Build a PDF to Spreadsheet Flow

Retyping invoice numbers into a spreadsheet is exactly the kind of error-prone chore to hand off. So here is a Gumloop workflow that reads a PDF, extracts the fields you care about, and files them in a sheet, no copy-paste.

What will you build?

A PDF-to-spreadsheet workflow. You give it a PDF, it reads the document, an AI step pulls out the fields (say vendor, date, total, line items), and they land as a row in a Google Sheet. Great for invoices, receipts, and reports.

What do you need to start?

• A free Gumloop account.
• A PDF to test on.
• A Google Sheet for the extracted rows.

How does it work?

Node	What it does
Input	The PDF you upload
PDF Reader	Parses the document text
Ask AI	Extracts the fields you name
Output	Adds a row to your sheet

How do you build it, step by step?

Start a New Workflow and type:

Take a PDF invoice, read it, and use AI to extract the vendor, date,
total, and line items. Add them as a new row in a Google Sheet.

Then sharpen the AI step so it is consistent:

Extract these fields from the document: vendor, date, total, and a
list of line items. If a field is missing, write "unknown".
Document: {{ pdf text }}

How do you set the model to Claude?

Open the AI step or agent and pick the model. Extraction rewards a careful model. For careful work I use Claude 4.8 Opus, the “Smartest” option; a faster model saves credits on simple runs.

How do you run it?

Upload a PDF and run it. Check the row it writes, then process the rest. For a folder of invoices, connect them in and let it work through.

What can go wrong?

• Scanned images. A photo of a page reads less cleanly than a true PDF. Use text PDFs where you can.
• Missing fields. The prompt tells it to write “unknown” rather than guess. Keep that line.
• Odd layouts. Unusual invoice layouts may need a tweak to the prompt. See the AI Tools Directory for more.

Common questions

Do I need to code it?

No. You describe the fields you want and Gumloop builds the steps.

What files work best?

Text-based PDFs read most cleanly. Scanned images are harder but often still work.

Where do the results go?

As rows in a Google Sheet, or back to you, whichever you choose for the output.

Can it handle many PDFs?

Yes. Connect a folder or list and it processes them one after another.

Does it use Claude?

The extraction step can. Claude 4.8 Opus is the “Smartest” option for careful reading.

Sources

• Gumloop documentation
• Gumloop official site
• Anthropic: Claude models

Last reviewed: June 2026. Gumloop changes often; check the official docs above if a button has moved.

You may also like

• AI Automation for Beginners (the hub)
• How to use Claude AI
• The AI Tools Directory
• Start Here: the beginner path
• The AI Glossary
• The daily Beginners in AI newsletter