Automate Meeting Summaries from Transcripts Using Python in 15 Minutes

You're drowning in meeting notes. Every day brings 3-4 meetings, each generating a transcript you need to read, summarize, and extract action items from. It takes 30 minutes per meeting just to create a useful summary.

That's 2.5 hours daily spent on post-meeting admin work.

Here's the problem: Meeting transcripts are verbose, unstructured, and filled with tangents. Finding the key decisions and action items means reading every word. Most people skip this step entirely, leading to missed commitments and confused teams.

But what if you could automate the entire process? Feed in a transcript, get back a clean summary with categorized action items in seconds.

I'll show you how to build this automation using Python and the OpenAI API. You'll have a working script in 15 minutes that can process any meeting transcript, regardless of format.

Why Automate Meeting Summary Generation

Before diving into code, understand the ROI:

Manual Process:

Read 30-page transcript: 15 minutes
Identify key points: 10 minutes
Extract action items: 5 minutes
Write summary: 10 minutes
Total: 40 minutes per meeting

Automated Process:

Run Python script: 30 seconds
Review AI summary: 3 minutes
Make minor edits: 2 minutes
Total: 5.5 minutes per meeting

For someone with 15 meetings per week, that's 8.5 hours saved - over a full workday.

Prerequisites and Setup

You'll need:

Python 3.8 or higher
OpenAI API key (get one at platform.openai.com)
Basic command line familiarity

Install Required Packages

bash

1pip install openai python-dotenv

Create a .env file for your API key:

bash

1OPENAI_API_KEY=your_api_key_here

Never commit API keys to version control. Use environment variables or .env files with .gitignore.

The Meeting Summary Automation Script

Here's the complete script that processes meeting transcripts:

python

1import os
2from openai import OpenAI
3from dotenv import load_dotenv
4from datetime import datetime
5
6# Load environment variables
7load_dotenv()
8
9# Initialize OpenAI client
10client = OpenAI(api_key=os.getenv('OPENAI_API_KEY'))
11
12def generate_meeting_summary(transcript_text, meeting_date=None):
13    """
14    Generate a structured meeting summary from transcript text.
15    
16    Args:
17        transcript_text (str): The full meeting transcript
18        meeting_date (str): Optional date of the meeting
19    
20    Returns:
21        str: Formatted meeting summary with action items
22    """
23    
24    # Construct the prompt for GPT-4
25    system_prompt = """You are an expert meeting summarizer. Your job is to:
26    1. Create a concise executive summary (2-3 sentences)
27    2. List key discussion points with brief explanations
28    3. Extract all action items with assigned owners and deadlines
29    4. Note important decisions made
30    5. Flag any unresolved questions or blockers
31    
32    Format your response as structured markdown with clear sections."""
33    
34    user_prompt = f"""Please analyze this meeting transcript and create a structured summary:
35
36TRANSCRIPT:
37{transcript_text}
38
39{"MEETING DATE: " + meeting_date if meeting_date else ""}
40
41Create a comprehensive but concise summary following best practices for meeting documentation."""
42
43    try:
44        # Call OpenAI API
45        response = client.chat.completions.create(
46            model="gpt-4-turbo-preview",
47            messages=[
48                {"role": "system", "content": system_prompt},
49                {"role": "user", "content": user_prompt}
50            ],
51            temperature=0.3,  # Lower temperature for more factual summaries
52            max_tokens=1500
53        )
54        
55        summary = response.choices[0].message.content
56        return summary
57        
58    except Exception as e:
59        return f"Error generating summary: {str(e)}"
60
61def process_transcript_file(file_path):
62    """
63    Read transcript from file and generate summary.
64    
65    Args:
66        file_path (str): Path to transcript file
67    
68    Returns:
69        str: Generated summary
70    """
71    try:
72        with open(file_path, 'r', encoding='utf-8') as f:
73            transcript = f.read()
74        
75        print(f"Processing transcript from: {file_path}")
76        print(f"Transcript length: {len(transcript)} characters\n")
77        
78        # Extract date from filename if possible
79        meeting_date = None
80        if '_' in file_path:
81            try:
82                date_part = file_path.split('_')[1].split('.')[0]
83                meeting_date = date_part
84            except:
85                pass
86        
87        summary = generate_meeting_summary(transcript, meeting_date)
88        return summary
89        
90    except FileNotFoundError:
91        return f"Error: File not found at {file_path}"
92    except Exception as e:
93        return f"Error processing file: {str(e)}"
94
95def save_summary(summary, output_path):
96    """
97    Save summary to a file.
98    
99    Args:
100        summary (str): The generated summary
101        output_path (str): Where to save the summary
102    """
103    try:
104        with open(output_path, 'w', encoding='utf-8') as f:
105            f.write(f"# Meeting Summary - {datetime.now().strftime('%Y-%m-%d')}\n\n")
106            f.write(summary)
107        print(f"\n✅ Summary saved to: {output_path}")
108    except Exception as e:
109        print(f"\n❌ Error saving summary: {str(e)}")
110
111def main():
112    """
113    Main function to run the meeting summary automation.
114    """
115    print("=== Meeting Summary Generator ===\n")
116    
117    # Get transcript file path from user
118    transcript_file = input("Enter the path to your transcript file: ").strip()
119    
120    if not transcript_file:
121        print("❌ No file path provided")
122        return
123    
124    # Process the transcript
125    summary = process_transcript_file(transcript_file)
126    
127    # Display the summary
128    print("\n" + "="*60)
129    print("GENERATED SUMMARY")
130    print("="*60 + "\n")
131    print(summary)
132    print("\n" + "="*60)
133    
134    # Ask if user wants to save
135    save_choice = input("\nSave this summary to a file? (y/n): ").strip().lower()
136    
137    if save_choice == 'y':
138        output_file = input("Enter output filename (default: meeting_summary.md): ").strip()
139        if not output_file:
140            output_file = "meeting_summary.md"
141        
142        save_summary(summary, output_file)
143
144if __name__ == "__main__":
145    main()

How the Script Works

Step 1: Prompt Engineering for Quality Summaries

The script uses a carefully crafted system prompt that instructs GPT-4 to:

Create executive summaries: 2-3 sentence overviews
Extract key points: Main discussion topics with context
Identify action items: With owners and deadlines
Document decisions: Important conclusions reached
Flag blockers: Unresolved issues needing attention

The temperature=0.3 setting ensures consistent, factual summaries rather than creative interpretations.

Step 2: Processing Different Transcript Formats

The script handles various transcript formats:

python

1# Works with plain text transcripts
2meeting_transcript_20260128.txt
3
4# Works with timestamped transcripts
5[00:01:23] John: We need to prioritize the Q2 roadmap.
6[00:01:45] Sarah: I agree. Let's focus on customer feedback.
7
8# Works with speaker-labeled transcripts
9John: We need to prioritize the Q2 roadmap.
10Sarah: I agree. Let's focus on customer feedback.

GPT-4 automatically identifies speakers, timestamps, and structure regardless of format.

Step 3: Structured Output Generation

The AI generates markdown-formatted summaries following this template:

markdown

1## Executive Summary
2[2-3 sentence overview]
3
4## Key Discussion Points
5- **Topic 1**: Brief explanation
6- **Topic 2**: Brief explanation
7
8## Action Items
9- [ ] **Task description** - Owner: [Name] - Due: [Date]
10- [ ] **Task description** - Owner: [Name] - Due: [Date]
11
12## Decisions Made
13- Decision 1 with context
14- Decision 2 with context
15
16## Unresolved Questions
17- Question or blocker 1
18- Question or blocker 2

This format is consistent, scannable, and ready to share with your team.

Running the Script

Basic Usage

bash

1python meeting_summarizer.py

When prompted, enter your transcript file path:

Enter the path to your transcript file: transcripts/meeting_20260128.txt

The script processes the file and displays the summary:

Prompt

Processing transcript from: transcripts/meeting_20260128.txt
Transcript length: 4523 characters

============================================================
GENERATED SUMMARY
============================================================

## Executive Summary
Team discussed Q2 product roadmap prioritization, focusing on customer feedback from recent surveys. Agreed to deprioritize feature X and accelerate feature Y timeline by 2 weeks.

[... rest of summary ...]

Saving Summaries

After generation, the script asks if you want to save:

Prompt

Save this summary to a file? (y/n): y
Enter output filename (default: meeting_summary.md): q2_planning_summary.md

✅ Summary saved to: q2_planning_summary.md

Advanced Usage: Batch Processing

Process multiple transcripts automatically:

python

1import os
2import glob
3
4def batch_process_transcripts(directory_path):
5    """
6    Process all transcript files in a directory.
7    
8    Args:
9        directory_path (str): Path to directory containing transcripts
10    """
11    # Find all .txt files in directory
12    transcript_files = glob.glob(os.path.join(directory_path, "*.txt"))
13    
14    print(f"Found {len(transcript_files)} transcript files\n")
15    
16    for transcript_file in transcript_files:
17        print(f"Processing: {transcript_file}")
18        
19        # Generate summary
20        summary = process_transcript_file(transcript_file)
21        
22        # Create output filename
23        base_name = os.path.splitext(os.path.basename(transcript_file))[0]
24        output_file = f"{base_name}_summary.md"
25        
26        # Save summary
27        save_summary(summary, output_file)
28        
29        print(f"Completed: {output_file}\n")
30
31# Usage
32batch_process_transcripts("./meeting_transcripts")

This processes an entire folder of transcripts, generating individual summary files for each.

Integrating with Meeting Tools

Connect to Otter.ai

Export transcripts from Otter and process automatically:

python

1import requests
2from datetime import datetime, timedelta
3
4def fetch_recent_otter_transcripts(api_key, days=7):
5    """
6    Fetch recent transcripts from Otter.ai API.
7    """
8    url = "https://otter.ai/forward/api/v1/speeches"
9    headers = {"Authorization": f"Bearer {api_key}"}
10    
11    # Calculate date range
12    end_date = datetime.now()
13    start_date = end_date - timedelta(days=days)
14    
15    params = {
16        "start_date": start_date.isoformat(),
17        "end_date": end_date.isoformat()
18    }
19    
20    response = requests.get(url, headers=headers, params=params)
21    return response.json()
22
23# Fetch and process
24transcripts = fetch_recent_otter_transcripts(os.getenv('OTTER_API_KEY'))
25for transcript in transcripts['speeches']:
26    summary = generate_meeting_summary(transcript['transcript'])
27    save_summary(summary, f"otter_summary_{transcript['id']}.md")

Connect to Microsoft Teams

Process Teams meeting recordings:

python

1from azure.identity import DefaultAzureCredential
2from microsoft.graph import GraphServiceClient
3
4async def get_teams_transcripts(meeting_id):
5    """
6    Fetch transcript from Microsoft Teams using Graph API.
7    """
8    credential = DefaultAzureCredential()
9    client = GraphServiceClient(credential)
10    
11    # Get meeting transcript
12    transcript = await client.users.by_user_id('me').onlineMeetings.by_online_meeting_id(meeting_id).transcripts.get()
13    
14    return transcript

Cost Analysis

Understanding API costs helps justify the automation:

OpenAI API Costs (GPT-4 Turbo)

Input: $0.01 per 1K tokens
Output: $0.03 per 1K tokens

Typical meeting transcript:

30-minute meeting: ~5,000 words = ~6,500 tokens input
Summary output: ~500 words = ~650 tokens output

Cost per summary: $0.06 + $0.02 = $0.08

For 15 meetings per week:

Weekly cost: $1.20
Monthly cost: $4.80
Annual cost: $62.40

Compare this to the 442 hours saved annually (8.5 hours/week × 52 weeks). At even $50/hour, you save $22,100 in time value.

ROI: 35,400%

Best Practices and Pro Tips

Optimize Prompt Quality

Your prompt quality directly affects summary accuracy:

Poor prompt:

python

1"Summarize this meeting."

Good prompt:

python

1"""Analyze this meeting transcript and create a structured summary including:
21. Executive summary (2-3 sentences)
32. Key discussion points with context
43. Action items with owners and deadlines
54. Decisions made
65. Unresolved questions or blockers
7
8Format as markdown with clear sections."""

The detailed prompt gives GPT-4 clear expectations, producing better results.

Handle Long Transcripts

GPT-4 Turbo has a 128K token context window, but very long meetings might exceed this:

python

1def chunk_long_transcript(transcript, max_tokens=100000):
2    """
3    Split long transcripts into processable chunks.
4    """
5    # Rough estimate: 1 token ≈ 4 characters
6    max_chars = max_tokens * 4
7    
8    if len(transcript) <= max_chars:
9        return [transcript]
10    
11    # Split into chunks at paragraph boundaries
12    paragraphs = transcript.split('\n\n')
13    chunks = []
14    current_chunk = ""
15    
16    for para in paragraphs:
17        if len(current_chunk) + len(para) < max_chars:
18            current_chunk += para + "\n\n"
19        else:
20            chunks.append(current_chunk)
21            current_chunk = para + "\n\n"
22    
23    if current_chunk:
24        chunks.append(current_chunk)
25    
26    return chunks
27
28# Process each chunk and combine summaries
29chunks = chunk_long_transcript(long_transcript)
30summaries = [generate_meeting_summary(chunk) for chunk in chunks]
31final_summary = combine_summaries(summaries)

Add Custom Fields

Customize output for your team's needs:

python

1system_prompt = """You are an expert meeting summarizer. Create a summary with:
21. Executive summary (2-3 sentences)
32. Attendees (if mentioned)
43. Key discussion points
54. Action items (format: - [ ] Task - Owner: Name - Due: Date)
65. Decisions made
76. Budget implications (if any)
87. Next meeting date/time (if mentioned)
98. Unresolved questions
10
11Use markdown formatting with emoji indicators:
12📋 for action items
13✅ for decisions
14❓ for questions
15💰 for budget items"""

This creates scannable summaries with visual indicators.

Troubleshooting Common Issues

Issue: "Rate limit exceeded"

Cause: Too many API requests too quickly

Solution: Add delays between requests

python

1import time
2
3for transcript in transcript_files:
4    summary = generate_meeting_summary(transcript)
5    save_summary(summary, f"summary_{i}.md")
6    time.sleep(2)  # 2-second delay between requests

Issue: Incomplete action items

Cause: Vague language in transcript

Solution: Add follow-up prompt to extract details

python

1# First, generate basic summary
2summary = generate_meeting_summary(transcript)
3
4# Then, extract action items specifically
5action_items_prompt = f"""Review this meeting summary and extract ALL action items, including:
6- Implied tasks (things someone mentioned they "should" or "will" do)
7- Tasks without explicit owners (assign to most relevant person mentioned)
8- Tasks without deadlines (infer reasonable deadline from context)
9
10Summary:
11{summary}
12
13Format each action item as:
14- [ ] [Task description] - Owner: [Name] - Due: [Date or "TBD"]"""
15
16detailed_actions = client.chat.completions.create(
17    model="gpt-4-turbo-preview",
18    messages=[{"role": "user", "content": action_items_prompt}],
19    temperature=0.2
20)

Issue: Summary too generic

Cause: Not enough context in prompt

Solution: Provide meeting metadata

python

1user_prompt = f"""Analyze this meeting transcript.
2
3MEETING CONTEXT:
4- Meeting Type: {meeting_type}  # e.g., "Sprint Planning", "Client Review"
5- Project: {project_name}
6- Team: {team_name}
7- Date: {meeting_date}
8
9TRANSCRIPT:
10{transcript_text}
11
12Create a detailed summary appropriate for a {meeting_type} meeting."""

Context helps GPT-4 understand what information matters most.

Conclusion

You now have a production-ready Python script that automates meeting summary generation. This single automation saves 8+ hours per week for meeting-heavy roles.

Key takeaways:

Simple setup: 15 minutes to deploy, works immediately
Massive ROI: $0.08 per summary vs 40 minutes manual work
Flexible: Handles any transcript format, integrates with meeting tools
Customizable: Adjust prompts for your team's specific needs
Scalable: Process one meeting or batch-process hundreds

The next step: Integrate this into your meeting workflow. Export transcripts from your meeting tool (Otter, Teams, Zoom) and run them through this script automatically.

Frequently Asked Questions

Can this work with transcripts in other languages?

Yes. GPT-4 supports 50+ languages. The script works identically with transcripts in Spanish, French, German, Chinese, Japanese, and many others. The summary will be generated in the same language as the input transcript.

How accurate are the AI-generated summaries?

GPT-4 achieves 90-95% accuracy on factual extraction from transcripts. Always review the summary before sharing, especially for critical meetings involving legal, financial, or strategic decisions. The 3-minute review step catches any misinterpretations.

What if my transcripts don't have speaker labels?

The script still works. GPT-4 can infer who said what from context clues, though speaker-labeled transcripts produce slightly better results. For unlabeled transcripts, add a note about attendees in the prompt.

Can I use a cheaper model like GPT-3.5?

Yes, but summary quality decreases noticeably. GPT-3.5 often misses nuanced action items and produces less structured output. For critical meetings, stick with GPT-4. For informal team syncs, GPT-3.5 at $0.001/1K tokens input works adequately.

How do I handle confidential meetings?

Either: (1) Use a local AI model like Llama 2 instead of OpenAI's API, or (2) Use OpenAI's API with the "zero data retention" option available on Enterprise plans. Never send confidential transcripts to standard API endpoints that may use data for training.

Automate Meeting Summaries from Transcripts Using Python in 15 Minutes

That's 2.5 hours daily spent on post-meeting admin work.

But what if you could automate the entire process? Feed in a transcript, get back a clean summary with categorized action items in seconds.

I'll show you how to build this automation using Python and the OpenAI API. You'll have a working script in 15 minutes that can process any meeting transcript, regardless of format.

Why Automate Meeting Summary Generation

Before diving into code, understand the ROI:

Manual Process:

Read 30-page transcript: 15 minutes
Identify key points: 10 minutes
Extract action items: 5 minutes
Write summary: 10 minutes
Total: 40 minutes per meeting

Automated Process:

Run Python script: 30 seconds
Review AI summary: 3 minutes
Make minor edits: 2 minutes
Total: 5.5 minutes per meeting

For someone with 15 meetings per week, that's 8.5 hours saved - over a full workday.

Prerequisites and Setup

You'll need:

Python 3.8 or higher
OpenAI API key (get one at platform.openai.com)
Basic command line familiarity

Install Required Packages

bash

1pip install openai python-dotenv

Create a .env file for your API key:

bash

1OPENAI_API_KEY=your_api_key_here

Never commit API keys to version control. Use environment variables or .env files with .gitignore.

The Meeting Summary Automation Script

Here's the complete script that processes meeting transcripts:

python

1import os
2from openai import OpenAI
3from dotenv import load_dotenv
4from datetime import datetime
5
6# Load environment variables
7load_dotenv()
8
9# Initialize OpenAI client
10client = OpenAI(api_key=os.getenv('OPENAI_API_KEY'))
11
12def generate_meeting_summary(transcript_text, meeting_date=None):
13    """
14    Generate a structured meeting summary from transcript text.
15    
16    Args:
17        transcript_text (str): The full meeting transcript
18        meeting_date (str): Optional date of the meeting
19    
20    Returns:
21        str: Formatted meeting summary with action items
22    """
23    
24    # Construct the prompt for GPT-4
25    system_prompt = """You are an expert meeting summarizer. Your job is to:
26    1. Create a concise executive summary (2-3 sentences)
27    2. List key discussion points with brief explanations
28    3. Extract all action items with assigned owners and deadlines
29    4. Note important decisions made
30    5. Flag any unresolved questions or blockers
31    
32    Format your response as structured markdown with clear sections."""
33    
34    user_prompt = f"""Please analyze this meeting transcript and create a structured summary:
35
36TRANSCRIPT:
37{transcript_text}
38
39{"MEETING DATE: " + meeting_date if meeting_date else ""}
40
41Create a comprehensive but concise summary following best practices for meeting documentation."""
42
43    try:
44        # Call OpenAI API
45        response = client.chat.completions.create(
46            model="gpt-4-turbo-preview",
47            messages=[
48                {"role": "system", "content": system_prompt},
49                {"role": "user", "content": user_prompt}
50            ],
51            temperature=0.3,  # Lower temperature for more factual summaries
52            max_tokens=1500
53        )
54        
55        summary = response.choices[0].message.content
56        return summary
57        
58    except Exception as e:
59        return f"Error generating summary: {str(e)}"
60
61def process_transcript_file(file_path):
62    """
63    Read transcript from file and generate summary.
64    
65    Args:
66        file_path (str): Path to transcript file
67    
68    Returns:
69        str: Generated summary
70    """
71    try:
72        with open(file_path, 'r', encoding='utf-8') as f:
73            transcript = f.read()
74        
75        print(f"Processing transcript from: {file_path}")
76        print(f"Transcript length: {len(transcript)} characters\n")
77        
78        # Extract date from filename if possible
79        meeting_date = None
80        if '_' in file_path:
81            try:
82                date_part = file_path.split('_')[1].split('.')[0]
83                meeting_date = date_part
84            except:
85                pass
86        
87        summary = generate_meeting_summary(transcript, meeting_date)
88        return summary
89        
90    except FileNotFoundError:
91        return f"Error: File not found at {file_path}"
92    except Exception as e:
93        return f"Error processing file: {str(e)}"
94
95def save_summary(summary, output_path):
96    """
97    Save summary to a file.
98    
99    Args:
100        summary (str): The generated summary
101        output_path (str): Where to save the summary
102    """
103    try:
104        with open(output_path, 'w', encoding='utf-8') as f:
105            f.write(f"# Meeting Summary - {datetime.now().strftime('%Y-%m-%d')}\n\n")
106            f.write(summary)
107        print(f"\n✅ Summary saved to: {output_path}")
108    except Exception as e:
109        print(f"\n❌ Error saving summary: {str(e)}")
110
111def main():
112    """
113    Main function to run the meeting summary automation.
114    """
115    print("=== Meeting Summary Generator ===\n")
116    
117    # Get transcript file path from user
118    transcript_file = input("Enter the path to your transcript file: ").strip()
119    
120    if not transcript_file:
121        print("❌ No file path provided")
122        return
123    
124    # Process the transcript
125    summary = process_transcript_file(transcript_file)
126    
127    # Display the summary
128    print("\n" + "="*60)
129    print("GENERATED SUMMARY")
130    print("="*60 + "\n")
131    print(summary)
132    print("\n" + "="*60)
133    
134    # Ask if user wants to save
135    save_choice = input("\nSave this summary to a file? (y/n): ").strip().lower()
136    
137    if save_choice == 'y':
138        output_file = input("Enter output filename (default: meeting_summary.md): ").strip()
139        if not output_file:
140            output_file = "meeting_summary.md"
141        
142        save_summary(summary, output_file)
143
144if __name__ == "__main__":
145    main()

How the Script Works

Step 1: Prompt Engineering for Quality Summaries

The script uses a carefully crafted system prompt that instructs GPT-4 to:

Create executive summaries: 2-3 sentence overviews
Extract key points: Main discussion topics with context
Identify action items: With owners and deadlines
Document decisions: Important conclusions reached
Flag blockers: Unresolved issues needing attention

The temperature=0.3 setting ensures consistent, factual summaries rather than creative interpretations.

Step 2: Processing Different Transcript Formats

The script handles various transcript formats:

python

1# Works with plain text transcripts
2meeting_transcript_20260128.txt
3
4# Works with timestamped transcripts
5[00:01:23] John: We need to prioritize the Q2 roadmap.
6[00:01:45] Sarah: I agree. Let's focus on customer feedback.
7
8# Works with speaker-labeled transcripts
9John: We need to prioritize the Q2 roadmap.
10Sarah: I agree. Let's focus on customer feedback.

GPT-4 automatically identifies speakers, timestamps, and structure regardless of format.

Step 3: Structured Output Generation

The AI generates markdown-formatted summaries following this template:

markdown

1## Executive Summary
2[2-3 sentence overview]
3
4## Key Discussion Points
5- **Topic 1**: Brief explanation
6- **Topic 2**: Brief explanation
7
8## Action Items
9- [ ] **Task description** - Owner: [Name] - Due: [Date]
10- [ ] **Task description** - Owner: [Name] - Due: [Date]
11
12## Decisions Made
13- Decision 1 with context
14- Decision 2 with context
15
16## Unresolved Questions
17- Question or blocker 1
18- Question or blocker 2

This format is consistent, scannable, and ready to share with your team.

Running the Script

Basic Usage

bash

1python meeting_summarizer.py

When prompted, enter your transcript file path:

Enter the path to your transcript file: transcripts/meeting_20260128.txt

The script processes the file and displays the summary:

Prompt

Processing transcript from: transcripts/meeting_20260128.txt
Transcript length: 4523 characters

============================================================
GENERATED SUMMARY
============================================================

## Executive Summary
Team discussed Q2 product roadmap prioritization, focusing on customer feedback from recent surveys. Agreed to deprioritize feature X and accelerate feature Y timeline by 2 weeks.

[... rest of summary ...]

Saving Summaries

After generation, the script asks if you want to save:

Prompt

Save this summary to a file? (y/n): y
Enter output filename (default: meeting_summary.md): q2_planning_summary.md

✅ Summary saved to: q2_planning_summary.md

Advanced Usage: Batch Processing

Process multiple transcripts automatically:

python

1import os
2import glob
3
4def batch_process_transcripts(directory_path):
5    """
6    Process all transcript files in a directory.
7    
8    Args:
9        directory_path (str): Path to directory containing transcripts
10    """
11    # Find all .txt files in directory
12    transcript_files = glob.glob(os.path.join(directory_path, "*.txt"))
13    
14    print(f"Found {len(transcript_files)} transcript files\n")
15    
16    for transcript_file in transcript_files:
17        print(f"Processing: {transcript_file}")
18        
19        # Generate summary
20        summary = process_transcript_file(transcript_file)
21        
22        # Create output filename
23        base_name = os.path.splitext(os.path.basename(transcript_file))[0]
24        output_file = f"{base_name}_summary.md"
25        
26        # Save summary
27        save_summary(summary, output_file)
28        
29        print(f"Completed: {output_file}\n")
30
31# Usage
32batch_process_transcripts("./meeting_transcripts")

This processes an entire folder of transcripts, generating individual summary files for each.

Integrating with Meeting Tools

Connect to Otter.ai

Export transcripts from Otter and process automatically:

python

1import requests
2from datetime import datetime, timedelta
3
4def fetch_recent_otter_transcripts(api_key, days=7):
5    """
6    Fetch recent transcripts from Otter.ai API.
7    """
8    url = "https://otter.ai/forward/api/v1/speeches"
9    headers = {"Authorization": f"Bearer {api_key}"}
10    
11    # Calculate date range
12    end_date = datetime.now()
13    start_date = end_date - timedelta(days=days)
14    
15    params = {
16        "start_date": start_date.isoformat(),
17        "end_date": end_date.isoformat()
18    }
19    
20    response = requests.get(url, headers=headers, params=params)
21    return response.json()
22
23# Fetch and process
24transcripts = fetch_recent_otter_transcripts(os.getenv('OTTER_API_KEY'))
25for transcript in transcripts['speeches']:
26    summary = generate_meeting_summary(transcript['transcript'])
27    save_summary(summary, f"otter_summary_{transcript['id']}.md")

Connect to Microsoft Teams

Process Teams meeting recordings:

python

1from azure.identity import DefaultAzureCredential
2from microsoft.graph import GraphServiceClient
3
4async def get_teams_transcripts(meeting_id):
5    """
6    Fetch transcript from Microsoft Teams using Graph API.
7    """
8    credential = DefaultAzureCredential()
9    client = GraphServiceClient(credential)
10    
11    # Get meeting transcript
12    transcript = await client.users.by_user_id('me').onlineMeetings.by_online_meeting_id(meeting_id).transcripts.get()
13    
14    return transcript

Cost Analysis

Understanding API costs helps justify the automation:

OpenAI API Costs (GPT-4 Turbo)

Input: $0.01 per 1K tokens
Output: $0.03 per 1K tokens

Typical meeting transcript:

30-minute meeting: ~5,000 words = ~6,500 tokens input
Summary output: ~500 words = ~650 tokens output

Cost per summary: $0.06 + $0.02 = $0.08

For 15 meetings per week:

Weekly cost: $1.20
Monthly cost: $4.80
Annual cost: $62.40

Compare this to the 442 hours saved annually (8.5 hours/week × 52 weeks). At even $50/hour, you save $22,100 in time value.

ROI: 35,400%

Best Practices and Pro Tips

Optimize Prompt Quality

Your prompt quality directly affects summary accuracy:

Poor prompt:

python

1"Summarize this meeting."

Good prompt:

python

1"""Analyze this meeting transcript and create a structured summary including:
21. Executive summary (2-3 sentences)
32. Key discussion points with context
43. Action items with owners and deadlines
54. Decisions made
65. Unresolved questions or blockers
7
8Format as markdown with clear sections."""

The detailed prompt gives GPT-4 clear expectations, producing better results.

Handle Long Transcripts

GPT-4 Turbo has a 128K token context window, but very long meetings might exceed this:

python

1def chunk_long_transcript(transcript, max_tokens=100000):
2    """
3    Split long transcripts into processable chunks.
4    """
5    # Rough estimate: 1 token ≈ 4 characters
6    max_chars = max_tokens * 4
7    
8    if len(transcript) <= max_chars:
9        return [transcript]
10    
11    # Split into chunks at paragraph boundaries
12    paragraphs = transcript.split('\n\n')
13    chunks = []
14    current_chunk = ""
15    
16    for para in paragraphs:
17        if len(current_chunk) + len(para) < max_chars:
18            current_chunk += para + "\n\n"
19        else:
20            chunks.append(current_chunk)
21            current_chunk = para + "\n\n"
22    
23    if current_chunk:
24        chunks.append(current_chunk)
25    
26    return chunks
27
28# Process each chunk and combine summaries
29chunks = chunk_long_transcript(long_transcript)
30summaries = [generate_meeting_summary(chunk) for chunk in chunks]
31final_summary = combine_summaries(summaries)

Add Custom Fields

Customize output for your team's needs:

python

1system_prompt = """You are an expert meeting summarizer. Create a summary with:
21. Executive summary (2-3 sentences)
32. Attendees (if mentioned)
43. Key discussion points
54. Action items (format: - [ ] Task - Owner: Name - Due: Date)
65. Decisions made
76. Budget implications (if any)
87. Next meeting date/time (if mentioned)
98. Unresolved questions
10
11Use markdown formatting with emoji indicators:
12📋 for action items
13✅ for decisions
14❓ for questions
15💰 for budget items"""

This creates scannable summaries with visual indicators.

Troubleshooting Common Issues

Issue: "Rate limit exceeded"

Cause: Too many API requests too quickly

Solution: Add delays between requests

python

1import time
2
3for transcript in transcript_files:
4    summary = generate_meeting_summary(transcript)
5    save_summary(summary, f"summary_{i}.md")
6    time.sleep(2)  # 2-second delay between requests

Issue: Incomplete action items

Cause: Vague language in transcript

Solution: Add follow-up prompt to extract details

python

1# First, generate basic summary
2summary = generate_meeting_summary(transcript)
3
4# Then, extract action items specifically
5action_items_prompt = f"""Review this meeting summary and extract ALL action items, including:
6- Implied tasks (things someone mentioned they "should" or "will" do)
7- Tasks without explicit owners (assign to most relevant person mentioned)
8- Tasks without deadlines (infer reasonable deadline from context)
9
10Summary:
11{summary}
12
13Format each action item as:
14- [ ] [Task description] - Owner: [Name] - Due: [Date or "TBD"]"""
15
16detailed_actions = client.chat.completions.create(
17    model="gpt-4-turbo-preview",
18    messages=[{"role": "user", "content": action_items_prompt}],
19    temperature=0.2
20)

Issue: Summary too generic

Cause: Not enough context in prompt

Solution: Provide meeting metadata

python

1user_prompt = f"""Analyze this meeting transcript.
2
3MEETING CONTEXT:
4- Meeting Type: {meeting_type}  # e.g., "Sprint Planning", "Client Review"
5- Project: {project_name}
6- Team: {team_name}
7- Date: {meeting_date}
8
9TRANSCRIPT:
10{transcript_text}
11
12Create a detailed summary appropriate for a {meeting_type} meeting."""

Context helps GPT-4 understand what information matters most.

Conclusion

You now have a production-ready Python script that automates meeting summary generation. This single automation saves 8+ hours per week for meeting-heavy roles.

Key takeaways:

Simple setup: 15 minutes to deploy, works immediately
Massive ROI: $0.08 per summary vs 40 minutes manual work
Flexible: Handles any transcript format, integrates with meeting tools
Customizable: Adjust prompts for your team's specific needs
Scalable: Process one meeting or batch-process hundreds

The next step: Integrate this into your meeting workflow. Export transcripts from your meeting tool (Otter, Teams, Zoom) and run them through this script automatically.

Frequently Asked Questions

Can this work with transcripts in other languages?

How accurate are the AI-generated summaries?

What if my transcripts don't have speaker labels?

Can I use a cheaper model like GPT-3.5?

How do I handle confidential meetings?

Automate Meeting Summaries from Transcripts Using Python in 15 Minutes

Why Automate Meeting Summary Generation

Prerequisites and Setup

Install Required Packages

The Meeting Summary Automation Script

How the Script Works

Step 1: Prompt Engineering for Quality Summaries

Step 2: Processing Different Transcript Formats

Step 3: Structured Output Generation

Running the Script

Basic Usage

Saving Summaries

Advanced Usage: Batch Processing

Integrating with Meeting Tools

Connect to Otter.ai

Connect to Microsoft Teams

Cost Analysis

OpenAI API Costs (GPT-4 Turbo)

Best Practices and Pro Tips

Optimize Prompt Quality

Handle Long Transcripts

Add Custom Fields

Troubleshooting Common Issues

Issue: "Rate limit exceeded"

Issue: Incomplete action items

Issue: Summary too generic

Conclusion

Frequently Asked Questions

Share this article

Automate Meeting Summaries from Transcripts Using Python in 15 Minutes

Why Automate Meeting Summary Generation

Prerequisites and Setup

Install Required Packages

The Meeting Summary Automation Script

How the Script Works

Step 1: Prompt Engineering for Quality Summaries

Step 2: Processing Different Transcript Formats

Step 3: Structured Output Generation

Running the Script

Basic Usage

Saving Summaries

Advanced Usage: Batch Processing

Integrating with Meeting Tools

Connect to Otter.ai

Connect to Microsoft Teams

Cost Analysis

OpenAI API Costs (GPT-4 Turbo)

Best Practices and Pro Tips

Optimize Prompt Quality

Handle Long Transcripts

Add Custom Fields

Troubleshooting Common Issues

Issue: "Rate limit exceeded"

Issue: Incomplete action items

Issue: Summary too generic

Conclusion

Frequently Asked Questions

Share this article