Home · Business · Finance & Accounting · Bookkeeping & reconciliation
As of 13 August 2026, AI can find duplicate transactions in your books.
This still needs a person who signs their name to it.
Can you do it?
15 minutesto a draft.
30 minutesto something you’d act on.
Cost, all in£0
Skill neededchat-fluent
Who has to check ityou
What the alternative costsThe supplied tool information gives no price for a bookkeeping or reconciliation alternative.
If this goes wrong: a genuine duplicate is missed or a valid recurring payment is removed, leaving your books and possibly your reports incorrect.
What to actually do
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.
Use a tool built for this
Second choiceDo it yourself
The distant thirdA chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.
How to actually do it
- Open your business bank or bookkeeping system and export the relevant transactions as CSV or XLSX, including row numbers or transaction IDs, dates, amounts, currencies, descriptions, references and payees where available.
- Remove unnecessary account numbers, customer details and other private fields, while keeping the fields needed to match transactions, then save a copy of the original export unchanged.
- Open ChatGPT, Claude or Gemini, attach the cleaned export and paste the prompt above.
- Ask the model to produce the duplicate table and separate high-confidence matches from medium- and low-confidence investigations.
- Open the original bank statement and your bookkeeping record, then compare every proposed match by row number, date, amount, payee, reference and supporting invoice or receipt.
- Mark each pair as confirmed duplicate, valid separate transaction or unresolved, and make any correction in your bookkeeping system only after checking the underlying bank records.
- Run the reconciliation again or ask the model to check the revised export for remaining likely duplicates, without allowing it to alter the records directly.
Prompt
I have attached a CSV or spreadsheet export of business bank transactions. Find likely duplicate transactions, but do not delete, alter or invent any records. Compare date, amount, currency, description, reference, payee and account fields where available. Allow for ordinary differences in posting dates and descriptions, and do not treat recurring bills, transfers between my own accounts, refunds or split payments as duplicates without explaining why. Return a table with: transaction IDs or row numbers, the matching transaction IDs or row numbers, the fields that support the match, the fields that differ, a confidence label of high, medium or low, and a short reason. Separate exact or near-exact matches from cases needing manual investigation. At the end, list the rows you could not assess because required fields were missing. Do not make accounting or tax adjustments. Keep the original row numbers so I can check every result against the bank statement.
Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.
What it gets wrong
- AI cannot know whether two similar payments are separate recurring transactions, a transfer, a refund or a genuine duplicate without your business context.
- AI cannot prove that a transaction belongs in the books merely because two rows have the same amount and description.
- AI cannot access missing invoices, bank feeds or accounting-system audit trails unless you provide them.
- AI cannot take responsibility for deleting a transaction or for errors in VAT, management accounts or other reporting.
- You still have to open the source records and investigate borderline matches.
Even on a YES, the friction has a name: judgement under ambiguity, verification cost and private data access.
How we scored this
Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.
| Axis | Score (0–2) |
|---|---|
| Output | 2 |
| Inputs | 2 |
| Verification | 1 |
| Liability | 1 |
| Effort delta | 2 |
| Total | 8 / 10 |
The methodology and its thresholds are published in full.
FAQ
- Can ChatGPT find duplicate transactions in a spreadsheet?
- Yes. Give it a transaction export and ask it to compare dates, amounts, descriptions, references and payees while preserving the original row numbers. It will identify likely matches, but you must check them against the bank statement before changing your books.
- How accurate is AI at finding duplicate bank transactions?
- It is useful for finding exact and near-exact matches, but accuracy falls when dates shift, descriptions change, payments recur or transfers are split. Treat its output as a list of candidates, not proof that a transaction should be deleted.
- Can AI remove duplicate transactions from my accounts?
- It can suggest which transactions may be duplicates, but it should not remove them automatically. Confirm each case against the bank record and supporting paperwork, then make the correction in your bookkeeping system.
- Is it safe to upload bank transactions to AI?
- Use a service with controls suitable for business financial data, and upload only the fields needed for matching. Remove unnecessary account and personal details, retain the original export, and check how the service handles uploaded data before using it.
Nearby answers
Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.
The newsletter
AI news, new answers and product picks, straight to your inbox.