Power BI Copilot “Something Went Wrong” Error (2025 Troubleshooting Guide)

The message appears when Copilot cannot complete its request in your tenant. Common reasons include a workspace on shared capacity, Copilot disabled in tenant settings, or a capacity region that cannot reach Azure OpenAI.

Follow four steps in order: move the workspace to Fabric F2+ or Premium P1+, enable Copilot and Azure OpenAI in tenant settings, confirm the capacity region is supported, and test Copilot against a small clean model. When all four checks pass, this error usually disappears.

This wording still points to configuration, not to a random glitch. Treat it like the main error and validate capacity, tenant settings, and region support instead of only retrying.

Closing the pane only refreshes the UI. If the message returns, the root cause sits in capacity, tenant policy, or region and needs an admin fix.

Smart Narrative uses Copilot. Failures usually mean the workspace is not on Copilot-ready capacity or the model is too large or complex for the current capacity tier.

This appears when Copilot cannot map your text to fields in the model. Use shorter prompts, reference table or column names, and prefer clear business terms over internal jargon.

Enabling Fabric is only the first step. You still need at least one workspace on Fabric or Premium capacity, Copilot toggles enabled in tenant settings, and data-movement consent where Azure OpenAI requires it.

If Copilot fails everywhere, treat it as a tenant issue. Ask an admin to review Copilot tenant settings, Azure OpenAI permissions, and capacity regions for the whole environment.

A disabled icon usually means your sign-in does not meet requirements. Confirm you have Pro or PPU, access to a Copilot-enabled workspace on F or P capacity, and a tenant where Copilot is turned on.

Missing buttons often point to shared capacity or unsupported region. Move the report to a workspace on Copilot-ready capacity and ask an admin to confirm Copilot is allowed for your security group.

Yes, Copilot runs only on dedicated capacity. A workspace must sit on Fabric F2 or higher, or on Premium P1 or higher, before Copilot can process prompts.

F2 is the entry tier for Fabric Copilot capacity and works for lighter workloads. Premium P1 aligns with an F64-level capacity and suits larger, multi-team deployments.

A Pro license alone does not unlock Copilot. You also need a workspace on Copilot-capable capacity; Copilot runs against that capacity, not against the shared pool for Pro.

PPU gives richer features but still depends on capacity. Copilot works when PPU content lives in a workspace that also runs on Copilot-enabled capacity.

Shared capacity is tuned for regular reports and dashboards. Copilot workloads run only on dedicated capacities where Microsoft can control performance and isolation.

Open workspace settings and look at the Premium or Advanced section. If it shows assignment to an F-series or P-series capacity and your tenant has Copilot enabled, the workspace is Copilot-ready.

Fabric admins can open the Copilot Capacity page in the admin portal to see capacity size, region, linked workspaces, and allowed users or groups. That page is the single source of truth.

In the Fabric or Power BI admin portal, go to Tenant settings and find the Copilot and Azure OpenAI section. Switch it on and scope it to the organization or to specific security groups, then save.

You must enable Copilot and Azure OpenAI, allow data processing in the Azure OpenAI regions your tenant uses, and optionally allow preview features if Copilot is still in preview for your org.

Many organizations pause Copilot until they finish security, privacy, or cost reviews. Ask the admin whether a pilot Copilot group exists and how you can be added.

Copilot sends prompts and metadata to Azure OpenAI. If tenant policies block that integration, Copilot fails regardless of workspace capacity or licenses.

This option controls whether Copilot can send data to Azure OpenAI for processing. Turning it off disables many Copilot features; enabling it for a scoped group provides control and flexibility.

When Copilot features run as previews, an admin must allow preview use and users may need to accept terms. If either piece is missing, Copilot icons can appear but requests still fail.

Copilot rolls out gradually to different regions. If your capacity region is not yet listed as supported, Copilot either stays hidden or offers only limited functionality.

The list appears in the official Copilot documentation on Microsoft Learn and changes over time. Always check that list and compare it with your capacity region before troubleshooting further.

You need a new capacity in the target region and a new workspace that uses it. After that, migrate or republish datasets and reports into that workspace.

Yes. If policy blocks data from leaving a specific geography, Azure OpenAI may not receive prompts, and Copilot fails. Many tenants solve this by allowing controlled data movement for a dedicated Copilot group.

Very large or complex models can hit capacity limits or timeouts when Copilot analyzes them. Reducing table size, pruning unused columns, and simplifying relationships usually improves stability.

Use a clean star schema, clear table and column names, and only the measures you need. A tidy model gives Copilot enough structure to understand your data and generate accurate summaries.

DirectQuery depends on live source performance, so slow queries or gateways can cause Copilot actions to time out. Testing the same model as Import, or limiting heavy DirectQuery tables, often makes Copilot more dependable.

Create a dataset with one small fact table and a few dimension tables, all with friendly names. Publish a report in the same workspace and run Copilot there to confirm that the environment can handle basic prompts.

Admins open the Fabric Capacity Metrics app, filter by capacity, and check CPU, memory, and throttling around the time of Copilot failures. Spikes or constant throttling point to capacity upgrades or workload tuning.

A non-admin can test Copilot in a private browser, disable extensions, and build one tiny test report in the same workspace. If every attempt still fails, the next step is to involve an admin with exact timestamps and workspace details.

Open a ticket after you confirm capacity, tenant settings, region, and a simple test model all look correct. Include screenshots, failure times, and capacity metrics so support can see what happens under the hood.

If multiple users across regions and workspaces see the same error and Microsoft’s service health page reports an incident, Copilot might be down. When only your tenant fails, or only certain workspaces fail, configuration is the more likely cause.