Categorise a month of bank transactions with one AI prompt
Sorting a bank export line by line is slow, and it is easy to forget what a payment was for. An AI assistant can do the first pass in seconds. You then spend your time only on the rows that need a human decision.
What you will be able to do: paste a bank-export CSV into an AI assistant and get back a table of transactions, each with a category, a confidence level and a "check me" flag, plus totals per category.
Why AI transaction categorization works (and where it does not)
AI models are good at reading messy bank descriptions such as ADOBE *CREATIVE CLOUD and recognising what they probably are. They cannot know your intentions. A coffee could be a client meeting or your weekend. A purchase from an online store could be a printer cartridge or a birthday present. Only you know.
So this prompt does two jobs. It categorises what is obvious, and it flags what is not so you decide those yourself. The aim is to save time on routine rows, not to hand your bookkeeping to a model.
Before you start: protect your data
- Export one month of transactions from your bank as CSV (most online banking has an "Export" or "Download" option).
- Delete columns you do not need. Keep only Date, Description and Amount. Remove account numbers, card numbers, balances and anything that names other people.
- Check your AI tool's data settings, including whether your chats may be used for training, and use a business or private workspace if you have one.
- If your work has confidentiality rules (client contracts, regulated industries), check them first.
The prompt
Copy this, edit the category list to match how you actually report your business, then paste your CSV rows at the end.
You are helping me categorise my business transactions. Use ONLY these categories: - Income – Client work - Income – Product sales - Software & subscriptions - Travel & transport - Meals - Office supplies & equipment - Transfer (not income or expense) - Personal / not business - Uncategorised For each row of the CSV below, return a table with these columns: Date | Description | Amount | Category | Confidence (High/Medium/Low) | Check me? (Yes/No) | Reason (max 8 words) Rules: 1. Copy Date, Description and Amount exactly. Never change amounts, never add or drop rows. 2. If you are not sure, use "Uncategorised", set Confidence to Low and Check me to Yes. 3. Set Check me to Yes for anything that could be either personal or business (for example meals, online shopping, fuel, ride-hailing) and for every transfer. 4. Do not guess what a purchase was. Say what information is missing instead. After the table, give: - the total amount per category - the number of rows marked "Check me" - a one-line reminder that I must confirm the flagged rows myself. CSV: [paste your Date,Description,Amount rows here]
Worked example
We ran the prompt above with Claude on 17 October 2026, using the 10 made-up transactions below. The data is illustrative, not from a real account. Your results will differ by tool, model and data.
Input (illustrative CSV)
Date,Description,Amount 2026-09-01,STRIPE PAYOUT,1250.00 2026-09-02,ADOBE *CREATIVE CLOUD,-59.99 2026-09-03,GRAB *TRIP 8842,-14.20 2026-09-05,NOTION LABS,-12.00 2026-09-08,WISE TRANSFER CLIENT ACME,2400.00 2026-09-09,STARBUCKS 0231,-6.80 2026-09-12,AMZN MKTP US,-89.45 2026-09-15,GOOGLE *WORKSPACE,-8.40 2026-09-18,TRANSFER TO SAVINGS,-500.00 2026-09-21,SHELL 1123,-62.30
Output
AI output for the sample data. Highlighted rows were flagged "Check me". On a phone, scroll the table sideways.
| Date | Description | Amount | Category | Confidence | Check me? | Reason |
|---|---|---|---|---|---|---|
| 2026-09-01 | STRIPE PAYOUT | 1250.00 | Income – Product sales | Medium | Yes | Payout could be clients or products |
| 2026-09-02 | ADOBE *CREATIVE CLOUD | -59.99 | Software & subscriptions | High | No | Design software subscription |
| 2026-09-03 | GRAB *TRIP 8842 | -14.20 | Travel & transport | Medium | Yes | Ride could be personal |
| 2026-09-05 | NOTION LABS | -12.00 | Software & subscriptions | High | No | Productivity app subscription |
| 2026-09-08 | WISE TRANSFER CLIENT ACME | 2400.00 | Income – Client work | High | No | Description names a client |
| 2026-09-09 | STARBUCKS 0231 | -6.80 | Meals | Low | Yes | Business purpose unknown |
| 2026-09-12 | AMZN MKTP US | -89.45 | Uncategorised | Low | Yes | Item unknown; check the receipt |
| 2026-09-15 | GOOGLE *WORKSPACE | -8.40 | Software & subscriptions | High | No | Email and office suite |
| 2026-09-18 | TRANSFER TO SAVINGS | -500.00 | Transfer (not income or expense) | High | Yes | Own-account move, not spending |
| 2026-09-21 | SHELL 1123 | -62.30 | Travel & transport | Medium | Yes | Fuel may be personal use |
Totals per category (we re-added these by hand: they match)
| Category | Total |
|---|---|
| Income – Client work | 2400.00 |
| Income – Product sales | 1250.00 |
| Software & subscriptions | -80.39 |
| Travel & transport | -76.50 |
| Meals | -6.80 |
| Uncategorised | -89.45 |
| Transfer (not income or expense) | -500.00 |
Rows marked "Check me": 6 of 10. That is a lot for a small sample, and it is deliberate: the prompt is cautious. On a typical month, most subscriptions and clearly named client payments will come back as High confidence and need only a glance.
Your 10-minute review
- Check the row count. The output should have exactly as many rows as your CSV. If not, ask: "You returned N rows but my CSV has M. List the missing rows."
- Spot-check three amounts against your bank statement.
- Decide every flagged row yourself. Look up receipts for "Uncategorised" items.
- Re-add one total (a calculator is fine). AI tools can make arithmetic mistakes.
- Save the final categories in your own record: a spreadsheet, your accounting tool or a Notion tracker.
Not tax or accounting advice. Which expenses are deductible, and how to treat mixed personal and business costs, depends on where you live and your situation. Use your tax authority's guidance or an accountant for those decisions.
Useful variations
- Your own rules: add lines such as "Anything from GRAB on weekdays between 9am and 6pm is Travel & transport, Medium confidence."
- Spreadsheet-ready output: add "Return the table as CSV in a code block" so you can paste it straight into a spreadsheet.
- Learn from last month: paste last month's reviewed table first and add "Use these as examples of how I categorise."
Next step
Keep your categories in one place
Once your categories are reviewed, they need a home. Our Notion Finance Dashboard includes an Income & Expense Tracker you can filter by category, client or date, alongside budgets, invoices and a tax estimator. The product page shows what is included, the requirements and the price.
See what's inside the Finance DashboardMore lessons
This week's theme is AI for your freelance money admin. Next: write a clear invoice and a polite overdue reminder with AI.
All lessonsHow we checked this lesson
- Prompt run with Claude on 17 Oct 2026 using the illustrative data above. The output table and totals are shown as returned; the totals were re-added by hand and match.
- This lesson relies on no product-specific features. Any general AI chat assistant that accepts pasted text should handle it, but we have only tested it with Claude.
- Last reviewed: 17 October 2026.