific visualizations.

After filters or slicers are changed, the narrative visual does not update automatically – the user must refresh the summary to reflect the new data context.

Microsoft 365 Ecosystem

Some work with Power BI data can take place outside the Power BI portal itself. Fabric IQ in Microsoft 365 Copilot Chat allows users to ask questions about data available in reports and semantic models without leaving their everyday work environment.

Copilot uses the user’s existing Power BI permissions, while data-based responses can be combined with broader Microsoft 365 context, such as files, emails, chats, and collaboration activity. Within the Microsoft IQ architecture, Fabric IQ provides the business data context and is built on OneLake, while Work IQ provides context related to how users work. This capability is currently available through the Frontier program and requires a Microsoft 365 Copilot Premium license.

What Are the Key Capabilities of Copilot in Power BI?

Key Copilot capabilities for business users, as well as BI analysts and developers.

Functional area	How does Copilot help in practice?
Business and end users
Natural language questions	Querying models without knowing table names and receiving answers in the form of text summaries, tables, or charts.
Executive summaries	Generating narrative business insights (Narrative Visual). After filters are changed, the user refreshes the summary to reflect the current data context.
Quick insights without opening reports	Checking information and finding relevant content from Copilot Hub or Microsoft 365 Copilot without manually searching for a specific report.
Explaining visualizations	Providing clear explanations of the context and relationships shown in rendered charts and tables.
BI developers and analysts
Report design	Automatically creating prototypes and complete page layouts with visualizations based on a text description.
Working with DAX code	Generating measures from scratch, optimizing and fixing errors, explaining and commenting on logic in DAX Query View.
Model documentation	Automatically generating business descriptions for measures and other model elements in the properties section.
AI vocabulary and synonyms	Mapping everyday language and company-specific terminology to technical field names in the semantic model.

How Does Copilot Work?

Whether you enter a prompt in DAX Query View in Power BI Desktop, use Copilot Hub, query an open report, or work with Power BI data through Microsoft 365 Copilot, the overall logic is similar. However, the technical details differ between individual features, so the five-step process below should be treated as a simplified flow model rather than an exact description of Microsoft’s internal implementation.

Step 1: Input and Context

The process begins with a user prompt or action. Copilot takes into account the environment in which the user is working, the current session, and the user’s permissions to reports, semantic models, and data.

the text of the user’s question or instruction,

the context of the Copilot experience currently in use,

the history of questions and answers from the current session, to the extent supported by the specific feature.

Step 2: Preprocessing and “Grounding”

Before the language model prepares a response, Copilot gathers business context from the surrounding environment. This process is called grounding, meaning that the response is anchored in the data and metadata available to the user.

When working with DAX and a semantic model, the context can include the schema of tables, columns, relationships, data types, measures, descriptions, and other model properties. In DAX Query View, it can also include the contents of the current query window – DAX code, comments, and previously generated queries.

In experiences designed for asking questions about data, indexed text values and elements prepared through Prep Data for AI can also be used.

In Microsoft 365 Copilot Chat, the response can additionally be combined with Microsoft 365 context, provided that the user has the appropriate permissions and the requirements for Fabric IQ are met.

Copilot operates within the user’s permission context. This means that data access controls such as Row-Level Security remain part of the response process.

Step 3: Processing by the Language Model

The prompt, enriched with grounding information, is processed by Azure OpenAI services used by Microsoft. The scope of data passed as context depends on the feature – it may include model metadata, selected properties, specific data points, or query results required to prepare the response.

Microsoft states that organizational data processed by Copilot is not used to train base AI models for other customers.

Step 4: Validation and Postprocessing

The response is then subject to security and processing mechanisms appropriate to the specific feature. Copilot uses Responsible AI filters as well as controls resulting from the user’s permissions. In DAX Query View, postprocessing uses a DAX parser to validate the generated query. In other experiences, the result can be transformed into a visualization, a model change, or a text response.

Generated responses, code, and model changes still require human review. Copilot is a nondeterministic tool and does not guarantee that every suggestion will be correct.

Step 5: Result for the User

The final form of the result depends on where the question was asked. It can be a generated DAX query, a suggested measure description, a new or modified report page, a visualization, a table, or a text response in the conversation window.

Simplified query processing cycle in Copilot for Power BI.

Where Does Copilot Get the Data It Uses to Answer a Question?

Copilot does not receive one fixed set of data for all features. Instead, it dynamically builds a business and technical context depending on where it is invoked and the type of task being performed. In some scenarios, the context is based primarily on model metadata, while in others, specific data points or query results needed to generate and explain the response may also be passed to Azure OpenAI.

A. Core Foundation – Semantic Model Metadata

Metadata forms the primary context layer when working with a semantic model. Depending on the feature, Copilot can use:

table, column, and DAX measure names, data types, relationships, and calculation groups,

field properties, including descriptions, data categories, formats, and DAX expressions; in certain scenarios, Copilot uses the first 200 characters of a description,

synonyms and the linguistic schema, which help map business language to technical names in the model.

B. Data Sampling and Indexing

To better recognize specific values that appear in the data – such as product names, cities, or customers — Power BI can create an index of text values used by question-answering mechanisms.

Indexing has an upper limit of 5 million value instances in the model; columns with lower cardinality are indexed first.

Power BI does not index text values that are 100 characters or longer.

For large models, there is also a limit on the number of model entities included in indexing.

In certain scenarios, Copilot can use statistical aggregations such as minimum and maximum column values, as well as query results needed to explain an answer.

C. Developer Controls – Prep Data for AI

A manually configured model preparation layer can reduce misinterpretation of data and improve the precision of responses.

AI data schema – allows the author to define which model elements should be used by Copilot and limit the use of technical fields that should not appear in business responses.

AI instructions – additional instructions describing business rules and how data should be interpreted, for example the definition of a high season or the correct measure to use for a specific type of question.

Verified answers – human-prepared and approved answers or visualizations associated with specific questions.

Descriptions and synonyms – help explain the meaning of fields and map terminology used within the organization to the structure of the model.

D. Interface and Current Session Context

The scope of context depends on the Copilot experience. When answering questions about report content, Copilot can use metadata from the current page and information from visualizations, including data behind supported filters and slicers. In Copilot Hub, it uses the information required to find content available to the user, while in DAX Query View it uses the semantic model schema and the contents of the current query window.

It is important to distinguish questions about report content from the Ask data questions feature: visualizations generated in the latter mode do not automatically inherit filters and slicers currently affecting visualizations on the report page.

Copilot can also use the history of the active conversation. This is session memory rather than persistent memory of all previous conversations.

E. Advanced Enterprise Context – Fabric IQ and Microsoft 365

In Microsoft 365 Copilot Chat, responses can be grounded in Power BI data through Fabric IQ and combined with the broader context of the user’s work. Fabric IQ provides business data context based, among other things, on semantic models and OneLake, while Work IQ contributes context related to people, collaboration, and Microsoft 365 workflows.

Access to data continues to rely on the user’s existing permissions to reports and semantic models. Fabric data agents are a separate mechanism and must be explicitly published to Microsoft 365 if they are to be used in Copilot Chat.

The “Share Fabric data with your Microsoft 365 services” setting affects the discoverability of Power BI content in Microsoft 365, but it is not required for every way of referencing a report – users can also point to a report directly, for example by name or link.

What Questions and Tasks Can Copilot Not Handle?

To avoid unrealistic expectations within an organization, it is important to understand the tool’s limitations. Some of these constraints relate to AI reasoning itself, while others result from the Power BI capabilities currently supported.

Key limitations of Copilot in Power BI.

Area	Can Copilot do it on its own?	Limitation
Deep causal analysis	×	Copilot does not support data questions that require generating new, deep insights, such as “Why do sales decline every July?”. It can help identify patterns or relationships visible in the data, but determining the actual business cause still requires analysis and domain expertise.
Forecasting future outcomes	×	Copilot does not natively answer questions that require forecasting, such as “How much will we sell next year?”. Such a result requires an appropriate model, measures, or a separate analytical method.
Questions about data outside the semantic model	×	If a question concerns organizational data, Copilot bases its answer on the available semantic model. If the question is unrelated to the model’s data, it may answer using general LLM knowledge, but such an answer is not based on the organization’s data.
Queries blocked by security filters	×	Built-in security mechanisms can reject content that violates Responsible AI policies or attempts to bypass security controls and permissions.
Models with unsupported architectures	×	Copilot cannot create report pages for models that use real-time streaming or models connected live to Analysis Services. Additional limitations also apply to areas such as custom visuals and styling changes.

Requirements and Licensing – What Do You Need to Get Started?

For Copilot to work across the touchpoints discussed above, an organization must meet requirements related to capacity, user licensing, administrative settings, and data access permissions.

Cloud Capacity and AI Billing

The primary requirement is a paid Microsoft Fabric F2 capacity or higher, or Power BI Premium P1 or higher, with Copilot enabled. Power BI Pro or Premium Per User licenses alone do not directly provide the required Copilot capacity.

Fabric Copilot Capacity (FCC) enables centralized billing of Copilot usage for assigned users. It can cover, among other things, Power BI Desktop and Copilot usage in Pro, PPU, and Fabric capacity-based workspaces.

In Power BI Desktop, the user must have Admin, Member, or Contributor access to an appropriate workspace backed by a supported capacity, or be assigned to Fabric Copilot Capacity. The selected Copilot workspace does not have to be the final destination where the report is published.

Using AI features consumes Fabric capacity compute resources. Tokens processed by Copilot are converted into Fabric Capacity Units (CU) usage; there is no separate charge for each individual question.

User Licenses

The required user license depends on the user’s role, the workspace mode, and how content is shared. Report authors and collaborators typically use Power BI Pro or PPU. In certain configurations, consumers can access content without a Pro license if it is shared from an appropriate Premium/Fabric capacity in accordance with Power BI licensing rules.

To use Power BI data through Fabric IQ in Microsoft 365 Copilot Chat, users need a Microsoft 365 Copilot Premium license, Frontier access, and licensed access to the reports and semantic models they are querying.

Administrative Settings

„Users can use Copilot and other features powered by Azure OpenAI” – the primary setting in the Fabric admin portal.

„Users can access a standalone, cross-item Copilot in Power BI experience (preview)” – required for Standalone Copilot.

„Show Copilot in app navigation” – the app setting required if Copilot should be available in the app navigation.

„Fabric data in Microsoft Copilot” – a Microsoft 365 setting that controls the use of Fabric/Power BI context in Microsoft Copilot.

„Share Fabric data with your Microsoft 365 services” – a setting that affects the discoverability of Power BI content in Microsoft 365.

Consent for cross-region processing (Cross-Geo) is required only when Azure OpenAI is not available in the geographic region of the capacity. It should not be treated as a mandatory setting for every European environment.

Permissions and Security

The user must have the appropriate permissions to the report or semantic model. The required level of access depends on whether the user only reads data or also edits the report or model.

Copilot respects existing Power BI security mechanisms, including Row-Level Security, and operates within the user’s permission context.

Key requirements for the different ways of using Copilot in Power BI.

User / environment	Tasks	Capacity / license	Settings
Power BI Service Individual report	Answers questions about the report and semantic model, generates summaries, and can create visualizations based on data questions.	Paid F2+ or P1+ capacity, or usage billed through an assigned Fabric Copilot Capacity (FCC). The user license depends on the workspace mode and how the content is shared.	“Users can use Copilot and other features powered by Azure OpenAI” + a supported region and appropriate permissions.
Power BI Service App Preview	Helps users find information across the reports and models shared in the app and directs them to the appropriate report.	Paid F2+ or P1+ capacity, or FCC; an appropriate license to access the content.	Primary Copilot setting + “Show Copilot in app navigation”.
Power BI Service Standalone Copilot Preview	Searches for reports and semantic models available to the user, allows them to be added as context, and supports asking questions without opening a report.	Paid capacity that supports Copilot or FCC.	Primary Copilot setting + “Users can access a standalone, cross-item Copilot in Power BI experience (preview)”.
Microsoft 365 Copilot Chat Fabric IQ Frontier	Allows users to ask questions about Power BI data in Microsoft 365 Copilot Chat and combine the response with the context of files, messages, and conversations.	Microsoft 365 Copilot Premium + Frontier + licensed access to the relevant Power BI reports and models.	“Fabric data in Microsoft Copilot”; “Share Fabric data with your Microsoft 365 services” affects content discoverability.
Power BI Desktop	Creating and editing report pages, working with DAX, generating measure descriptions, and supporting work with the semantic model.	F2+/P1+ with Admin, Member, or Contributor access to the appropriate workspace, or assignment to FCC.	Primary tenant-level Copilot setting + access to a supported capacity.
Power BI Service Model editing view Preview	Model analysis and editing: tables, columns, measures, relationships, DAX, RLS roles, descriptions, and display folders.	Supported capacity or FCC, plus Write permission to the active semantic model.	Primary Copilot setting + permission to edit semantic models in Power BI Service.

Due to the rapid development of the technology described in this article, the specifications and technical information presented here reflect the state of knowledge as of the publication date, September 11, 2026.

Summary and Next Steps

Copilot in Power BI can reduce the time required to build reports, make working with DAX easier, support model documentation, and help business users interact with data more efficiently. However, success depends not only on enabling AI features, but also on properly preparing semantic models, descriptions, synonyms, business rules, and permissions.

The better structured the model and the more clearly defined the business context, the greater the chance of obtaining useful responses. In practice, Copilot should therefore be treated as part of a broader Power BI and Microsoft Fabric architecture, rather than as a standalone add-on to existing reports.

Planning an implementation or looking to strengthen your team’s skills?
Visit powerbicourse.com and explore the dedicated Power BI Copilot training program.

Do You Need Architectural Support?

Contact the Antdata team. We help companies audit Power BI environments, prepare semantic models for AI, optimize Fabric capacity usage, and implement solutions tailored to real business processes.

Choose a date for a phone consultation.