FAQ answers completed by AI for each row of a Google Sheet

FAQ answers completed by AI for each row of a Google Sheet

An n8n workflow you start by hand. It reads four sheets of one Google Sheets file and takes every row whose status is
empty. For each row it builds five questions with answer starts and lets OpenAI finish each answer. It saves the
result as a JSON text file in a Google Drive folder and writes done into the row.

What is inside

• workflow.json: the workflow as n8n exports it, 36 nodes. 29 are working nodes and seven are sticky notes that
  explain the steps. Four of the working nodes (Strapi, WordPress, Webflow and HTTP Request) are not connected to
  anything. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• Strapi sat at the end of the run with no content set and no account. A run that reached it ended on an error,
  after the Drive file and the done mark were already made. Mindket's copy leaves it, with the WordPress, Webflow and
  HTTP Request nodes, unconnected on the canvas as CMS options for you to wire up.
• The sticky note about the sheets linked to a sample spreadsheet of its author. The link is gone, and the note now says
  which columns each sheet needs.
• The WordPress node was linked to a credential in its author's n8n. The link is gone.
• The file had no settings block, so Mindket's copy adds executionOrder v1, the setting n8n gives a new workflow.
• Nothing else. The question templates, the prompt and the model are the file's own.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A Google Sheets file with four sheets named Single Integration Native, Single Integration Cred-only,
  Single Integration Non-native and Categories. The first three need the columns displayName and status. The
  fourth needs Category name and status. A row with an empty status is picked up.
• A Google Drive folder for each sheet, and its id.
• A Google account for the Sheets and Drive credentials, and an OpenAI API key.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json. Save it at once,
   because Execute Workflow calls this same workflow by its id.
2. Add a Google Sheets credential. Select it in Get Services and in Update Row Status, and choose your spreadsheet
   in both.
3. Add a Google Drive credential and select it in Create From Text.
4. Open Prepare Job. Its outdir field holds one placeholder line for each sheet. Replace each with the id of the
   Drive folder for that sheet.
5. Add an OpenAI credential and select it in OpenAI Chat Model. The file uses gpt-4o-mini, which OpenAI shuts down
   on 14 January 2027. Choose gpt-6-luna there if you will run the workflow later.
6. Put two or three test rows in a sheet and leave status empty. Press Execute workflow.
7. Open the Drive folder for the sheet. Each row has a file, and its status cell now says done.

How a run goes

1. The manual trigger starts the run. Define Sheets lists the four sheet names and Sheets To List... splits them.
2. For Each Sheet... (Split In Batches 3) takes one sheet at a time. Get Services (Google Sheets 4.3) reads the rows
   whose status is empty. It always returns an item, so an empty sheet gives a blank one. Prepare Job adds the
   sheet's name and Drive folder to each row.
3. When every sheet is read, For Each Service... takes the rows one at a time. If has Data drops the blank items.
4. Execute Workflow starts a second run of this same workflow for each row. In it, Execute Workflow Trigger hands
   the row to Switch, which picks one of four templates by the sheet's name.
5. The template (a Set node in raw JSON mode) lists five questions, each with the start of an answer. `Question to
   List1 splits them and For Each Question...1` takes them one by one.
6. Needs AI Completion?1 sends a question to AI Completion1 (Basic LLM Chain 1.4) when its ai_completion is true,
   which it is for all of them. The model finishes the answer in up to three sentences of plain text, in the tone of
   the start it was given. Format QA Pair1 joins the start, the model's words and any append text.
7. Questions to Object...1 gathers the pairs and Format DisplayName + Questions1 names them after the row.
   Create From Text saves a file named after the row and the date, with the questions and answers as JSON.
   Update Row Status then writes done into the row.

Change the look

• Questions and answer starts: the four template nodes. Set ai_completion to false on a question to keep its answer
  as written.
• The instructions to the model: the system message and prompt text in AI Completion1.
• Model: model in OpenAI Chat Model.
• Sheets: the list in Define Sheets, a branch in Switch and a template node for each sheet you add, plus a line for
  it in the outdir field of Prepare Job.
• Where the answers go: connect one of the CMS nodes at the bottom of the canvas after Create From Text and fill its
  fields from the JSON the question loop produces.

Good to know

• The templates are written for n8n's own integration pages. Their questions and answer starts talk about setting up a
  service in n8n and about n8n's pricing, so rewrite them for your own site first.
• OpenAI bills five calls for every row.
• If outdir keeps its placeholders, the Drive step fails. A status column must exist in every sheet, or the read
  stops with an error about the column.
• A row is marked done only after its Drive file is saved, so a failed run leaves the row to be picked up again.
• Each row's name and the templates go to OpenAI. The answers go to Google Drive.
• The file holds no credential and no key.