AI Agent in n8n to Automatically Extract Data from PDF Invoices

This initial step lets you manually test your PDF invoice processing. The Manual Trigger node in n8n is perfect for simulating the automation step-by-step and making sure every field (amount, supplier, IBAN, etc.) is correctly extracted.

It’s the best way to confirm that your AI agent reads each file and pulls the correct data before adding a real trigger like new email received, file added to drive, or API call.

Start your first test by clicking “Execute Workflow” in the n8n editor.

This step automatically scans a specific folder in your Google Drive to fetch all PDF invoices ready for processing. Each file will then be analyzed individually by the AI agent.

💡 Tip: To find your Drive folder ID, open the folder in your browser—the ID appears in the URL after /folders/.

This step uses a Loop node in n8n to process each invoice PDF individually. It ensures that every file is analyzed one by one, preventing data overlap or workflow collisions.

Looping over the list of files lets your automation treat every invoice as a separate item—from text extraction to AI analysis and data structuring.

This step automatically downloads the invoice PDF file from your Google Drive folder, using the dynamic file ID retrieved during the previous loop step.

💡 You can also replace Google Drive with a Gmail module, a webhook trigger, or your ERP's API if invoices come from another source.

The invoice is saved in PDF format inside a temporary server folder (/tmp/doc.pdf). This step is required to make the file accessible for text extraction using a terminal command.

This method works with any type of PDF: customer invoice, vendor invoice, credit note, purchase order, etc.

In this step, we use the pdftotext command (included in the Poppler library) to convert the PDF invoice into a plain text file. This format is essential for the AI agent to analyze and structure the information extracted from the invoice.

➡️ Command executed: pdftotext /tmp/doc.pdf /tmp/doc.txt

This method extracts all visible fields from an invoice: number, date, line items, VAT, total amount, IBAN, and more.

Not sure how to install pdftotext? Ask ChatGPT depending on your system (Ubuntu, Docker, Mac…) or contact us.

In this step, we use the Read File from Disk node to load the plain text content previously extracted from the PDF invoice. This data will then be passed to the AI agent for analysis and structured extraction.

This step is crucial to ensure that the AI receives clean, readable input data for accurate processing.

The raw text extracted from the invoice may contain unwanted line breaks, extra spaces, or repeated headers. This step is used to clean and standardize the content so it can be properly interpreted by the AI agent.

The cleaned output is stored in $json.data, ready to be passed to the analysis step. This ensures the AI model receives clear, usable input such as dates, invoice numbers, product lines, VAT, totals, and more.

The cleaned text is sent to an AI agent powered by GPT-4o, using LangChain. This agent is trained to automatically extract all key invoice data: invoice number, date, client, vendor, subtotal, total with tax, IBAN, product lines, and more.

➡️ Prompt: JSON-formatted output with standardized fields, optimized for Google Sheets (e.g. using apostrophes to prevent number formatting issues).

The JSON generated by the AI agent is transformed into a flat structure, with standardized fields (e.g., invoice_number, invoice_date, total_amount, client_name, etc.) for direct integration into Google Sheets.

💡 You can easily adapt this structure for other tools like Notion, Airtable, or your accounting database depending on your stack.

The extracted invoice information (amount, date, client, supplier, IBAN, etc.) is automatically added as a new row in a Google Sheet. Each column corresponds to a clearly defined field.

➡️ Connection: Google Sheets linked to your account

You can easily replace this output with Notion, Airtable, an ERP, an invoicing tool, or a SQL database depending on your needs.