How to assess Dragon Copilot issues

  • 6 minutes

As a trainer, you're often the first person users turn to when something isn't working as expected in the application. By using a clear, structured assessment, you can diagnose problems quickly and coach learners with confidence. This unit provides a step-by-step approach you can use across common categories of problems.

Note

This unit applies to Dragon Copilot and DAX Copilot. The features and functionality available to you vary based on your organization's configuration, language settings, and whether you're using the web, mobile, desktop, or embedded app.

How to analyze Dragon Copilot (physicians) problems

Step 1: Gather information

Start by understanding the situation thoroughly. Ask open-ended, learner-centric questions and tailor your follow-ups based on what you hear. Examples:

  • Is the problem affecting one session or multiple sessions?
  • Has the clinician used Dragon Copilot successfully before?
  • Is the problem primarily technical, workflow-related, output-related, or something else?
  • Did anything change in their setup? For example, a new location, device, network, or electronic health record (EHR) workflow?
  • What output did the clinician expect, and what did they receive?
  • Did any settings recently change, such as extensions, style or format settings, or microphone source?
  • What was the input ambient recording vs. dictation?
  • Did the clinician follow verbalization best practices?
  • What mitigation steps has the clinician already tried?
  • Does the problem affect other clinicians?

When you begin with questions, you prevent assumptions and surface the most likely paths to resolution.

Step 2: Attempt to reproduce the problem

Ask the clinician to demonstrate the exact steps they took when the error or problem occurred. If you can't reproduce the actions, the problem might be a singular occurrence or already resolved. If you or the clinician can reproduce the actions, you gain precise insight into triggers and sequence, which accelerates troubleshooting.

Step 3: Map to a root cause category

Use the evidence you gathered to isolate variables and narrow the likely cause. Typical categories include:

  • Expectation versus product behavior: Misaligned assumptions about how the product should work or a misunderstanding of Dragon Copilot feature scope.

  • Environment or connection: A network or location problem. For example, is the internet spotty in an area?

  • Device or input source: Device settings such as do not disturb on the mobile device or microphone selection.

  • Electronic health record (EHR) workflow: Where did the clinician edit the note? Were they in an unsupported text field?

  • Configuration or settings: What extensions are enabled? What style and formatting settings are selected? Was a specialty or role selected? Were default templates applied?

  • Clinician workflow or verbalization: Does the output lack clarity? Is it missing context? Are there contradictions to the transcript?

  • Cursor placement or template usage: Did the text land at end of a note or in the incorrect section?

  • Prompt quality: Is a key component of the prompt missing? Does the prompt contain a goal, context, source, and expectations? Does it contain vague instructions?

  • Application or technical problem: Are there unexpected errors that require escalation?

When in doubt, isolate one variable at a time and know when to escalate.

Things to consider

  • Cursor placement: Confirm the cursor is in the intended location. Remember, the text goes where the cursor is blinking.

  • Spoken forms and keywords: Verify the spoken form of a command and advise the clinician of any default keywords, such as insert.

  • Workflow steps: Ensure each workflow step mirrors a keystroke sequence that works on the physical keyboard.

  • Prompt clarity: Check for the four elements (goal, context, source, expectations) and simplify to a single focus request.

  • Revisions needed: If you don't achieve the desired output, update vocabulary, prompts, customizable templates, or workflows in Dragon Copilot’s library.

  • Extensions and sources: Confirm that Information Assist is enabled and at least one trusted source or extension is selected.

These small checks often yield significant improvements in output reliability and learner confidence.

Key takeaway

The most reliable path to resolution is a structured assessment: gather clear details, reproduce the problem, and map evidence to the likely root cause. This approach helps you coach learners more effectively and build long-term self-sufficiency across your organization.