Skip to main content
Back to Library
Prompt Engineering Guide

Mastering Meeting notes extraction
on DeepSeek V3

Stop guessing. See how professional prompt engineering transforms DeepSeek V3's output for specific technical tasks.

The "Vibe" Prompt

"Extract the key points, action items, and decisions from the following meeting transcript. Summarize briefly. Make sure to capture who is responsible for what. Find any upcoming deadlines. Meeting Transcript: [insert transcript here]"
Low specificity, inconsistent output

Optimized Version

STABLE
You are an expert meeting assistant. Your task is to meticulously extract and structure information from the provided meeting transcript. Follow these steps: 1. **Identify Speakers and Core Topics**: Read through the transcript to understand who spoke and the main subjects discussed. 2. **Extract Key Decisions**: For each decision, identify: - **Decision Point**: The core outcome or choice made. - **Decided By**: The person(s) or group who made the decision. - **Impact**: Brief note on what the decision affects. 3. **Extract Action Items**: For each action item, identify: - **Action**: The specific task to be performed. - **Owner**: The person(s) responsible for completing the task. - **Due Date (if specified)**: The deadline for completion. - **Status (if specified)**: Any current status (e.g., 'pending', 'in progress'). 4. **Summarize Discussion Points**: Condense the main discussion points that led to decisions or informed action items. Do not repeat decisions or action items here. 5. **Identify Open Questions/Next Steps (excluding decided actions)**: Note any topics that require further discussion or unresolved questions. Structure your output as a JSON object with the following keys: `meeting_summary`: A concise, high-level overview of the meeting. `decisions`: An array of objects, each with `decision_point`, `decided_by`, `impact`. `action_items`: An array of objects, each with `action`, `owner`, `due_date` (nullable string), `status` (nullable string). `discussion_points`: An array of strings summarizing key discussions. `open_questions_next_steps`: An array of strings for unresolved items. If a category is empty, provide an empty array. Meeting Transcript: [insert transcript here]
Structured, task-focused, reduced hallucinations

Engineering Rationale

The optimized prompt uses a chain-of-thought approach by breaking down the extraction process into sequential, logical steps. It explicitly defines the types of information to be extracted for each category (decisions, action items) and provides a clear, structured JSON output format. This reduces ambiguity, guides the model to perform specific sub-tasks, and ensures consistency in the output. By specifying nullable fields and empty arrays for missing categories, it handles edge cases gracefully. The 'vibe_prompt' is too general and relies heavily on the model's interpretation, leading to varied and potentially incomplete results.

5%
Token Efficiency Gain
The optimized prompt consistently returns valid JSON.
The optimized prompt accurately extracts specific owners and due dates for action items, where the naive prompt might miss them or present them inconsistently.
The optimized prompt clearly separates decisions from general discussion points, unlike the naive prompt which might conflate them.

Ready to stop burning tokens?

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

Optimize My Prompts