Support chatbot webhook that logs chats to Google Sheets

Support chatbot webhook that logs chats to Google Sheets

An n8n workflow that answers customer questions over a webhook. Your website or app sends a question as a POST request. An OpenAI
model answers from the company details you type into one node, and the reply comes back as JSON with a confidence score and a
category. Every question and answer is also added to a Google Sheet. There is no vector database and no memory between messages.

What is inside

• workflow.json: the workflow as n8n exports it, seven nodes. Six of them run and one is a sticky note. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• AI: Generate Response was an n8n-nodes-base.openAi node at version 1.8. In n8n's current source, read on 12 November 2026,
  that type exists only at versions 1 and 1.1, for the old text-completion call. The node that takes chat messages is
  @n8n/n8n-nodes-langchain.openAi. The copy uses it at version 1.8 with the same settings: model gpt-4o-mini, the two messages,
  temperature 0.3 and 500 tokens. Output Content as JSON stands in for the old responseFormat setting.
• That node hands back JSON already parsed, where the old one gave text. Merge Response read it as text, so the copy accepts
  both forms.
• Validate Input made its IDs with Node's crypto module, which a Code node refuses unless the instance allows it. The copy
  makes the two IDs with a short function and needs no module.
• The system message read the company name and the company details from environment variables, and Log to Sheets read the sheet
  ID from a third. You can set environment variables only when you run n8n yourself, so the copy takes these values from the
  nodes. The placeholders are <YOUR_COMPANY_NAME>, <YOUR_COMPANY_INFO> and YOUR_SHEET_ID. The support address in rule 3 is
  <YOUR_SUPPORT_EMAIL>.
• The webhook path is support-chatbot now.
• The sticky note matches these changes. The credential links and the webhook ID are gone.
• Nothing else. The rules in the system message and the sheet columns are as the file had them.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• A Google account for a Google Sheets credential, and a sheet with a tab named Lite Conversations. Its first row needs these
  headings: Message ID, Session ID, Customer, Question, Answer, Confidence, Category, Received and Responded.
• A page, a script or a tool that can send a POST request to the webhook.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. In Log to Sheets, replace YOUR_SHEET_ID with the ID from the sheet's address, the part between /d/ and /edit. Select a
   Google Sheets credential.
3. Add an OpenAI credential and select it in AI: Generate Response.
4. In the same node, edit the first message. Replace <YOUR_COMPANY_NAME> with your name, <YOUR_COMPANY_INFO> with what the bot
   should know, and <YOUR_SUPPORT_EMAIL> with the address it hands over to. The brackets go.
5. Press Test workflow and send a request to the test address, shown in Receive Message:

   curl -X POST "<YOUR_N8N_URL>/webhook-test/support-chatbot" -H "Content-Type: application/json" -d '{"message":"What are your opening hours?","session_id":"demo-1","customer_name":"Sam"}'

1. Check the reply, then check the sheet for a new row.
2. Press Publish and call the production address, which uses /webhook/ in place of /webhook-test/.

How a run goes

1. Receive Message takes a POST on the path support-chatbot. It answers later, through Respond to Customer.
2. Validate Input reads message, or text, from the request body. It takes session_id, which it makes up if missing, and
   customer_name, which falls back to Customer. A message shorter than 2 characters stops the run with an error.
3. AI: Generate Response sends the system message and Customer <name> asks: with the question to gpt-4o-mini. The system
   message asks for friendly, short answers under 150 words, based only on your company details. The model must reply with JSON
   holding answer, confidence and category, one of faq, billing, technical, complaint or other.
4. Merge Response joins that reply with the request's fields. If the reply cannot be read, it answers `Sorry, I had trouble
   processing that. Please try again. and sets confidence` to 0.
5. Log to Sheets appends a row to Lite Conversations, and Respond to Customer returns this JSON to the caller:
   success, messageId, answer, confidence and category. Both branches leave Merge Response together.

Change the look

• The company details, tone and rules: the system message in AI: Generate Response.
• The model, the length cap and the temperature: the same node.
• The categories: the line in the system message that lists them.
• The sheet's tab or columns: Log to Sheets.
• Where the answers are logged: swap Log to Sheets for Airtable, a database or another node.

Good to know

• The webhook has no sign-in. Anyone who learns its address can send questions on your OpenAI key and add rows to your sheet. Keep
  the address private, or add Header Auth in Receive Message. A page on another site also needs that node's allowed origins set.
• Each message is answered alone. session_id is only written to the sheet, so the bot does not remember an earlier question.
• Your company details travel with every request. Keep them short and keep secrets out of them.
• confidence is a number the model writes, not a measured value. Do not route on it without testing.
• A message that is empty or one character long ends in an error, not in a JSON reply. Check the length on your side first.
• OpenAI's deprecations page, read on 12 November 2026, lists gpt-4o-mini-2024-07-18 for shutdown on 14 January 2027, with
  gpt-6-luna as the replacement. Pick another model in AI: Generate Response before then.
• Where the data goes: each question and your company details go to OpenAI, the rows go to Google Sheets, and the answer goes back
  to the caller.
• Billing: OpenAI charges for each message, in proportion to the size of your company details and the answer.