Power BI Copilot: What It Is & How To Use It In 2026

Power BI Copilot is Microsoft's AI assistant inside Power BI: it summarizes reports, answers plain-English questions about a semantic model, writes DAX and drafts report pages. It only runs on a paid Fabric capacity (F2 or larger) or a Power BI Premium capacity (P1 or larger), so a Pro or Premium Per User license on its own doesn't switch it on, and an admin has to keep the Copilot tenant settings enabled. In practice the cheapest route is an F2 capacity at about $262.80 a month pay-as-you-go, on top of the Pro ($14) or Premium Per User ($24) seats your people already need (per user a month, billed annually), as of October 2026. How good the answers are depends mostly on how well the semantic model behind them is prepared.

I checked the requirements and prices below on October 29, 2026. Copilot changes monthly, so every claim links its source.

Power BI Copilot at a glance	As of October 2026
Cheapest entry	F2 at about $262.80 a month pay-as-you-go, or about $156 on a one-year reservation
How usage is billed	Tokens converted into capacity units: 100 CU-seconds per 1,000 input tokens, 400 per 1,000 output tokens
Main limits	Answers are only as good as the model; the report pane sees only the open report; no forecasting, anomaly detection or key-influencer questions; English only
Power BI Q&A	Being retired by the end of December 2026; Microsoft points Q&A users to Copilot

What Copilot in Power BI is: four places it shows up

Power BI has had AI visuals such as key influencers and the decomposition tree for years; Copilot is its generative AI feature, and "Copilot in Power BI" is one brand over several experiences with different scopes and maturity. Knowing which one you're looking at explains most of the confusion in forum threads.

The Copilot pane in a report

The panel next to a report in the Power BI service summarizes the report and answers questions only about the report you have open (Microsoft Learn: standalone Copilot). Filters and slicers set on the page don't apply to its answers, so a user looking at "EMEA, this quarter" can get a global, all-time number, and English is the only supported language (Microsoft Learn: ask data questions with Copilot and Fabric IQ).

Standalone Copilot (preview)

A full-screen Copilot that finds and answers questions about any report, semantic model or Fabric data agent you have access to. It is still preview, isn't available yet in several Asian and European regions, and has its own tenant setting, "Users can access a standalone, cross-item Power BI Copilot experience" (Microsoft Learn: standalone Copilot). Microsoft turned that setting on by default, starting on or after September 5, 2025, for tenants that already had Copilot enabled (Microsoft Learn: enable Copilot in Power BI).

Copilot for report authors

Copilot drafts report pages from a prompt and writes or explains DAX queries in DAX query view (Microsoft Learn: DAX query view with Copilot). Page creation doesn't work on semantic models that use real-time streaming, on live connections to Analysis Services, or on models where implicit measures are disabled (Microsoft Learn: create reports with Copilot).

Power BI inside Microsoft 365 Copilot

Microsoft's September 2026 feature summary made Power BI data answers in Microsoft Copilot Chat generally available: people attach a report or paste a link and ask questions, answers use the model's existing measures, existing Power BI permissions still apply, and Purview data loss prevention policies are supported. The same release added a Microsoft 365 Copilot experience inside Power BI in preview, off by default and limited to people with a Microsoft 365 Copilot license (Power BI September 2026 feature summary; eigenx plain-language summary).

Power BI Copilot requirements

Capacity: F2 or P1, paid

Copilot needs a paid Fabric capacity (F2 or higher) or Power BI Premium capacity (P1 or higher) in a supported region. Trial capacities, free SKUs and sovereign clouds are not supported (Microsoft Learn: Copilot for Power BI overview). Older guides that say F64 predate Microsoft lowering the entry point to F2. Power BI Desktop is no way around this: Copilot in Desktop needs write access to a workspace on a paid Fabric or Premium capacity.

Licenses: Pro and PPU need a Fabric Copilot capacity

Pro and Premium Per User workspaces don't support Copilot directly (Microsoft Learn: enable Copilot in Fabric). Microsoft's route is a Fabric Copilot capacity: an F2 or P1 (at minimum) that collects and bills one group's Copilot usage, including usage from Power BI Desktop and from Pro and PPU workspaces. A Fabric admin enables it, the capacity admin assigns users, and it only works in the tenant's home region (Microsoft Learn: Fabric Copilot capacity). So "Copilot for Pro users" still means buying a capacity, with the $14 Pro or $24 PPU seats on top.

How to enable Copilot in Power BI

1. Put the workspace on a paid F2+ or P1+ capacity, or have Pro and PPU users assigned to a Fabric Copilot capacity.
2. In the admin portal's tenant settings, check that "Users can use Copilot and other features powered by Azure OpenAI" is on. It is enabled by default, but your admin may have turned it off (Microsoft Learn: Copilot for Power BI overview).
3. If your capacity is outside the US or EU, an admin also has to let data sent to Azure OpenAI be processed outside your capacity's geographic region, compliance boundary or national cloud instance. That is a compliance decision, so loop in whoever owns data residency.

What Power BI Copilot costs in practice

There is no Copilot line item. Copilot spends your capacity, metered as tokens converted into capacity-unit seconds: 100 CU-seconds per 1,000 input tokens, 10 per 1,000 cached input tokens, and 400 per 1,000 output tokens (Microsoft Learn: Copilot consumption in Fabric).

Microsoft's own worked example, a request with 2,000 input tokens and 500 output tokens, comes to 400 CU-seconds: about two cents of capacity at the US West 2 pay-as-you-go rate of $0.18 per CU-hour (Microsoft Fabric blog: Copilot pricing example). An F2 has 172,800 CU-seconds a day, so it could absorb roughly 430 requests of that size if it did nothing else. Treat that as a floor: Copilot draws on the same units as your refreshes and report queries, and heavy prompting can push a small capacity into throttling, so watch the Fabric Capacity Metrics app (diyai Power BI Copilot review). Prices also vary by region (Microsoft Learn: Azure billing for Fabric).

Common setups at US list prices as of October 2026, seats billed annually:

Setup	Monthly cost	What to know
10 people on Pro, plus an F2 as the Fabric Copilot capacity	About $403 (10 × $14, plus $262.80 pay-as-you-go)	A typical small-team setup
10 people on Premium Per User, plus an F2 Copilot capacity	About $503 (10 × $24, plus $262.80)	PPU doesn't include Copilot; the capacity is still needed
Workspaces already on F64	About $5,000 reserved or about $8,410 pay-as-you-go, plus Pro seats for authors	Viewers need no seat at F64. Copilot competes with every other workload for the same units

Microsoft announced a zero-provisioned F0 SKU with on-demand billing at FabCon Europe 2026; as of October 2026 it is not generally available and its price hasn't been announced (Solv: Fabric F0 and on-demand billing). For the full seat and capacity math, see our Power BI pricing guide and Microsoft Fabric pricing guide.

Where Power BI Copilot falls short

The semantic model decides the answer

Copilot reads your model's names, descriptions and relationships. If a measure is called Measure 7 and nobody wrote a description, Copilot guesses. Microsoft's answer is "Prep data for AI", still in preview: an AI data schema that limits which fields Copilot considers, verified answers that tie trigger phrases to visuals an author has checked, and AI instructions that add business context (Microsoft Learn: prepare your data for AI). It works, but it is hand-written work per model. Our guide to the Power BI semantic model covers what to fix first.

One report or model at a time

The standalone experience can find any model you have access to, but a Fabric community answer says Copilot doesn't query across multiple semantic models simultaneously (Fabric community thread). That reply is about a year old, so test it, but plan on one model per answer: if "revenue" lives in the Finance model and "pipeline" in the Sales model, Copilot won't join them for you.

Questions it won't answer

Microsoft lists as unsupported anything that needs new insight generated, such as anomaly detection, forecasting or finding key influencers, with examples like asking why sales fall every July or for next year's sales forecast (Microsoft Learn: ask data questions with Copilot and Fabric IQ). Those are the questions a leadership team asks after the number moves.

Smaller things that trip people up

• Data questions work only in the Power BI service for now; Desktop support is listed as coming later.
• Copilot's data questions need Q&A enabled on the model, because they use the same engine.
• When Fabric IQ is enabled, Copilot doesn't return verified answers.
• Every surface is a Microsoft one. If your company runs on Slack, or the people asking never open Power BI, Copilot isn't where they are.
• Because Copilot spends shared capacity, its cost shows up as throttling or a capacity upgrade, not per question.

The first three come from Microsoft's page on asking data questions.

Fabric data agents, Fabric IQ and the end of Q&A

Fabric data agents

A Fabric data agent is a conversational agent a data team builds and publishes for a domain. Unlike the Copilot pane, it can sit over several sources: lakehouses, warehouses, Power BI semantic models, KQL databases, mirrored databases and ontologies, up to five per agent. It needs a paid F2 or higher Fabric capacity, or Premium P1 or higher with Fabric enabled; the data agent tenant settings; cross-geo settings where your region needs them; and read access to each source. Row-level and column-level security still apply on semantic models (Microsoft Learn: create a Fabric data agent).

Agents can be used from Microsoft 365 Copilot, which adds a Microsoft Copilot license or Office 365 commercial subscription plus user licenses for each person using the agent (Microsoft Learn: data agent in Microsoft Copilot), and from Microsoft Foundry agents (Microsoft Learn: Fabric data agent in Foundry).

Fabric IQ

Fabric IQ is Microsoft's business-context layer for Fabric and OneLake data: an ontology (business entities, how they relate, the rules that govern them), Power BI semantic models, a graph, and data and operations agents (Microsoft Fabric IQ). It is not a chat window but the context Copilot, data agents and outside agents draw on. Microsoft announced it as generally available at Build in June 2026, but Microsoft Learn still labels the IQ workload preview, and the ontology item is still in preview as of October 2026 (Atlan: what is Fabric IQ). Fabric IQ in Microsoft Copilot Chat and Copilot Cowork was announced as generally available at FabCon Europe 2026 (Solv: FabCon Europe 2026 recap).

Power BI Q&A is being retired

Microsoft's Message Center notice MC1218421 says Q&A experiences and Q&A Setup will be fully retired by the end of December 2026. Existing Q&A visuals in reports, dashboards, mobile and embedded scenarios will stop working and be removed, Q&A Setup tools such as synonyms and "teach Q&A" go too, and Microsoft recommends replacing Q&A visuals with Copilot (MC1218421; Power BI blog: deprecating Q&A). The catch: Q&A didn't need a capacity, and Copilot does. A Pro-only tenant that relies on Q&A should price an F2, or another route, before January 2027.

Power BI MCP servers

Microsoft also exposes semantic models to outside AI assistants through MCP servers, which our Power BI MCP server guide covers.

Alternatives to Copilot for Power BI on your existing models

These are other ways to put AI on the semantic models you already have. Moving off Power BI itself is a different decision, covered in our Power BI alternatives guide.

• Fabric data agent: the Microsoft-native answer when Copilot's one-model scope is the problem. Choose it when your data already lives in OneLake, your users live in Microsoft 365, and someone will own the agent like a product.
• MCP server plus an AI assistant: fits analysts and developers already in Claude, VS Code or GitHub Copilot. Microsoft's recommended server, Fabric IQ MCP, is generally available and read-only, but each person's assistant and its data terms sit in the path of your company data, and unless the model carries your business definitions, the assistant starts from table and measure names alone.
• Microsoft 365 Copilot Chat: no new tool if your company already pays for Microsoft 365 Copilot, but it inherits Copilot's dependence on model quality.

GetDot: your semantic model in Slack, Teams and email

GetDot is an AI data analyst that connects to your Power BI semantic models, and to the warehouse under them, and answers in Slack, Microsoft Teams, email, its web app or any MCP client. From our Power BI integration page and release notes:

• It reads the model and lets Power BI do the math. Dot signs in with Microsoft, reads each model's tables, columns, measures (with their DAX and descriptions) and relationships over the XMLA endpoint, and sends DAX to the Power BI REST API's executeQueries endpoint, so your measures compute the numbers.
• It reads reports too. Dot can return the exact number a visual shows, re-run that visual for another period or slice, explain how a measure is built, and trace each table back to the warehouse table named in its Power Query. Decomposition tree, key influencers, Q&A, R/Python and custom visuals are blocked rather than approximated, because no DAX query reproduces their result.
• It goes past one model and past DAX. Dot can run SQL against a connected warehouse in the same analysis and uses Python for analysis DAX can't express. Root Cause Analysis runs multi-step "why did this move" investigations, the question type Copilot's docs exclude.

Dot is not a way around Premium: XMLA exists only on Premium, Premium Per User and Fabric capacity, so models in Pro-only workspaces are not indexed. A Power BI admin must enable the "Dataset Execute Queries REST API" tenant setting, and Dot's Microsoft account must be a workspace member (Contributor or higher to read reports). The practical difference from Copilot is that PPU is enough, with no F2 to add. For Pro-only workspaces, Dot can query the warehouse under them instead, through connectors such as Microsoft Fabric.

Permissions work differently than in Copilot. Power BI evaluates its row-level security against the account that authenticated the connection, not the person asking in Slack, so on the Team plan an admin sets Dot's row-level rules per model, and Dot refuses a query when a rule can't be applied. Dot is SOC 2 Type II and GDPR compliant, and its LLM providers run with zero data retention.

Our March 2026 release added Microsoft 365 Copilot and GitHub Copilot as MCP clients (changelog), and Claude, Cursor and Codex work too (MCP integration). Since September 2026, Dot can also rebuild a Power BI report as a Dot dashboard and compare the two visual by visual. The full walkthrough is in how Dot uses your Power BI semantic models.

Pricing as of October 2026: Dot bills on credits, not seats, with unlimited users. Free is $0 with 300 one-time credits and full Pro features. Pro is $180 a month billed annually ($200 month-to-month) with 150 credits a month; extra credits are $1.80 on annual billing or $2.00 monthly. Team is $720 a month billed annually ($800 month-to-month) with 800 credits a month; extra credits are $1.44 annual or $1.60 monthly. Enterprise is custom.

Where Copilot wins: Dot reads Power BI but never writes back to it, so authors drafting pages and DAX need Copilot; Power BI has native mobile apps where Dot works in a mobile browser; and Dot is one more vendor to put through procurement and security review.

Choose Copilot when…

• You already run a Fabric F2+ or Premium P1+ capacity with headroom to spare.
• Your users work in the Power BI service and Microsoft 365, not Slack.
• Your semantic models are documented, and someone will do the Prep data for AI work per model.
• Most questions are "what" and "how much" on one model at a time.
• Security wants no new vendor in the path of your data.

Choose something else when your questions span several models or reach the warehouse, people ask "why" more than "how much", the people asking work in Slack, email or another AI assistant, or you would rather pay per use than size a capacity. You can try Dot on your Power BI models for free, or see the side-by-side in our Power BI Copilot vs Dot comparison.

FAQ