n8n AI Meeting Summary: Extract Decisions & Action Items to Notion, Slack, and Gmail

Webhook meeting transcript → GPT-4 structured analysis → multi-channel sync of summaries, tasks, and reminders

Who

Product teams, remote companies, and operations teams needing automated meeting follow-up

You get

Within 8-12 seconds after a meeting ends, generate structured summaries, decision lists, and prioritized action items with owners, then sync to Slack, Notion, Gmail, Google Calendar, and Sheets

You need

n8n · webhook · code · openai · slack · gmail · google-sheets

Steps

1. Configure webhook trigger and input source

   Add a Webhook node in n8n, set method to POST, path to meeting-transcript. Use the generated URL as the webhook endpoint for Zoom, Google Meet, Otter.ai, etc., or create a manual submission form. Supports JSON, form data, and other input formats.

2. Configure OpenAI analysis node

   Add an OpenAI node, select Chat resource, and configure API credentials. Choose GPT-4 or GPT-3.5-turbo, temperature 0.3, max tokens 1500. The node includes a system prompt defining the output JSON structure, and a user prompt template injecting meeting metadata and transcript text.

   Execution prompt · Source text。OpenAI node system message field

   Copy prompt

   You are an executive assistant AI specializing in meeting analysis. Extract and structure information from meeting transcripts. Return ONLY valid JSON with these exact keys:
   
   {
     "executive_summary": "2-3 sentence overview",
     "key_decisions": ["decision 1", "decision 2"],
     "action_items": [
       {"task": "description", "owner": "name", "due_date": "YYYY-MM-DD or 'not specified'", "priority": "High/Medium/Low"},
     ],
     "discussion_topics": ["topic 1", "topic 2"],
     "open_questions": ["question 1"],
     "next_steps": ["step 1", "step 2"],
     "risks_identified": ["risk 1"],
     "follow_up_meeting_needed": true/false,
     "sentiment": "Positive/Neutral/Negative",
     "engagement_level": "High/Medium/Low"
   }

   Execution prompt · Source text。OpenAI node user message field (n8n expression, variables resolved at runtime)

   Copy prompt

   Meeting: {{ $json.title }}
   Date: {{ $json.date }}
   Attendees: {{ $json.attendees.join(', ') }}
   Duration: {{ $json.duration }} minutes
   
   Transcript:
   {{ $json.transcript }}

3. Configure intelligence synthesis and scoring code node

   Add a Code node (Synthesize Intelligence) running JavaScript to calculate action item priority scores (0-40 range) and meeting completeness score (0-100). The code parses OpenAI's JSON response, generates unique IDs for each action item, sorts them, and flags unassigned tasks. Thresholds can be customized as needed.

   Execution prompt · Source text。Synthesize Intelligence Code node jsCode field

   Copy prompt

   const meetingData = $('Parse Meeting Input').first().json;
   const aiResponse = JSON.parse($input.first().json.choices[0].message.content);
   
   // Calculate priority scores
   function calculatePriorityScore(actionItem) {
     const priorityScores = { 'High': 3, 'Medium': 2, 'Low': 1 };
     const hasOwner = actionItem.owner && actionItem.owner !== 'not specified' ? 1 : 0;
     const hasDueDate = actionItem.due_date && actionItem.due_date !== 'not specified' ? 1 : 0;
     return (priorityScores[actionItem.priority] || 2) * 10 + hasOwner * 5 + hasDueDate * 5;
   }
   
   // Enrich action items
   const enrichedActionItems = aiResponse.action_items.map((item, index) => ({
     id: `${Date.now()}-${index}`,
     task: item.task,
     owner: item.owner || 'Unassigned',
     due_date: item.due_date !== 'not specified' ? item.due_date : '',
     priority: item.priority,
     priority_score: calculatePriorityScore(item),
     status: 'Not Started',
     meeting_title: meetingData.title,
     meeting_date: meetingData.date,
     created_at: new Date().toISOString()
   }));
   
   // Sort by priority score
   enrichedActionItems.sort((a, b) => b.priority_score - a.priority_score);
   
   return {
     json: {
       // Meeting metadata
       meeting_id: `meeting-${Date.now()}`,
       meeting_title: meetingData.title,
       meeting_date: meetingData.date,
       attendees: meetingData.attendees,
       attendee_count: meetingData.attendee_count,
       duration: meetingData.duration,
       organizer: meetingData.organizer,
       meeting_url: meetingData.meeting_url,
       recording_url: meetingData.recording_url,
       
       // AI Analysis
       executive_summary: aiResponse.executive_summary,
       key_decisions: aiResponse.key_decisions,
       action_items: enrichedActionItems,
       action_item_count: enrichedActionItems.length,
       high_priority_count: enrichedActionItems.filter(a => a.priority === 'High').length,
       discussion_topics: aiResponse.discussion_topics,
       open_questions: aiResponse.open_questions,
       next_steps: aiResponse.next_steps,
       risks_identified: aiResponse.risks_identified,
       follow_up_meeting_needed: aiResponse.follow_up_meeting_needed,
       sentiment: aiResponse.sentiment,
       engagement_level: aiResponse.engagement_level,
       
       // Calculated flags
       requires_urgent_attention: enrichedActionItems.some(a => a.priority === 'High'),
       has_unassigned_tasks: enrichedActionItems.some(a => a.owner === 'Unassigned'),
       completeness_score: calculateCompletenessScore(enrichedActionItems, aiResponse)
     }
   };
   
   function calculateCompletenessScore(actionItems, analysis) {
     let score = 50; // Base score
     
     // Penalties
     if (actionItems.length === 0) score -= 20;
     if (actionItems.filter(a => a.owner === 'Unassigned').length > 0) score -= 10;
     if (actionItems.filter(a => !a.due_date).length > 0) score -= 10;
     if (analysis.open_questions.length > 3) score -= 10;
     
     // Bonuses
     if (analysis.key_decisions.length > 0) score += 15;
     if (analysis.next_steps.length > 0) score += 10;
     if (actionItems.every(a => a.owner !== 'Unassigned')) score += 15;
     if (actionItems.every(a => a.due_date)) score += 10;
     
     return Math.max(0, Math.min(100, score));
   }

4. Configure Slack summary post

   Add a Slack node, select OAuth2 authentication, create a Slack app with chat:write scope. Replace the channel ID placeholder with your actual channel (e.g., #meeting-summaries). The message template uses mrkdwn format with meeting summary, decisions, action items, and completeness score.

   Execution prompt · Source text。Slack node text field (mrkdwn message template)

   Copy prompt

   📋 *Meeting Summary: {{ $json.meeting_title }}*
   
   🗓️ *Date:* {{ $json.meeting_date.split('T')[0] }} | *Duration:* {{ $json.duration }}min
   👥 *Attendees:* {{ $json.attendee_count }} people
   📊 *Completeness:* {{ $json.completeness_score }}%
   
   *Executive Summary:*
   {{ $json.executive_summary }}
   
   *✅ Key Decisions ({{ $json.key_decisions.length }}):*
   {{ $json.key_decisions.map((d, i) => `${i+1}. ${d}`).join('\n') }}
   
   *🎯 Action Items ({{ $json.action_item_count }}):*
   {{ $json.action_items.slice(0, 5).map(a => `• [${a.priority}] ${a.task} - *${a.owner}* ${a.due_date ? '(Due: ' + a.due_date + ')' : ''}`).join('\n') }}
   {{ $json.action_item_count > 5 ? `\n...and ${$json.action_item_count - 5} more` : '' }}
   
   {{ $json.has_unassigned_tasks ? '⚠️ *Warning:* Some tasks are unassigned' : '' }}
   {{ $json.recording_url ? '🎥 <' + $json.recording_url + '|Watch Recording>' : '' }}

5. Configure Notion database sync

   Create two databases in Notion: Meeting Notes (fields: title, date, attendees, summary, action items, completeness, recording URL) and Action Items (fields: title, assigned to, due date, priority, status, meeting relation). Share both with your integration and add the Notion token. Configure Create Meeting Note and Create Task in Notion nodes with the database page IDs.

6. Configure email notifications and calendar reminders

   Add a Gmail node, configure OAuth2 or SMTP credentials. The email template is HTML with priority color-coding and meeting context. Add a Google Calendar node, enable Calendar API, create reminder events at 9:00 AM on due dates (adjustable), adding the task owner as attendee.

   Execution prompt · Source text。Gmail node message field (HTML email template)

   Copy prompt

   =<h2>New Action Item from Meeting</h2>
   
   <p><strong>Meeting:</strong> {{ $json.meeting_title }}<br>
   <strong>Date:</strong> {{ $json.meeting_date.split('T')[0] }}</p>
   
   <h3>Your Task:</h3>
   <p>{{ $json.task }}</p>
   
   <p><strong>Priority:</strong> <span style="color: {{ $json.priority === 'High' ? 'red' : $json.priority === 'Medium' ? 'orange' : 'green' }}">{{ $json.priority }}</span><br>
   <strong>Due Date:</strong> {{ $json.due_date || 'Not specified' }}</p>
   
   {{ $json.meeting_url ? '<p><a href="' + $json.meeting_url + '">View Meeting Details</a></p>' : '' }}
   
   <hr>
   <p style="font-size: 12px; color: #666;">This action item was automatically extracted from the meeting transcript using AI.</p>

7. Configure analytics tracking and test validation

   Create a Google Sheet with columns: Meeting_ID, Title, Date, Attendees, Duration, Action_Items, High_Priority, Decisions, Completeness, Unassigned_Tasks, Follow_Up_Needed. Configure the Google Sheets node to append rows. Test by POSTing a sample transcript, then verify Slack messages, Notion entries, emails, calendar events, and Sheets records.

You are an n8n workflow setup assistant. The user wants to replicate the "AI Meeting Summary Auto-Distribution" template. First confirm they have: n8n instance, OpenAI API key, Slack bot token (chat:write), Notion integration token, Google Workspace OAuth2 credentials. Then guide them step by step through configuring the webhook, OpenAI node, Slack/Notion/Gmail/Calendar/Sheets nodes, and remind them to replace all placeholders (channel ID, database page IDs, Sheet ID, email domain).