n8n Google Drive Organizer: AI File Classification

To build an n8n Google Drive organizer, watch one inbox folder with the Google Drive Trigger, download each new file, extract its text, let an AI node pick a category, then rename the file and move it to that category's folder. Scanned PDFs need one extra step, Google Drive's own OCR, and anything the model cannot place goes to a review folder instead of being guessed.

Checked October 2026 against n8n's docs for the Google Drive Trigger and its common issues, the Google Drive file operations, the Text Classifier and Information Extractor, plus Google's Drive push notifications, uploads and conversion and search query guides.

This is a docs-based build: every node setting below comes from n8n's and Google's documentation, and the design choices (one inbox, a review branch, a processed marker) come from running file automations at agency scale, where the expensive failure is a document filed in the wrong place without anyone noticing. If you want more workflows in the same style, the n8n templates and automation library is the hub for this series.

What the n8n Google Drive organizer does

You drop a file into one Drive folder called Inbox. Within about a minute it has a clean name like 2026-09-14 Acme Invoice and sits in /Finance/Invoices. If the file is unclear, it lands in /Inbox/Needs review with its name unchanged, so nothing is silently misfiled.

Before you start

Create the destination folders before you build. The Move operation needs a target folder, and choosing it from a list in the node is easier than wiring folder creation into the workflow. If you have not connected Google to n8n before, n8n's Google credential docs cover OAuth2 and service accounts.

Step 1: set up the n8n Google Drive trigger

Add a Google Drive Trigger node. Set Trigger On to Changes Involving a Specific Folder, choose your Inbox folder, and set Watch For to File Created. Under Options you can restrict File Type if you only want, say, images or Google Docs; for a mixed inbox leave it on all types.

Three behaviours from the docs shape the rest of the build:

• It polls. The node checks Drive at a set interval, once every minute by default, and starts the workflow for each matching event (n8n common issues page, checked October 2026).
• It batches. If several files arrive in one interval, they come through as several items in a single run. Build every later node to handle many items, not one.
• It ignores subfolders. The node shows a notice that changes inside subfolders will not trigger it. That suits this design: Needs review sits inside Inbox, so files you park there do not re-enter the workflow.

Expected result: clicking Test step returns the most recent file created in Inbox. If Inbox is empty, the manual run throws an error, because the docs say a manual execution returns the last matching event and errors when none exists. Drop a sample file in and test again.

n8n Google Drive trigger: polling or push notifications?

People often ask whether they should replace the polling trigger with Drive push notifications for an instant reaction. The Drive API does support push through files.watch and changes.watch, and you can receive those calls with n8n's Webhook node. Reading Google's guide, though, push adds work in four places:

The details behind that table, all from Google's push notifications guide (checked October 2026): the callback address must use HTTPS with a valid certificate, so self-signed certificates are rejected. Channels expire after at most 86,400 seconds (one day) for a file and 604,800 seconds (one week) for changes, and there is no automatic renewal; you call watch again with a new channel ID. Notification bodies are empty, so the workflow still has to call changes.list or files.get to learn what changed. And you should set a channel token and check it, so the webhook can reject calls it did not ask for.

For filing documents, a one-minute delay costs nothing, so the built-in trigger wins on simplicity. Push makes sense when a person is waiting on the result, for example a form upload that should be acknowledged within seconds, and you are willing to run a second, scheduled workflow that renews channels before they expire.

Step 2: download the file and extract its text

The trigger hands you file metadata, not content. Add a Google Drive node with Resource File and Operation Download, and pass the file ID from the trigger. The file arrives as binary data. For native Google files, the node's Google File Conversion option sets the export format; pick Text for Google Docs so the next step receives plain text.

Then route by file type with a Switch node on the MIME type:

• PDFs go to an Extract From File node set to Extract From PDF, which pulls the text layer out of the binary.
• Google Docs already arrive as text from the download conversion.
• Images (JPEG, PNG and similar) have no text layer and go straight to the OCR fallback below.

Step 3: the OCR fallback for scanned files

A scanned PDF looks like a document but is really a picture, so Extract From PDF returns empty or near-empty text. Add an IF node after extraction that checks whether the text is shorter than a threshold you choose (for example, under 50 characters as an illustrative starting point), and send those files, along with images, to OCR.

Google Drive does the OCR for you. Its upload and conversion guide says that when you convert an image to a Google Doc, Drive uses optical character recognition and puts the extracted text in the document next to the image, and that JPEG, PNG, GIF, BMP and PDF files can be converted to Docs. To convert, you create the file with the Google Docs MIME type, and an ocrLanguage parameter improves accuracy when you know the language (checked October 2026).

Expected result: a phone photo of a receipt comes back as readable text. OCR quality depends on the scan; a crooked, low-light photo can still fail, and that is what the review branch is for.

Step 4: classify the file with AI

Add n8n's Text Classifier node and connect a chat model to it. Feed it the extracted text, ideally only the first couple of thousand characters, since the category is almost always clear from the first page and shorter input costs less. Each category has a name and a description, and the description is where the accuracy comes from: tell the model what separates an invoice from a receipt in your files.

Two settings matter. Turn Allow Multiple Classes To Be True off, so each file gets exactly one folder. Set When No Clear Match to Output on Extra, 'Other' Branch; the default, Discard Item, silently drops files the model cannot place, which in a filing workflow means a document that simply never moves. The Text Classifier docs describe both options (checked October 2026).

Step 5: rename and move each file

A good file name needs a date and a party, which the classifier does not give you. Add an Information Extractor node on each category branch with a small schema, for example document_date and party_name, using the From Attribute Descriptions option. Tell it to return an empty value rather than guess when a field is missing.

The processed marker earns its place the first time you run a backfill. Drive search supports queries such as properties has { key='organized' and value='true' }, per Google's search guide, so you can find files the workflow already handled, or the ones it missed, with the Google Drive node's Search operation and its Advanced Search query string.

What else can n8n Google Drive automation do?

The same node set covers most Drive jobs people automate. The Google Drive node can copy, create from text, delete, download, move, share, update and upload files, search files and folders, create and share folders, and manage shared drives, per the node overview (checked October 2026). Common patterns built on this organizer:

• Backfill an old folder. Swap the trigger for a manual or scheduled run, Search the folder with 'FOLDER_ID' in parents and trashed = false, and push each result through the same classify path.
• Log every file. Append the name, category and link to a spreadsheet so you have an index you can filter.
• Share by category. Use the Share operation to give your accountant reader access to the Invoices folder only.
• Send files to an agent. Feed the extracted text into a retrieval store so an assistant can answer questions about your documents, the pattern in our guide to n8n AI workflows with OpenAI.

When the Drive node lacks an operation, the n8n docs recommend the HTTP Request node with your existing Drive credential, the same route the OCR step uses. Our walkthrough of connecting any API in n8n covers that pattern in more depth.

If you want to package automations like this into a product clients pay for, with a front end, auth and billing around the workflow, our AI SaaS Builder program covers that build step by step.

Troubleshooting

• 401 unauthorized_client on the trigger. Enable the Google Drive API in your Cloud project. For a service account, enable domain-wide delegation and add the Drive API to it (n8n common issues page).
• Only one file processed when three arrived. A node is reading only the first item. Check expressions that reference a single item, and use a Loop Over Items node if a step must handle files one at a time.
• Empty text from a PDF. It is a scan. Confirm your length check routes it to the OCR branch.
• Files reprocessed after renaming. You watched for File Updated as well as File Created. Watch only File Created, and check the processed property before acting.
• Workflow fails halfway and leaves files in Inbox. Add an error workflow so failures reach you; our n8n error handling guide shows the setup.

n8n Google Drive: FAQ