ChatGPT can sort a month of bank transactions into categories, add them up and point out odd entries, as long as you give it clean rows and check its sums yourself. The reliable route is a CSV or a pasted block of rows, one question at a time, and a final comparison with the totals printed on your statement.
What ChatGPT is good at with a statement, and what it is not
ChatGPT is usually described as an AI assistant people use to analyse pasted statements. It reads messy narration such as UPI strings well, so grouping a line like a food-delivery debit under eating out is easy for it. It is weaker at long arithmetic over hundreds of rows, and it has no memory of last month unless you paste last month again.
- Good: grouping merchants, spotting repeat debits, drafting a category list, explaining an unfamiliar narration.
- Shaky: exact totals over a long list, keeping the same categories from one chat to the next.
- Not its job: telling you what to cut or what to invest in.
How do I prepare the statement before pasting?
- Download the statement as CSV or Excel if your bank offers it; a PDF often breaks columns when pasted.
- Keep only four columns: date, description, amount, and whether it is a debit or credit.
- Delete the header block with your name, address and account number, and the running balance column.
- Check the row count and the statement's own total debits and total credits, and write them down.
Step four is the one people skip. Without it you have nothing to test the answer against.
A worked example with made-up numbers
Say a statement has these rows after cleaning. All figures are invented for illustration.
| Date | Description | Amount (₹) | Type |
|---|---|---|---|
| 01 Nov | Salary credit | 60,000 | Credit |
| 02 Nov | Rent transfer | 18,000 | Debit |
| 05 Nov | Grocery store | 7,200 | Debit |
| 09 Nov | Food delivery | 2,600 | Debit |
| 12 Nov | Electricity bill | 1,900 | Debit |
| 15 Nov | Fuel | 3,000 | Debit |
Your first prompt could be: Group these debits into categories, show a total per category, and list the rows you were unsure about. A good answer gives rent 18,000, groceries 7,200, food delivery 2,600, bills 1,900 and fuel 3,000. Now add them yourself: 18,000 + 7,200 + 2,600 + 1,900 + 3,000 = 32,700. Credits are 60,000, so the net for these rows is 27,300 in the positive.
If the chatbot says total debits are 33,700, you have caught a slip worth a thousand rupees. Ask it to list every row it counted with a running sum and the difference shows up quickly.
Which follow-up questions give useful answers?
- Which debits repeat every month, and on which dates?
- Which merchants appear more than twice, and what did they add up to?
- Are there two debits of the same amount to the same name within two days?
- What share of total debits went to each category, rounded to the nearest rupee?
- Which rows would you need me to clarify before you categorise them?
Ask for a table in each reply so you can paste it into a sheet. Ask for the working, not just the answer.
What should I check before trusting the result?
Remove anything you would not hand to a stranger: account numbers, your full name, addresses and balances. Check the privacy settings of your chatbot account, and remember that rules differ between free and paid plans, so read what yours says.
When a chat stops being enough
A chat is fine for one statement. It gets tiring once you repeat it monthly, across two banks and a card, with the same corrections each time. That is where an expense tracker earns its place. SpendixAI takes bank and card statements as PDF, CSV or Excel, shows you the rows to review before saving, and remembers your corrections as history, so March does not start from zero. You can also connect ChatGPT to your SpendixAI workspace through MCP, choose a scope, start read-only and disconnect whenever you like, which means you can still ask questions in chat without pasting statements.
A short checklist to finish
- Statement exported as CSV, personal details removed.
- Row count and printed totals written down first.
- One question per message, table in every answer.
- Totals recomputed by you for at least the biggest category.
- Corrections saved somewhere permanent, not only in the chat.
Useful ideas start with a clearer picture of your own spending.
Explore your reports
