The useful result
A one-month budget summary by category, where every total is backed by listed rows and three totals are checked by hand.
- Remove private details from a bank export before it goes near an AI chat.
- Get category totals plus the exact rows behind each figure.
- Recheck three totals by hand and say whether the AI's answer can be trusted.
You ask an AI where your money went, and it hands you tidy totals. Nobody can tell whether they are right. This lesson fixes that, and it works with any AI chat and any bank that lets you download a CSV.
You will remove the private parts (redact) of an exported file, ask for category totals with the rows behind each figure, then add three totals up yourself. If they match, you can trust the method for that file.
We have not run this on a live account, and the examples are illustrations. This is a budgeting exercise, not financial advice. It needs no bank link, so it works anywhere.
Before you start
- A CSV export of one to three months from your own bank, and a spreadsheet app to open it.
- Any AI chat assistant. About 25 minutes.
Review evidence
Documentation-based review · 2026-11-09
Sources opened and checked on 9 November 2026. Figures are dated and describe what the sources reported, not rules. Prompts and examples are illustrative; no model output or outcome is claimed.
Redact the file so only date, description and amount remain
Open your export in a spreadsheet. First paste only the header row into an AI chat (no data) and ask which columns to keep. Delete the others. Then add a Row column numbered 1, 2, 3. Use one month first. Before you paste, read the Description column line by line. Replace or delete anything that names a person, your employer, your landlord, a doctor, lawyer or clinic, or holds a phone number, an address, a card or account number, or a long reference code. If you would not say it to a stranger, change it to a short label such as EMPLOYER or RENT. Delete any row you cannot make safe. A file you paste into an AI chat is shared with that service, so check its privacy settings before you paste.
Below is ONLY the header row of a bank export. It has no data. For each column name, answer KEEP or REMOVE. Keep only what a spending-by-category review needs, which is usually date, description and amount. Remove anything that could identify me or is not needed. Then list three things I should search for by hand in the file before I paste any rows. Do not ask me for any rows.
HEADER ROW:
[PASTE HEADER ROW ONLY]Example / expected result
Illustrative example; not a recorded model result.
Header row: Date, Description, Amount, Balance, Account Number, Account Holder, Reference KEEP: Date, Description, Amount REMOVE: Balance, Account Number, Account Holder, Reference Search by hand for: your full name, your street or phone number, and the last four digits of any account.
Check your result
Only Date, Description, Amount and Row remain. Searching the sheet for your name, address and account digits finds nothing. No description line contains a name, address, phone number or long number.
Ask for category totals with the rows behind each figure
Paste the redacted rows with their Row numbers. Choose five or six categories yourself. Tell the AI that negative amounts are spending and positive amounts are income, which stays out of the totals. A total with no rows behind it cannot be checked, so always ask for both.
This is a redacted bank export with a Row number. Use ONLY these rows. Negative amounts are spending. Positive amounts are income: leave them out of the totals. Put every spending row into exactly one of these categories: [YOUR CATEGORIES]. If you are unsure about a row, put it in Unsure and give the reason in five words. Output 1) a table with Category, Total, Row count and Row numbers, then 2) for each category, every row as: Row | date | description | amount. Do not round. Do not estimate. Do not give advice.
ROWS:
[PASTE REDACTED ROWS]Example / expected result
Illustrative example; not a recorded model result.
Rows (imagined): 1 GROCERY MART -42.10; 2 CITY TRANSIT -2.75; 3 STREAMING CO -9.99; 4 GROCERY MART -35.60; 5 CITY TRANSIT -2.75; 6 PIZZA PLACE -18.40; 7 GROCERY MART -51.30; 8 PAYROLL +1,200.00 Groceries | 129.00 | 3 rows | 1, 4, 7 Transport | 5.50 | 2 rows | 2, 5 Subscriptions | 9.99 | 1 row | 3 Eating out | 18.40 | 1 row | 6 Unsure | 0.00 | 0 rows Row 8 left out as income.
Check your result
Every spending row appears in exactly one category, income rows are left out, and each total comes with listed rows.
Recheck three totals by hand
Pick the biggest category, the smallest, and one that surprised you. Add each one yourself with a calculator or a spreadsheet sum, using the rows the AI listed. If your number differs, give both numbers to the AI and ask it to list the rows with a running total. Then also add all your category totals and compare with the sum of every spending row.
My hand total for [CATEGORY] is [MY TOTAL]. Your total was [AI TOTAL]. List every row you counted in that category, one per line, with its amount and a running total. Tell me if any row is counted twice, left out, or in the wrong category. Do not change my category rules.Example / expected result
Illustrative example; not a recorded model result.
Imagined case: the AI first said Groceries 164.60, my hand total is 129.00. Row 1 | -42.10 | running 42.10 Row 4 | -35.60 | running 77.70 Row 7 | -51.30 | running 129.00 Row 4 was counted twice in my first answer. The correct total is 129.00. Check: 129.00 + 5.50 + 9.99 + 18.40 = 162.89, the sum of all spending rows.
Check your result
All three of your hand totals match the AI's to the cent, and the category totals add up to the sum of all spending rows.
When it goes wrong
If the AI gives totals without rows, say: "Show every row behind each total, with its Row number." If it says it estimated, say: "Use only the rows I pasted. Do not estimate." If a row sits in two categories, say: "List any Row number that appears more than once." If the file is long, paste one month at a time. If you pasted an account number by mistake, tell your bank. Deleting the chat may not remove it from the service's records. Matching totals show the sums are right, not that the categories make sense.
Make it your own
Redact one month of your own export and run all three steps. Score yourself: how many of your three hand totals matched to the cent? Three means the sums are right for that file. It does not mean the categories make sense. Fewer means fix the rows first, then try again.
Useful next resources
Sources and review
Last checked Nov 9, 2026. Features and account limits can change. Verify important results against your original material.
Did this help you complete your task?
Tell us what needs improving. No account or private output required.
Report incorrect information or a broken linkGenius Forges Editorial
Practical exercises based on checked sources. Examples are illustrations, not recorded model results.
