Skip to main content
Back to Library
Prompt Engineering Guide

Mastering Meeting notes extraction
on Perplexity Online 70B

Stop guessing. See how professional prompt engineering transforms Perplexity Online 70B's output for specific technical tasks.

The "Vibe" Prompt

"Extract the key points, decisions, action items, and attendees from these meeting notes: [MEETING_NOTES_TEXT]"
Low specificity, inconsistent output

Optimized Version

STABLE
You are a highly efficient meeting assistant tasked with generating concise and accurate summaries from raw meeting transcripts. Your output must strictly adhere to the specified JSON schema. Prioritize clarity and completeness. If a field is not explicitly present in the provided notes, output 'N/A' for its value. Do not infer or hallucinate information not directly stated. Before generating the final JSON, follow these steps: 1. **Identify Core Participants**: Scan the text for names or roles mentioned as attending or speaking, and list them. 2. **Summarize Key Discussions**: Read through the notes and distill the main topics and discussions. Group related points thematically. 3. **Extract Decisions**: Look for explicit statements of decisions made, agreements reached, or conclusions drawn. Use keywords like 'decided', 'agreed', 'resolved'. 4. **Identify Action Items**: Search for tasks assigned, next steps, or responsibilities. Note who is responsible and by when, if stated. Use keywords like 'will', 'to do', 'action', 'follow up'. 5. **Synthesize**: Combine the refined information into the final JSON structure. Meeting Notes: """ [MEETING_NOTES_TEXT] """ Output JSON strictly in this format: { "meeting_title": "(Inferred or Stated Title, N/A if not found)", "date": "(Date of meeting, N/A if not found)", "attendees": [ "Attendee Name 1", "Attendee Name 2" ], "key_discussions": [ "Main point of discussion 1", "Main point of discussion 2" ], "decisions_made": [ "Decision 1 (e.g., 'Approved budget for Q3')", "Decision 2 (e.g., 'Website redesign phase 2 to start next month')" ], "action_items": [ { "task": "Action item description", "owner": "Responsible Person", "due_date": "(Due date, N/A if not specified)" } ], "next_meeting_date": "(Date of next scheduled meeting, N/A if not specified)" }
Structured, task-focused, reduced hallucinations

Engineering Rationale

The 'optimized_prompt' works better due to several factors: 1. **Explicit Role Assignment**: It assigns the model a clear role ('highly efficient meeting assistant'), setting expectations for its behavior. 2. **Clear Instructions and Constraints**: It provides detailed instructions on what to extract, the output format (strict JSON), and how to handle missing information ('N/A'). It explicitly forbids inference/hallucination. 3. **Chain-of-Thought (CoT)**: The numbered steps (Identify Participants, Summarize Discussions, Extract Decisions, Identify Action Items, Synthesize) guide the model through a logical processing sequence, mimicking human thought and improving accuracy. 4. **Structured Output Schema**: Providing a detailed JSON schema with example values minimizes ambiguity and ensures consistent, machine-readable output. 5. **Keyword Guidance**: Suggesting keywords helps the model identify specific types of information (e.g., 'decided' for decisions, 'will' for action items). 6. **Temperature and Determinism**: By asking for 'strictly adhere' and 'strictly in this format', it encourages a more deterministic, less creative output from the model, which is crucial for data extraction.

0%
Token Efficiency Gain
The output for 'optimized_prompt' will always be valid JSON.
The 'optimized_prompt' will extract action items with owners and due dates when present, unlike the 'vibe_prompt'.
The 'optimized_prompt' will correctly identify and list attendees as a distinct array.

Ready to stop burning tokens?

Join 5,000+ developers using Prompt Optimizer to slash costs and boost LLM reliability.

Optimize My Prompts