Ever wondered how to draft professional meeting minutes without relying on MS Co-Pilot? While tools like Microsoft Teams can record meetings and generate transcripts, they often come with limitations. For instance, MS Teams requires an MS Co-Pilot subscription to analyze transcripts and create meeting minutes, and even with that, crafting effective prompts for such tools is essential for generating useful outputs.

Recently, a colleague sent a meeting recording—without a transcript—and asked us to create the minutes. Here’s how we accomplished this task, step by step.

---

Step 1: Transcribing the Meeting Recording

Since AI models cannot directly process audio or video, the first step was to generate a text transcript of the recording. I used Microsoft Word’s Dictate → Transcribe feature, but encountered a roadblock: the recording exceeded the tool’s 300MB file size limit (it was 550MB).

To bypass this, I extracted the audio from the video using VLC Media Player, a versatile media tool:

1. Open VLC and go to Media → Convert/Save.
2. Add the video file.
3. Select Convert/Save and set the output format to .mp3.
4. Start the process, and VLC will generate an audio file.

With the audio file ready, I returned to Microsoft Word. This time, the smaller file successfully transcribed into a 45-page text document of decent quality.

---

Step 2: Crafting a Prompt for Meeting Minutes

Creating effective meeting minutes with an AI model requires a detailed, structured prompt. Think of it as giving precise instructions to a chef—vagueness leads to unsatisfactory results.

I started with a simple XML-style prompt for ChatGPT (GPT-4), using tags to organize key elements:

plaintextCopyEditYou are an expert in creating meeting minutes from a given transcript. Analyze the provided transcript and generate professional meeting minutes with the specified structure.  

<transcript>
{{meeting_transcript.docx}}
</transcript>  

<structure>
- Main Points Discussed  
- Decisions, Resolutions, and Agreements  
- Summary of Differing Opinions (if any)  
- Action Items: Tasks assigned, responsible parties, and deadlines  
- Follow-Ups: Topics to revisit in future meetings  
</structure>  

<instructions>  
- Stick strictly to the transcript content.  
- Do not invent or infer information.  
- Keep the minutes objective, factual, and concise.  
- Ensure clarity and self-containment for future reference.  
</instructions>

This prompt acted as a baseline, providing clarity and structure for the model to extract and summarize relevant details from the transcript.

---

Step 3: Refining the Prompt Using Anthropic’s Workbench

To improve the clarity and effectiveness of the prompt, I used Anthropic’s Workbench, which offers an automatic prompt enhancement tool. The goal was to refine the structure and optimize the instructions.

Here’s the improved version generated by Anthropic:

plaintextCopyEditYou are an expert in creating professional meeting minutes from transcripts. Analyze the provided transcript and organize the information systematically before drafting the minutes.  

<meeting_transcript>
{{meeting_transcript.docx}}
</meeting_transcript>  

<analysis_structure>
1. Main Points Discussed:  
   - Key topics with relevant quotes from the transcript.  
2. Decisions and Agreements:  
   - Summary of resolutions with supporting quotes.  
3. Differing Opinions (if any):  
   - Notable disagreements or alternative viewpoints.  
4. Action Items:  
   - Tasks, responsible parties, and deadlines.  
5. Follow-Up Topics:  
   - Issues or items to revisit in future meetings.  
</analysis_structure>  

<guidelines>  
- Follow the analysis structure before drafting the final minutes.  
- Use clear, concise language and a professional tone.  
- Avoid unnecessary details and stick to transcript content.  
- Ensure the minutes are self-contained and explanatory.  
</guidelines>

This enhanced prompt incorporated a “chain-of-thought” methodology, guiding the model to analyze and organize the information step by step before drafting the final minutes.

---

Exploring Other Tools: OpenAI’s Prompt Improver

I also tested OpenAI’s Prompt Improver in its Chat Playground, which generated a similarly refined prompt:

plaintextCopyEditCreate professional meeting minutes from the provided transcript. Use the following structure and guidelines to ensure accuracy and clarity:  

**Transcript:**  
- File: {{meeting_transcript.docx}}  

**Structure:**  
- Main Points Discussed  
- Decisions and Agreements  
- Differing Opinions (if any)  
- Action Items  
- Follow-Up Topics  

**Instructions:**  
- Maintain objectivity and stick to the transcript content.  
- Use concise yet explanatory language.  
- Adhere strictly to the structure for clarity and reference.  
- Avoid unnecessary embellishments or personal insights.  

**Output Format:**  
- Use bullet points for clarity, with no more than one level of indentation.  
- Ensure the minutes are self-contained and useful for future reference.  

While effective, OpenAI’s output lacked the chain-of-thought methodology and example formatting provided by Anthropic’s tool, which resulted in less structured meeting minutes.

---

Key Takeaways

1. Prompt Design Matters: A detailed, structured prompt ensures that AI models generate accurate and useful outputs. Refining prompts with tools like Anthropic’s Workbench can significantly enhance clarity and results.
2. Adaptability: The approach for crafting meeting minutes can be applied to various knowledge-extraction tasks, including document summarization and insight generation.
3. Privacy Concerns: When using AI tools, ensure sensitive data is handled securely. For ChatGPT, disable data-sharing settings and consider enterprise-grade solutions for confidential content.
4. Cross-Verification: Always compare AI-generated meeting minutes with the original transcript to ensure accuracy and reliability.

By following this approach, you can produce professional meeting minutes efficiently—no MS Co-Pilot subscription required.