n8n Automated Multilingual Gmail Draft Reply with GPT-4o

Gmail label trigger → GPT-4o assess & detect language → branch to generate Japanese/English draft → save as Gmail reply draft

Who

Customer support or business users handling a high volume of multilingual Gmail inquiries who want to reduce manual drafting

You get

When a new email with a specified label arrives, automatically assess whether a reply is needed, detect the language, and generate a draft reply in that language saved into the original thread

You need

n8n · gmail

Steps

1. Configure the Gmail trigger

   Add a Gmail Trigger node in n8n and connect your Gmail OAuth2 credential. In Gmail, apply a label (e.g., Inquiry) to target emails and configure the same label filter in the node. Note: the source trigger's Search query (filters.q) is empty, so it applies no label filter and the scope is too broad; before the first test, create a dedicated test label in a test mailbox and set filters.q to label:your-test-label, then run the test. Adjust the polling interval (pollTimes) as needed. The trigger outputs fields including subject, textAsHtml, headers.from, and threadId for downstream use.

2. Assess the email and identify language

   Add a Basic LLM Chain node (Assess if a message needs a reply and identify the language), connect an OpenAI GPT-4o model node (model=gpt-4o, temperature=0, responseFormat=json_object), and a Structured Output Parser node. In the LLM Chain, set the prompt field to the email subject and body; in messages, place the system prompt asking the model to return needsReply (boolean) and message_language (English / Japanese / Other) as JSON. The Parser’s JSON Schema must require needsReply and message_language.

   Execution prompt · Source text。Basic LLM Chain node prompt field (email content input)

   Copy prompt

   =Subject: {{ $json.subject }}
   Message:
   {{ $json.textAsHtml }}

   Execution prompt · Source text。Basic LLM Chain node messages.messageValues[0].message (system prompt)

   Copy prompt

   Your task is to :
   - identify the language of the message and set the value to field "message_language" : "English", "Japanese" or "Other"
   - assess if the message requires a response and set the value to field "needsReply" : true if it does, false otherwise.
   Return in JSON format.
   Marketing emails don't require a response.

   Execution prompt · Source text。Structured Output Parser node jsonSchema field

   Copy prompt

   ={
     "type": "array",
     "items": {
       "type": "object",
       "properties": {
         "needsReply": {
           "type": "boolean"
         },
         "message_language": {
           "type": "string"
         }
       },
       "required": ["needsReply", "message_language"]
     }
   }

3. Branch by detected language

   Add an If node to check whether needsReply is true; if true, route to a Switch node. The Switch branches on message_language: equals Japanese goes to the Japanese branch, equals English goes to the English branch. fallbackOutput is set to 1 (English). To add more languages, create additional rules in the Switch and connect corresponding generation nodes.

4. Generate a draft reply in the detected language

   On the Japanese branch, place a Basic LLM Chain node (Generate email for a Japanese client); on the English branch, place another (Generate email reply for an English client). Each connects to its own OpenAI GPT-4o model sub-node. In both nodes, the text field references the Gmail trigger’s subject and textAsHtml; the messages field contains the Japanese or English system prompt defining tone, salutation, closing, placeholder rules, and yes-no dual-response format.

   Execution prompt · Source text。Generate email for a Japanese client and Generate email reply for an English client nodes text field (email content input)

   Copy prompt

   =Subject: {{ $('Gmail trigger').item.json.subject }}
   Message: {{ $('Gmail trigger').item.json.textAsHtml }}

   Execution prompt · Source text。Generate email for a Japanese client node messages.messageValues[0].message (Japanese generation system prompt)

   Copy prompt

   You're a helpful personal assistant and your task is to draft replies on my behalf to my incoming emails. Whenever I provide some text from an email, return an appropriate draft reply for it and nothing else.
   Ensure that the reply is suitable for a professional email setting and addresses the topic in a clear, structured, and detailed manner.
   Do not make things up.
   
   Detailed instructions:
   - Be concise and maintain a business tone (desu-masu).
   - Follow the following content 
   [INSERT CLIENT NAME] 様
   お世話になっております。
   株式会社 [INSERT YOUR COMPANY NAME] の[INSERT YOUR NAME]です。
   
   BODY (main content)
   
   引き続き、何卒どうぞ宜しくお願いいたします。
   
   - When replying to yes-no questions, draft 2 responses: one affirmative and one negative separated by " - - - - - - - OR - - - - - - - "
   - If you don't know an answer, you can leave placeholders like "[YOUR_ANSWER_HERE]".
   - Don't use any special formatting, only plain text.
   - Make sure each request is separated with a new line break
   - Reply in the same language as the inbound email.

   Execution prompt · Source text。Generate email reply for an English client node messages.messageValues[0].message (English generation system prompt)

   Copy prompt

   You're a helpful personal assistant and your task is to draft replies on my behalf to my incoming emails. Whenever I provide some text from an email, return an appropriate draft reply for it and nothing else.
   Ensure that the reply is suitable for a professional email setting and addresses the topic in a clear, structured, and detailed manner.
   Do not make things up.
   
   Detailed instructions:
   - Be concise and maintain a business casual tone.
   - Start with "Dear,", and end with "Best regards,"
   - When replying to yes-no questions, draft 2 responses: one affirmative and one negative separated by " - - - - - - - OR - - - - - - - "
   - If you don't know an answer, you can leave placeholders like "[YOUR_ANSWER_HERE]".
   - Don't use any special formatting, only plain text.
   - Reply in the same language as the inbound email.

5. Save as a Gmail reply draft

   Add a Gmail node (Gmail - Create Draft), set resource to draft and emailType to html. The message field replaces newlines with <br />\n; the subject prefixes Re: and references the original subject; options.sendTo references the original sender and options.threadId references the original thread ID so the draft appears as a reply. Connect your Gmail OAuth2 credential.

   Execution prompt · Source text。Gmail - Create Draft node message field (HTML conversion)

   Copy prompt

   ={{ $json.text.replace(/\n/g, "<br />\n") }}

   Execution prompt · Source text。Gmail - Create Draft node subject field

   Copy prompt

   =Re: {{ $('Gmail trigger').item.json.headers.subject }}

   Execution prompt · Source text。Gmail - Create Draft node options.sendTo field

   Copy prompt

   ={{ $('Gmail trigger').item.json.headers.from }}

   Execution prompt · Source text。Gmail - Create Draft node options.threadId field

   Copy prompt

   ={{ $('Gmail trigger').item.json.threadId }}

Import template 4870 into n8n, connect Gmail OAuth2 and OpenAI API credentials, confirm the label name and polling interval, then activate the workflow. For the first run, test with a single email to verify language detection and draft generation.