Key principle: Clear, specific prompts produce better results. The LLM's output quality depends directly on your prompt clarity and context.
High-Level Insights
The real value lies in extracting patterns, themes, and sentiments across multiple conversations—insights that aren't obvious from individual transcripts.
Ask for Themes or Patterns
Use phrases like "What are the key themes..." or "Identify common patterns..." LLMs analyze context to find implied meanings and emotions, not just exact words.
Example: "What recurring concerns do patients express about managing diabetes?" might reveal themes like side-effect fears or lifestyle change difficulties.
In Transcription Hub, select the Research tab to access the agent research page
You can also minimize the left side navigation panel by selecting the panel card on the bottom left corner. Select the icon with speech bubbles to access the Research page
Type your question into the chat box
Review the response and continue the conversation
Inquire about Sentiment or Tone
Ask how patients or physicians feel about specific topics: "How do patients generally feel about chemotherapy in our transcripts—optimistic, frustrated, or confused?"
Navigate back to the Research page and type your question into the chat box
Review the response and continue the conversation
Request Summaries or Comparisons
Summaries: "Summarize how doctors typically explain treatment X for disease Y"
Response summary for 'Summarize how doctors typically explain treatment options for diabetes'
Deep Research
Comparisons: "Do patients express different concerns about symptoms related to Atrial Fibrillation vs Hypertension?" or "How does side effect discussion differ between cardiology and neurology transcripts?"
Select 'Run Deep Research' to return a comprehensive analysis from the data
Specify whether the data the agent is preparing is correct
Review the response and continue the conversation
General Prompting Best Practices
Be Clear and Specific: Replace vague requests with precise details. Say "summarize patient concerns about medication side effects" instead of "analyze this conversation"
Provide Context: Set the scene—"You are a healthcare researcher analyzing diabetes patient-doctor conversations"—to focus the LLM appropriately
Define Output Format: Specify what you want—bullet points, paragraphs, tables Example: "List the top 3 themes with brief explanations"
Set Focus Points: Guide attention to relevant aspects: "Focus on patient sentiments, skip technical lab details" or "Consider only oncology transcripts from the last 2 years"
Match Your Audience: Specify tone—"explain in simple terms" for general audiences or "use clinical terminology" for experts
Iterate and Refine: Review initial outputs and ask follow-up questions for clarity or additional detail. Iterative prompting often yields better insights