Power BI Q&A is going away—do not migrate only the question box

Microsoft now says the Power BI Q&A visual is scheduled for deprecation in December 2026. The visual currently lets report users type natural-language questions and receive a generated visualization. That gives teams a deadline, but it does not make the replacement a one-click interface change. Microsoft’s visualization overview states the date and current behavior.

Preserve the work behind the box

A useful migration inventory starts with the deployed Q&A visuals, the semantic models behind them and the questions people actually ask. It should also capture measures, synonyms, hidden fields, row-level security, expected answers and known failure phrasings. Spotonix’s migration guide makes the practical point that teams need to preserve business questions, semantic preparation, access rules, expected results and the human review path—not merely recreate a text field. The guide also cautions that Copilot is not automatically equivalent to Q&A. Its proposed migration process freezes a baseline, prepares a target, tests meaning and access, then releases workflow by workflow.

That advice aligns with Microsoft’s own Copilot preparation model. Power BI’s Prep data for AI workflow stores three kinds of context on the semantic model: AI data schemas, verified answers and AI instructions. Authors can attach trigger phrases to verified visuals, test changes in the Copilot report pane and inspect diagnostics and “How Copilot arrived at this” information. Microsoft explicitly says this preparation cannot guarantee the same output every time because Copilot is nondeterministic. The product documentation describes those controls and the testing workflow.

Run the old and new paths against the same fixture

The defensible cutover test is therefore paired, not anecdotal. Run a representative question set through the existing Q&A experience and the proposed Copilot configuration against the same semantic model and user identities. Compare the selected metric, population, period, filters, grain, returned values and access outcome. Record where either path needs a rephrased question or analyst intervention.

Do not use an “answer appeared” check as the release gate. A fluent answer can still apply the wrong measure or scope. Microsoft’s tooling supports iterative testing, verified answers and diagnostics, but its nondeterminism warning means the team still needs repeated runs and reviewer-owned expected results. Marking a model Approved for Copilot changes how the standalone experience treats that model; it is not an accuracy certificate. Microsoft documents both the approval setting and its effect.

The December date should trigger discovery now: find every Q&A surface, establish owners and build the fixture before users lose the old path. The migration is complete only when the replacement preserves meaning, permissions and reviewability for the workflows that depended on Q&A—not when a new chat box renders.