RFP questions answered into a Google Doc by AI

RFP questions answered into a Google Doc by AI

An n8n workflow that takes a request for proposal (RFP) as a PDF sent to a webhook. It creates a Google Doc for the answers and
asks a model to pull every question out of the PDF. An AI agent then answers the questions one at a time from a company
reference text you paste into its system message, and each question and answer goes into the Doc. When the last one is
written, Gmail and Slack tell the team.

What is inside

• workflow.json: the workflow as n8n exports it, fourteen nodes and ten sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The question answering used n8n's OpenAI Assistant node. OpenAI's deprecations page, read on 12 November 2026, gives
  26 August 2026 as the shutdown of the Assistants API, with the Responses API and Conversations API as the replacement. The
  step is now an AI Agent, Answer Question with Context, with its own model node, OpenAI Chat Model1, calling
  gpt-5.6-terra. The author's assistant held company documents. Here you paste a company reference text into the agent's
  system message instead, in place of <PASTE YOUR COMPANY REFERENCE TEXT HERE>. The system message is ours, and it tells the
  agent to say what is missing when the text does not cover a question.
• The first model node set no model, so n8n used its default, gpt-5-mini. OpenAI's page gives 11 December 2026 as its
  shutdown, with gpt-5.6-terra as the replacement, so OpenAI Chat Model now names gpt-5.6-terra.
• Item List Output Parser had no Number Of Items setting. With an empty setting n8n's code asks the model for three items,
  stops after three and fails when fewer come back, so only the first three questions of an RFP would be answered. It is now
  set to -1, which turns the limit off.
• Send Chat Notification and Send Email Notification read $('Set Variables').item.json.title, a field that node does not
  produce. They now read doc_title.
• The webhook path was a generated id. It is now rfp-questionnaire.
• Create new RFP Response Document pointed at one of the author's Drive folders. It now uses default, the top of your
  Drive.
• Add Metadata to Response Doc wrote an execution link that began with http://localhost:5678. The start is now
  https://<YOUR_N8N_HOST>.
• The text recorded for each question was taken from the output of the assistant node. It now reads the question from
  For Each Question... and the answer from the agent.
• Three sticky notes told you to set up an assistant and linked to the author's files. They now describe the agent and the
  pasted text.
• The Google Docs, Gmail, Slack and OpenAI nodes lost their links to credentials in the author's n8n. Select your own.

Before you start

• An n8n instance, on n8n Cloud or your own server, that can receive a webhook from where you send the RFP.
• An OpenAI API key that can use gpt-5.6-terra.
• A Google account for Google Docs and Gmail.
• A Slack workspace, an app that can post messages and a channel.
• An RFP as a PDF with text you can select, and a short company document to paste.
• A tool that can send a form with a file, such as curl or Postman.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential and select it in OpenAI Chat Model and OpenAI Chat Model1.
3. Open Answer Question with Context and replace <PASTE YOUR COMPANY REFERENCE TEXT HERE> in the system message with your
   company text.
4. Add a Google Docs credential and select it in Create new RFP Response Document, Add Metadata to Response Doc and
   Record Question & Answer in Response Doc.
5. In Add Metadata to Response Doc, replace <YOUR_N8N_HOST> with the address of your n8n.
6. Add a Gmail credential and select it in Send Email Notification. Add a Slack credential and select it in
   Send Chat Notification. The file names the channel RFP-channel, so pick yours.
7. Click Execute workflow. Wait for Request listens at the test URL, /webhook-test/rfp-questionnaire on your n8n. Send it
   a form with the fields id, title and reply_to and the PDF in a field named data. The sticky note on the canvas holds a
   curl example.
8. Open the new Doc in your Drive and read the answers. When they look right, publish the workflow and use the production URL,
   /webhook/rfp-questionnaire.
9. Before you share the production URL, add authentication to Wait for Request.

How a run goes

1. Wait for Request takes the POST. It answers the sender at once, before the work is done.
2. Get RFP Data reads the text of the PDF from the field data.
3. Set Variables builds three values: doc_title from title, doc_filename from the id, the title and the time, and
   reply_to.
4. Create new RFP Response Document makes an empty Google Doc with that file name at the top of your Drive.
5. Add Metadata to Response Doc writes the title, the date, the requester and a link to the execution at the top of the Doc.
6. Extract Questions From RFP (Basic LLM Chain, with OpenAI Chat Model) is told to take every question meant for the
   supplier, exactly as written. Item List Output Parser splits the reply at each new line, so each question becomes an item.
7. For Each Question... hands the questions on one at a time. Answer Question with Context answers one from the reference
   text, and Record Question & Answer in Response Doc adds a numbered question and answer to the end of the Doc.
8. When the loop is done, Send Email Notification emails the person in reply_to, then Send Chat Notification posts to Slack.

Change the look

• The answer rules and the company text: the system message in Answer Question with Context.
• Which questions count: the prompt in Extract Questions From RFP.
• The layout of the Doc: the text in the two Docs update nodes.
• The file name and the folder: doc_filename in Set Variables, and the folder in Create new RFP Response Document.
• Who is told and how: Send Email Notification and Send Chat Notification.
• The models: OpenAI Chat Model and OpenAI Chat Model1.

Good to know

• Anyone who knows the webhook address can start a run, spend your OpenAI key and make your Gmail email any address they put in
  reply_to. Add header authentication to Wait for Request before you use it outside a test.
• The answers are drafts. The agent writes from your pasted text and is told not to guess, but a person should read every
  answer before it goes into a bid.
• The pasted text has to fit in the agent's prompt next to each question. A large library of documents needs a different
  approach, such as a search tool for the agent.
• A scanned PDF has no text to read, so Get RFP Data returns little and the model finds few questions.
• The whole PDF text goes to the model in one prompt. A long RFP can exceed what the model takes.
• The email says the document is complete but carries no link, although a sticky note says it does. Add the Doc link to
  Send Email Notification if you want one.
• Each question is its own agent call, plus one call to extract the list, so a long RFP costs many calls.
• The model reads your RFP and your company text, the Doc holds the answers in your Drive, and the document title goes to
  Slack and the requester's email.
• The execution link in the Doc opens only for people who can sign in to your n8n.