ChatGPT budget prompts: how to analyze your spending from a CSV

Step 1: export a clean CSV

AI assistants are much more reliable with a file than with numbers pasted into the chat. Prepare the export first:

Pick full months. Three to six complete months are enough to compare. A half-finished month makes spending look lower than it is.
Keep the columns simple. A machine-readable format with ISO dates and dot decimals is the easiest for AI to read.
Know what’s inside. Merchant names, notes and account names all go into the file. Rename accounts like “Main card” instead of card numbers, and leave notes out if they contain anything private.

In Budget AI, open Export to CSV in settings, choose the period, accounts and categories, and pick the Data format. Turn off Export note if you don’t want to share your notes. You can also tap Share in Analytics to export exactly what you’re looking at, with the filters you already set. The file has these columns: date, type, amount, currency, amount_default (the amount in your main currency), category, subcategory, account and note, plus transfer details.

Step 2: check what you share

Uploading a file to a chatbot means sending your spending history to that service. Before you do:

Check the data settings. ChatGPT lets you turn off “Improve the model for everyone” in Data Controls and offers temporary chats. Claude lets you change your model improvement settings, and its incognito chats aren’t used to improve Claude.
Remove what you don’t need. Full card numbers, names of other people and personal notes don’t help the analysis.
Use a work account carefully. If your chatbot account belongs to your employer, your personal finances may be visible to its administrators.

Step 3: prompts that work

Attach the file and start with a setup prompt. It tells the AI how to read the columns, so later answers don’t mix up income, spending and transfers.

Setup

The attached CSV is my transaction history from a budgeting app. Columns: date, type (income, expense or transfer), amount, currency, amount_default (the amount in my main currency), category, subcategory, account, note. Expense amounts are negative. Ignore transfers. Use amount_default for totals. First, show the date range, the number of transactions and total income and expenses by month in a table. Don’t give advice yet.

Compare months

Compare my spending by month. Show the five categories where the last full month differs most from the average of the previous months, with amounts and the change in percent.

Find recurring payments

Find payments that repeat every month or every week with a similar amount. For each, show the category or note, the typical amount, the dates and the estimated cost per year. Mark the ones that look like subscriptions.

Check the 50/30/20 split

Classify each category as a need, a want or savings and debt payments. Show my actual split by month next to the 50/30/20 rule, and list the categories you weren’t sure how to classify.

Spot small leaks and outliers

Group expenses under 10 in my main currency by category and show how much they add up to each month. Then list unusual transactions: amounts much larger than usual for that category, or two expenses with the same amount on the same day.

Plan the next month

Based on the last three months, suggest monthly limits for my flexible categories that would free up 200 in my main currency. Keep fixed costs such as rent and loan payments as they are, and explain each change in one line.

Save for a goal

I want to save 3,000 by June. Using my average monthly income and spending, how much would I need to set aside each month, and which two categories would be the easiest to cut to get there?

Replace the amounts and dates with your own. If you track several currencies, add: “Show totals in each currency separately, then in my main currency using amount_default.”

Step 4: check the answer

AI assistants can misread a column or miscount, and they say it with the same confidence as a correct answer. A quick check saves you from decisions based on a wrong number:

Ask for the working. “Show the calculation as a table with monthly totals” makes errors visible.
Compare one total. Check one month’s total spending against the same period in your budgeting app.
Watch for usual mistakes: transfers counted as spending, currencies added together, negative and positive amounts mixed up, a missing month.
Ask where a number came from. “Which rows did you use for this?” is a good follow-up when something looks off.

What is AI good at here — and what isn’t it?

AI is good at grouping transactions, noticing patterns, comparing periods and drafting a first plan. It doesn’t know your context: that a big purchase was a one-off gift, that rent rises next month or what you’re saving for — tell it. And it isn’t a financial adviser: treat its suggestions as ideas to check, not instructions.

How do you make it a monthly routine?

On the first day of the month, export the previous three months.
Run the same setup and comparison prompts. Saved prompts make months comparable.
Pick one change for the new month and set it as a limit in your budget.

The analysis is only as good as the data. Fewer missed purchases means better answers, which is where automatic tracking helps.

Frequently asked questions

Which AI assistant works best?

Any assistant that accepts file uploads, such as ChatGPT or Claude. Results vary between models and even between runs, so check the totals either way.

Can I paste transactions instead of uploading a file?

For a few dozen rows, yes. For months of data, a file is more reliable: long pasted text is easier to cut off or misread.

Is it safe to upload my spending history?

It depends on the service and its settings. Remove what you don’t need, check whether chats are used to improve the model and consider a temporary or incognito chat.

Can I do the same in a spreadsheet?

Yes. The CSV opens in Excel, Numbers or Google Sheets, and a pivot table gives totals by month and category. AI is faster for questions you haven’t planned in advance.

Sources

This article is for general information and isn’t financial advice.