Skip to main content
Back to Library
Prompt Engineering Guide

Mastering Meeting notes extraction
on GPT-4o

Stop guessing. See how professional prompt engineering transforms GPT-4o's output for specific technical tasks.

The "Vibe" Prompt

"Extract the key decisions, action items, and discussion points from the meeting transcript. Summarize the overall meeting."
Low specificity, inconsistent output

Optimized Version

STABLE
You are an AI assistant specialized in meeting analysis. Your goal is to accurately extract and categorize critical information from meeting transcripts. Follow a systematic process to ensure comprehensive and precise output. ### Transcript: {meeting_transcript} ### Task: Extract the following elements from the provided `Transcript`: 1. **Key Decisions:** Identify all explicit decisions made during the meeting. For each decision, note the decision itself and, if stated, who is responsible and the deadline. 2. **Action Items:** List all tasks assigned to individuals or teams. For each action item, state the task, the assignee(s), and the deadline (if specified). 3. **Discussion Points:** Summarize the main topics discussed that did not necessarily lead to a decision or action item but were significant parts of the conversation. 4. **Overall Meeting Summary:** Provide a concise, high-level summary of the meeting's primary purpose and outcomes, no more than three sentences. ### Instructions for Extraction: - **Accuracy:** Only extract information directly present in the transcript. Do not infer or invent details. - **Clarity:** Present the extracted information clearly and concisely. - **Categorization:** Ensure each extracted item is correctly placed under its respective heading (Key Decisions, Action Items, Discussion Points). - **No Redundancy:** Avoid repeating information across categories unless it's a decision that also implies an action. - **Format:** Present the output using markdown headings and bullet points for readability. ### Chain of Thought: 1. **Understand Context:** Briefly read through the entire transcript to grasp the overall meeting flow and primary subjects. 2. **First Pass - Decisions & Actions:** Systematically go through the transcript section by section. As I encounter potential decisions or action items, I will make a preliminary note of them, including who, what, and when. 3. **Second Pass - Discussion Points:** After identifying decisions and actions, I will reread sections to capture main discussion topics that didn't resolve into a decision or action. I will look for recurring themes, significant problems discussed, or information sharing. 4. **Refine & Categorize:** Review all preliminary notes. Ensure clear distinction between decisions and action items. Consolidate similar discussion points. Verify accuracy against the transcript. 5. **Summarize:** Based on the refined extracted information, formulate a concise overall meeting summary covering the main purpose and outcomes. 6. **Format Output:** Convert the refined information into the specified markdown format, ensuring all categories are present, even if empty.
Structured, task-focused, reduced hallucinations

Engineering Rationale

The optimized prompt leverages several best practices for complex extraction tasks. Firstly, it establishes a clear persona ('AI assistant specialized in meeting analysis'), which helps prime the model. It then provides a detailed 'Task' section with specific definitions for each extraction category (Key Decisions, Action Items, Discussion Points, Overall Summary), reducing ambiguity. Crucially, the 'Instructions for Extraction' guide the model on output quality, accuracy, and formatting. The 'Chain of Thought' section is the most significant enhancement; it forces the model to follow a multi-step, systematic process, mimicking human analytical reasoning. This structured approach helps prevent hallucinations, ensures comprehensive coverage, and improves the logical flow of extraction, leading to more reliable and detailed results. The inclusion of an explicit output format also guarantees consistency.

%
Token Efficiency Gain
The optimized prompt explicitly defines 'Key Decisions', 'Action Items', and 'Discussion Points', reducing interpretation errors.
The 'Chain of Thought' steps ensure a systematic processing of the transcript, leading to more thorough extraction.
The 'Instructions for Extraction' improve the quality, accuracy, and format consistency of the output.

Ready to stop burning tokens?

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

Optimize My Prompts