Power BI: Add Synonyms for Smarter AI

Key insights

• In a YouTube video, Greg Trzeciak shows how adding Copilot-friendly synonyms helps the assistant understand the language your organization uses and reduces “I don’t understand” responses.
  
• He demonstrates where to manage terms: open the Model view → Q&A setup in Power BI (or Fabric) to edit synonyms for tables, columns, and measures.
  
• Using the Synonyms pane, Greg runs a baseline test (total sales by reseller) and a failure test (total sales by “friends”), then adds synonyms like “friends” or “bros” to fix interpretation without changing the data model.
  
• To confirm outputs, he inspects the generated DAX and the Copilot explanation so you can validate how the answer was produced and trust the result.
  
• Practical tip: Avoid ambiguous synonyms by choosing clear, organization-specific terms and scope synonyms to the right fields to prevent incorrect matches.
  
• Benefit summary: adding synonyms is a small, Prepped for AI step that delivers high impact—better Copilot accuracy, faster user adoption, and fewer support requests.
  

Overview of the video

Pragmatic Works published a YouTube video in which Greg Trzeciak demonstrates how to make semantic models more conversational-ready for Copilot. The video focuses on using Synonyms in the Q&A configuration so that natural language queries map to the correct tables, columns, and measures. Greg frames the process as a way to reduce “I don’t understand” responses and to teach Copilot the language your organization uses. As a result, business users get more accurate answers with less frustration.

Walkthrough and practical demo

Trzeciak opens the Copilot experience inside a Microsoft Fabric workspace and runs a baseline query that succeeds, such as “total sales by reseller.” He then shows a failing query—“total sales by friends”—to highlight how the AI cannot connect casual terms to model elements. Next, he uses the Synonyms pane from Model view → Q&A setup to add alternatives like “friends” and “bros” to the reseller field, and then reruns the prompt. Finally, he validates Copilot’s revised answer by checking how the result was generated, including the returned DAX expression.

Where to set up and why it still matters

Although the Q&A visual will be deprecated in December 2026, Trzeciak emphasizes that the underlying Q&A setup remains critical because it teaches AI how to interpret your semantic model. You manage Synonyms in the Model view under Q&A setup, and these mappings apply to tables, fields, and measures without changing the data model itself. This preparation helps Copilot and other data agents understand everyday language used across the business. Therefore, preparing models for AI is not just a temporary tweak but a lasting part of model governance and usability.

Benefits and immediate impacts

Adding synonyms tends to produce quick wins: users ask questions in plain language and receive relevant answers more often, which lowers support overhead. Furthermore, the ability to check the DAX behind an answer gives analysts confidence and a way to audit how results were derived. By contrast, leaving a model unprepared leaves Copilot guessing, and that increases the chance of incorrect or incomplete outputs. Thus, synonyms improve both user experience and traceability.

Tradeoffs and operational challenges

While synonyms can greatly improve interpretability, they also introduce new responsibilities for model owners: maintaining synonyms at scale can become time-consuming and may require cross-team agreement on terminology. If teams add ambiguous synonyms, the AI might map the same term to multiple fields, producing incorrect answers; therefore, careful curation matters. Automated generation of synonyms speeds initial rollout but often needs manual refinement to avoid collisions and preserve data integrity. Balancing automation with human review provides the best path forward, though it requires disciplined processes and governance.

Validation and governance practices

Trzeciak shows practical validation steps, such as rerunning prompts after edits and inspecting the generated DAX, which exposes how Copilot translated the query. Organizations should adopt testing routines that cover common queries and edge cases, prioritizing high-value measures and popular reports first. Additionally, documenting synonyms and the rationale for changes helps prevent confusion across teams and supports compliance needs. With a lightweight review cadence, teams can keep synonyms useful without letting them drift into inconsistency.

Recommendations for rollout

Start small by identifying a few critical measures or tables that business users query frequently, then add synonyms for those items and validate results with Copilot. Next, observe query logs and user feedback to guide incremental expansion while avoiding blanket synonym additions that could create ambiguity. Training power users to inspect returned DAX and to flag mismatches accelerates improvement and distributes stewardship across the organization. Over time, this iterative approach reduces support tickets and makes conversational analytics more reliable.

Conclusion: simple steps, high impact

Greg Trzeciak’s demonstration underscores that adding Synonyms is a low-effort, high-impact step toward making semantic models AI-friendly. Although the upcoming removal of the Q&A visual changes the UI, the core benefits of teaching Copilot your organization’s language remain important. By balancing automated suggestions with human curation and by validating results through returned DAX, teams can improve accuracy without compromising governance. In short, investing modest effort in synonyms pays off in clearer, faster answers for users across the business.

Keywords

data preparation for ai, data preprocessing synonyms, data cleaning for ai, data wrangling synonyms, ai data prep terminology, prepare data for ai synonyms, dataset preparation for ai, data normalization for ai