Receipts tend to live in all the wrong places: a pocket, the camera roll, an email, the bottom of a bag. AI can help turn those images and PDFs into an organized log. The key is keeping each extracted entry connected to the original receipt so questions are easy to resolve.

Start with a simple intake folder

For a small business, create one place for new receipts and a separate place for processed files. Photograph the entire receipt, including the merchant, date, totals, and any handwritten tip. If the paper is long, take overlapping images and identify them as parts of the same purchase.

Provide a short list of categories you already use, such as materials, software, travel, or office supplies. These are working labels for organization. A receipt image alone cannot establish the business purpose of a purchase or determine its tax treatment.

Choose the fields you need to review

A practical log includes a record ID, receipt date, merchant, total, currency, suggested category, business purpose, source filename, and review status. Add subtotal, tax, and tip when they are shown and useful. Do not combine different currencies into a single total without a defined conversion method.

Ask the assistant to leave unreadable fields blank and explain the uncertainty. A faint 8 can look like a 3; a credit-card authorization slip can be mistaken for the final bill. Confident formatting does not make the extracted amount correct.

  1. Run a small batch that includes a clear receipt, a faded one, and a duplicate.
  2. Compare every extracted amount in that batch with its original image.
  3. Confirm that duplicate detection flags possible repeats without deleting originals.
  4. Append reviewed entries to the log and keep an explicit queue for unresolved receipts.

Handle the awkward cases deliberately

Suppose a store receipt shows a $64.00 subtotal, $5.12 tax, and $69.12 paid. Those fields should reconcile. If a second image shows the same purchase, the assistant should flag a likely duplicate rather than count another $69.12. If the final total is unclear, the row belongs in review.

Refunds, split purchases, and missing receipts need their own handling. A refund should remain a separate record linked to the original purchase. A mixed personal and business purchase needs your allocation; the assistant should not decide it from the store’s name. A missing image should remain visible as a missing document.

Make repeat runs predictable

Use a stable receipt ID and a processing log so rerunning the same folder does not append the same expenses twice. Keep the receipt date distinct from the date you uploaded or processed it. Move a file to processed only after its log entry has been successfully saved, and preserve the original image.

Once the workflow works on representative receipts, you can consider scheduling it if your tools support that setup. Still review exceptions regularly and compare the log with your payment records. The payoff is less transcription and a clearer trail for your bookkeeping, with uncertain items waiting in one place instead of being hidden in the totals.

A prompt to make it your own

TRY THIS PROMPT

Process the receipts in [folder] into [expense log]. Extract receipt date, merchant, subtotal, tax, tip, total, and currency where shown. Use these suggested categories: [list]. Include a stable record ID, original filename, business-purpose field, and review status. Leave unreadable values blank and explain uncertainties. Flag possible duplicates without deleting originals. Reconcile subtotal plus tax and tip to the total when applicable. Do not infer business purpose or tax deductibility. Append without duplicating existing entries. Move a receipt to processed only after its row is saved, preserve the original, and return a concise exceptions list.

Replace the brackets with your details.

Automate the transcription while keeping the original receipt and unresolved questions easy to find.

Use-case inspiration: 9x’s ChatGPT Work video and companion article. Examples, instructions, and prompts written for AI Prompts.

Use the prompts with the tools and permissions available in your AI environment. Some workflows require connected accounts or uploaded files.