What you’ll be able to do

Right, so the goal here is simple. You take a statement export from your bank, the CSV or PDF you can download in about two clicks, and you get an AI to turn it into something useful. Categories. Monthly totals. That subscription you forgot you were paying for since March. All without giving anyone your online banking login, because you should never do that, ever.

Before you start

You need three things. Your bank export sitting somewhere on your computer, five minutes, and a decision about what you’re comfortable letting an AI see.

That last one matters more than people think. Your statement has your name on it, maybe your account number, maybe a sort code, and definitely a list of exactly where you spend money and when. None of that needs to go anywhere near a chat window. You’re going to strip it out before you upload anything, and I’ll show you how in a second.

You don’t need any AI experience for this. You don’t need to know what a “prompt” is beyond typing a sentence and pressing enter. If you can use online banking, you can do this.

Get set up

First, get the file. Every UK bank, and I mean every one, lets you export statements as a CSV or a PDF from the app or the website. Look for something like “statements” or “download transactions” or “export”. Grab three months’ worth if you can. One month is too small to spot patterns, six months starts getting messy for a first go.

Next, open the file. If it’s a CSV, Excel or Google Sheets or even Numbers will open it fine. If it’s a PDF, just open it as normal.

Now the important bit: delete anything you don’t want an AI to see. That’s usually your account number, your sort code, and your name if it appears anywhere in the file. If you’re using a CSV this is easy, just delete the column. If it’s a PDF, you might need to black it out or retype the transaction list into a spreadsheet without the personal details. A bit fiddly, but worth it.

Then pick your tool. ChatGPT can analyse an uploaded CSV directly, it actually runs code behind the scenes and can give you charts back, which is genuinely useful for this. Claude and Gemini can also read an uploaded CSV and work through it. If you’re already using Excel or Google Sheets, Copilot in Excel and Gemini in Sheets can do similar analysis on eligible plans, working right there in your spreadsheet rather than a separate chat.

If you’d rather not do any of this manually at all, open banking apps like Emma, Snoop, or Monzo’s own budgeting tools connect to your account with your consent and categorise spending continuously, no exporting or uploading required. Different approach, same goal. Worth mentioning honestly: if you’re not fussed about doing this yourself every few months, one of those might suit you better than the DIY route below. But if you want to understand your spending without linking another app to your bank account, keep going.

Try it yourself

Upload your cleaned file to whichever tool you’ve chosen, then start with this:

I've uploaded my bank transactions for the last three months. Please
group them into spending categories (like groceries, transport,
subscriptions, eating out, bills) and give me a total for each
category. Show it as a simple table.

That gets you the shape of where your money’s going. Once you’ve got that, ask for the month-by-month view:

Now break these totals down by month instead of by category. I want
to see how much I spent in total each month, and whether any
category jumped noticeably from one month to the next.

This is the one that catches things. A category that quietly doubled in month two, say. Or a month where eating out went through the roof and you can actually remember why (that was the month your mate visited, wasn’t it).

Then the one that does the real work:

Looking at these transactions, can you identify any recurring
payments or subscriptions? List them with how often they charge and
roughly how much per year each one costs me. Then tell me honestly:
if I had to cut one thing from this list first, what would you
suggest and why?

That last question, the “what would I cut first”, is doing real work. It forces the AI to actually reason about your spending rather than just listing it back at you.

Once you’ve got answers you trust, set one rule for yourself. Just one. Something like “no more than four takeaways a month” or “cancel anything I haven’t used in 60 days”. Don’t try to overhaul your entire financial life off the back of one chat. One rule, stick to it, revisit in a few months.

Check the result

Don’t just take the output and run with it. Check it.

Start with the categories. Do they match what you’d guess from memory? If you know you spend a lot on coffee and the AI has lumped it all under “miscellaneous”, that’s a sign it needs a nudge, so tell it: “please put anything from cafes into its own coffee category and redo the totals.”

Check the maths adds up, roughly. Add up two or three categories yourself, quickly, on a calculator, and see if it’s in the right ballpark. AI tools that run actual code on your CSV (which ChatGPT does) tend to be accurate with the numbers themselves, the bit that goes wrong is usually the categorising, not the arithmetic.

Look for anything obviously miscategorised. A big one-off purchase, like a holiday or new laptop, sometimes gets folded into a regular spending category and skews the monthly total. Ask the AI to flag anything unusually large as a one-off, separate from your normal monthly spend.

And the subscription list: actually check each one. Do you recognise it? Is it still something you use? This is the single most useful check in the whole process, because it’s the one most likely to save you actual money.

If it doesn’t work

Sometimes the AI gets confused, especially with PDF statements where the formatting is messy or the transaction descriptions are cryptic bank shorthand. If that happens, try converting to CSV first if you can, it’s a much cleaner format for this kind of work. Or ask directly: “some of these transaction descriptions are unclear, can you group anything you’re unsure about into an ‘unclear, please check’ category rather than guessing.”

If the categories come back completely wrong or nonsensical, it’s usually a sign the file didn’t upload properly, or there’s a formatting issue in the CSV. Re-export and try again before troubleshooting further.

Now the hard rule, and this is the one that actually matters: never give an AI your banking login. No AI assistant needs it, no legitimate tool will ask you to type your online banking password into a chat window, and anyone or anything asking you to do that is not something you should trust with your money. This whole guide works entirely from a downloaded file. That’s the point.

Strip identifiers before you upload, every time. Account numbers, sort codes, your full name, your address if it appears. If you’re ever unsure whether something’s sensitive, take it out. You can always add detail back into a follow-up question without it living permanently in a chat log.

And one more thing worth saying plainly: this is spending analysis, not financial advice. An AI can show you patterns in your own data, it can’t tell you whether you’re saving enough for retirement or whether to overpay your mortgage. For anything like that, talk to an actual adviser.

Keep exploring

If you liked pulling structure out of messy data, “How to get AI to make a spreadsheet” is the natural next step, useful well beyond bank statements. If you’re trying to cut spending on food specifically, “How to get AI to make a meal plan” pairs nicely with what you’ve just found in your grocery category. And if the subscription audit above tempted you into a bigger purchase decision, “How to get AI to compare products before I buy” will help you think it through properly before you spend the money.

Sources and review notes

Review date: 15 September 2026 — every source above was opened and checked against the text on that date.