As of 13 August 2026, AI can detect duplicate invoices.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.

   How to actually do it

   1. Open your accounting system or invoice folder and export the relevant invoice register, including supplier, invoice number, dates, net, VAT and gross amounts, purchase-order references, currency and payment status.
   2. Collect the original PDF or image files for the invoices, plus the purchase-order records and payment ledger for the period being checked.
   3. Remove unrelated personal information and redact bank details unless they are necessary for your internal matching process, then give each invoice a stable record ID.
   4. Paste the register into a chatbot or upload the register and invoice files, then use the supplied prompt to produce exact and likely duplicate pairs.
   5. Open every flagged invoice pair and compare the original documents, purchase order and payment ledger against the model's matching and conflicting fields.
   6. Mark each pair in your accounting system as duplicate, legitimate repeat charge, unresolved or not a duplicate, and ask your finance lead or accountant to resolve any unresolved pair before cancelling payment or altering the ledger.

   Prompt

   I need to detect possible duplicate invoices in a UK business invoice register. I will provide either a spreadsheet, invoice files, or both.
   
   Compare the records using invoice number, supplier name, supplier reference, invoice date, due date, net amount, VAT amount, gross amount, currency, purchase-order number, description and bank details where available. Normalise spacing, punctuation, capitalisation and common supplier-name variations before matching.
   
   Do not mark records as duplicates solely because they have the same amount or supplier. Flag exact matches and likely matches separately. Treat credit notes, recurring legitimate invoices, partial invoices, corrected invoices, invoices in different currencies and invoices with different purchase orders as possible reasons not to call two records duplicates.
   
   Return a table with these columns: record ID 1, record ID 2, match category, matching fields, conflicting fields, confidence as high, medium or low, and a short reason. Then provide a separate list of records that could not be assessed because required fields or source documents were missing. Do not invent missing values and do not make a final accounting decision. State which flagged pairs I must compare against the original invoice files, purchase-order records and payment ledger before blocking payment or changing the accounts.
   
   Data and files:
   [PASTE THE INVOICE REGISTER OR ATTACH THE FILES HERE]

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.