Use ChatGPT to Categorize Expenses and Tidy Your Books as a Freelancer

Export your bank or card transactions as a CSV, paste the descriptions into ChatGPT or Claude with your own list of categories, and it labels every row in one pass. Then ask it to flag anything odd and total your spend by category. Review the results before you trust them, never paste full card numbers, and treat the output as bookkeeping prep, not tax advice.

Month-end bookkeeping for a freelancer is mostly one dull job: opening a statement and deciding which bucket each line belongs in. Forty transactions, forty small decisions, cryptic merchant names you half-recognize. ChatGPT and Claude are good at exactly this. You hand over the list of descriptions and your own categories, and the AI labels every row in seconds while you just review. The point is not to replace your judgment. It is to do the boring sorting so your attention goes only to the few charges that genuinely need a human.

Before you open any AI, decide your buckets. A fixed list is the single thing that separates tidy books from forty slightly different labels you clean up later. Keep it short and concrete, matched to how you actually work:

• Software, Travel, Meals, Office supplies
• Marketing, Subcontractors, Bank fees, Other

Now every transaction has exactly one home. When the AI is unsure where something goes, it drops it in "Other" and flags it instead of guessing a brand-new label. Reuse the same list every month and your categories stay consistent, which is what makes the year-end total actually mean something.

Log into your bank or card account and download the month as a CSV. Almost every bank has this under statements or activity, usually as "Export" or "Download." Open the file in Google Sheets or Excel and keep just three columns: Date, Description, Amount. Then delete anything sensitive. You do not need the full card number to categorize, so removing that column means it never leaves your bank. Add one empty column called Category for the results. Clean input here makes every step after it faster and safer.

Copy the whole Description column and paste it into ChatGPT or Claude with one clear instruction. The key phrase is "same order," so the answers line up when you paste them back into your sheet.

Prompt: Categorize each transaction into exactly one of: Software, Travel, Meals, Office, Marketing, Subcontractors, Bank fees, Other. Return only the category, one per line, in the same order. If unsure, use "Other (check)".

Paste the results into your Category column. Dozens of rows, sorted in the time it takes to read this sentence. The AI already knows that "ADOBE" is software and "UBER" is travel, so most rows land correctly on the first pass.

This is the step that keeps it honest. Filter your sheet for anything marked "Other (check)" and scan those rows. They are the genuinely ambiguous merchants, a name that could be a client lunch or a personal one. Fix just those by hand. While you are at it, ask the AI to help spot trouble.

Prompt: From this list, flag any duplicate charges and anything unusually large for its category.

That catches a double charge, a subscription you forgot, or a payment in the wrong bucket. Reviewing ten flagged rows takes two minutes versus the half hour it takes to label everything from zero.

Now the payoff. Paste your categorized list back in, or keep it in the same chat, and ask for the summary that makes your books readable at a glance.

Prompt: Total the Amount by category for this month, and list my three biggest spending areas.

In seconds you get a clean breakdown you can drop into a tracker or hand to your accountant. Ask a follow-up like "list every recurring charge you can see" and you catch the tools you are quietly paying for but no longer use. This monthly total, kept consistently, is the difference between guessing your numbers and knowing them.

Two habits make this low-risk. First, never paste full card or account numbers. You stripped that column in Step 1, so there is nothing sensitive in the merchant name and amount. Second, mind whose data it is. Personal spending in a personal account is fine. If any of these transactions belong to a client or a registered company, follow their rules on which tools are approved and use a business plan where your input is not used to train the model. Low-stakes data, sensible account, and you keep the convenience without the exposure.

Be clear about what you are getting. The AI sorts and totals your spending so your records are clean and your accountant has tidy numbers to work from. It does not know your country's rules, your tax situation, or what counts as deductible, and it cannot check current law. So use this to prepare, not to decide. Filing your taxes, or judging whether something is a business expense, is a question for a real tax professional looking at clean books, which is exactly what this gives them.

Open last month's statement and export it as a CSV. Trim it to Date, Description, Amount, delete anything sensitive, and run the categorize prompt with your own bucket list. Review only the flagged rows, then ask for the monthly total. One real month is the fastest way to see how much bookkeeping this quietly removes, and to fine-tune your categories before the next statement lands.

Export last month's transactions to CSV now — just the one month, not the year — and run the categorisation prompt on it. Check every 'other'. If the result survives your inspection, do the backlog next weekend.