A stack of supplier invoices as PDF sits in the inbox, and every figure has to be typed into a table by hand. ChatGPT takes over this step: uploaded receipts are read and automatically turned into a structured table. The following workflow shows the path from the first file to the finished Excel table.
A chat turns uploaded receipts into a table
ChatGPT relies on its data analysis feature, which, according to the OpenAI help page, was specifically improved for text-rich documents such as PDFs. Uploaded files are evaluated in the background with self-generated program code, and the result appears as a table in the chat.
- Upload PDF invoices or photos of receipts in a new ChatGPT chat using the file icon
- Prompt ChatGPT to transfer date, amount, VAT, and supplier for each receipt into a table
- Check the resulting table in the chat against the original receipts
- Download the table as a CSV file using the download button above the table
- Open the CSV file in Excel and carry it over into the accounting records
The workflow mainly saves typing time
Instead of typing every line by hand, the table is ready within minutes – whether ten or fifty receipts are involved. For office staff and bookkeeping, the task shifts to checking the numbers rather than copying them out. Similar automation is offered by Copilot in Excel for month-end closing for the calculations once the receipts are on file.
Free to start, a subscription pays off for more receipts
A free ChatGPT account is enough for a first test; according to the help page on file uploads, free accounts may upload up to three files per day. For regular or larger volumes of receipts, OpenAI offers the ChatGPT Plus subscription, whose price for Germany is shown in euros only at checkout. Access runs through the browser at chatgpt.com as well as the mobile apps, and an account is required in every case.
For customer invoices containing personal data, one rule applies: according to the OpenAI help page on ChatGPT Business, content from Business and Enterprise accounts would not be used for model training by default, while private Free and Plus accounts are not automatically covered by that standard. Companies handling sensitive receipts should therefore consider a business account or redact customer data before uploading.
The reading feature reaches its limits with poorly scanned documents, handwritten notes, or unusual invoice layouts – wrong amounts can easily slip into the table there. A full comparison with the original receipts therefore remains mandatory before any figure is booked; for the actual bookkeeping entry, the workflow does not replace accounting software.
