Auto-categorize Gmail and draft replies with Groq in n8n

Gmail trigger → field extraction → Groq AI analysis → JSON parsing → priority alert + Google Sheets logging → conditional draft reply

Who

Users who need to automate Gmail inbox processing with AI-assisted categorization, summarization, and pre-drafted replies

You get

New emails are auto-categorized, summarized, and draft-replied; high-priority emails trigger instant alerts; all analyses are logged to Google Sheets

You need

n8n · gmail · google-sheets

Steps

1. Configure Gmail trigger and content filter

   Add a 'Gmail Trigger' node in n8n, set polling interval to every minute (pollTimes: everyMinute). Add an 'If' node (Check Email Content) with condition {{ $json.text || $json.snippet }} not equal to empty string, skipping emails without body or snippet.

2. Extract key email fields

   Add an 'Edit Fields (Set)' node mapping: sender (from address), sender_name (from name), subject (default '(No Subject)'), body (text or snippet truncated to 3000 chars), email_id, received_date.

3. AI analysis and categorization

   Add an 'AI Agent' node (Email Analysis Agent), set promptType to define. Fill the text field with the user prompt and options.systemMessage with the system prompt. Connect a 'Groq Chat Model' sub-node, select model llama-3.3-70b-versatile, temperature 0.3, maxTokensToSample 2048.

   Execution prompt · Source text。AI Agent node text field (user prompt)

   Copy prompt

   =Analyze and categorize this email, then draft an appropriate response.
   
   FROM: {{ $json.sender_name }} <{{ $json.sender }}>
   SUBJECT: {{ $json.subject }}
   BODY:
   {{ $json.body }}
   
   Respond ONLY with valid JSON.

   Execution prompt · Source text。AI Agent node options.systemMessage field

   Copy prompt

   You are an expert email assistant. Your job is to analyze incoming emails, categorize them by urgency and type, and draft professional responses.
   
   You MUST respond ONLY with valid JSON. No markdown, no code blocks.
   
   Required JSON structure:
   {
     "category": "One of: URGENT | ACTION_REQUIRED | FOLLOW_UP | FYI | NEWSLETTER | SPAM | PROMOTION | PERSONAL",
     "priority": "One of: HIGH | MEDIUM | LOW",
     "summary": "2-3 sentence summary of what this email is about",
     "key_asks": ["What the sender wants — each ask as a separate item"],
     "sentiment": "One of: positive | neutral | negative | frustrated",
     "requires_response": true,
     "suggested_response_time": "Respond within: 1 hour / 4 hours / 24 hours / 48 hours / No response needed",
     "draft_response": {
       "subject": "Re: original subject or new subject if needed",
       "body": "Professional, concise draft reply. Match the tone of the original email. If the email is a newsletter or promotion, leave this empty.",
       "tone": "The tone used: professional / friendly / formal / apologetic / enthusiastic"
     },
     "action_items": ["Any to-do items this email creates for you"],
     "labels_suggested": ["Suggested Gmail labels: work, personal, finance, travel, project-X, etc."]
   }
   
   Rules:
   - SPAM/NEWSLETTER/PROMOTION: set requires_response to false, draft_response body to empty string
   - URGENT: anything with deadlines within 24hrs, payment issues, or critical requests
   - ACTION_REQUIRED: needs you to do something but not time-critical
   - FOLLOW_UP: someone following up on a previous conversation
   - FYI: informational, no action needed
   - Draft responses should be concise (under 150 words)
   - Never be sycophantic in drafts — be professional and direct
   - If the email is in a non-English language, respond in the same language
   - Extract ALL action items, even implicit ones

4. Parse and validate AI output

   Add a 'Code' node (Parse AI Analysis) running JavaScript to parse the AI JSON response. Try JSON.parse first, fallback to markdown code block extraction or regex match for {…}, return parse_status: failed if all fail. On success, normalize all fields (category, priority, summary, key_asks, sentiment, requires_response, suggested_response_time, draft_subject, draft_body, draft_tone, action_items, labels).

   Execution prompt · Source text。Parse AI Analysis node jsCode field

   Copy prompt

   const items = $input.all();
   const output = items[0].json.output;
   
   let parsed;
   try {
     parsed = JSON.parse(output);
   } catch (e) {
     const jsonMatch = output.match(/```(?:json)?\s*([\s\S]*?)```/);
     if (jsonMatch) {
       try { parsed = JSON.parse(jsonMatch[1].trim()); } catch (e2) { parsed = null; }
     }
     if (!parsed) {
       const objMatch = output.match(/\{[\s\S]*\}/);
       if (objMatch) {
         try { parsed = JSON.parse(objMatch[0]); } catch (e3) { parsed = null; }
       }
     }
   }
   
   if (!parsed) {
     return [{ json: { parse_status: 'failed', error: 'Could not parse AI analysis' } }];
   }
   
   return [{
     json: {
       parse_status: 'success',
       timestamp: new Date().toISOString(),
       sender: items[0].json.sender || '',
       sender_name: items[0].json.sender_name || '',
       original_subject: items[0].json.subject || '',
       email_id: items[0].json.email_id || '',
       category: parsed.category || 'FYI',
       priority: parsed.priority || 'LOW',
       summary: parsed.summary || '',
       key_asks: Array.isArray(parsed.key_asks) ? parsed.key_asks.join('; ') : '',
       sentiment: parsed.sentiment || 'neutral',
       requires_response: parsed.requires_response || false,
       suggested_response_time: parsed.suggested_response_time || '',
       draft_subject: parsed.draft_response?.subject || '',
       draft_body: parsed.draft_response?.body || '',
       draft_tone: parsed.draft_response?.tone || '',
       action_items: Array.isArray(parsed.action_items) ? parsed.action_items.join('; ') : '',
       labels: Array.isArray(parsed.labels_suggested) ? parsed.labels_suggested.join(', ') : ''
     }
   }];

5. Priority check, alert, and logging

   Add an 'If' node (Check Parse Success) verifying parse_status equals success. On success, route to 'If' (Evaluate Priority Level) checking if priority equals HIGH: if yes, trigger 'Gmail' node (Send High Priority Alert) to send an HTML alert to your address. Regardless of priority, append the record via 'Google Sheets' node (Append to Google Sheets) to the 'Email Log' sheet.

6. Conditional draft reply creation

   After Google Sheets logging, add an 'If' node (Assess Reply Necessity) checking if requires_response is true. If true, trigger 'Gmail' node (Draft Gmail Reply) with operation createDraft to create a draft reply; if false, the workflow ends.

You are setting up an n8n workflow to monitor Gmail, categorize emails with Groq AI, and draft replies. Please confirm you have ready: 1) Gmail OAuth2 credential (read + draft permissions); 2) Groq API key; 3) Optional Google Sheets credential and Sheet ID. Then import the workflow JSON or create nodes manually, checking that the AI Agent's systemMessage and Parse AI Analysis field mappings match downstream If conditions.