Using AI on your own money, and what never to paste into it

Ask a chatbot how to save more money and you get a list. Cook at home. Cancel subscriptions. Build an emergency fund.

None of it is wrong and none of it is any use. You already knew all of it, which is why you asked a machine at eleven at night in the first place.

The problem is not the model. It is that you asked the average question, so you got the average answer.

Context is the whole trick

A model has no idea who you are. Give it nothing and it answers for everybody, which means it answers for nobody.

Give it this instead, once, at the start:

What you earn and how it is structured. Base, bonus, commission, whether any of it is actually reliable.

What you spend, from your statements rather than your memory.

What you have already, and roughly what it is in.

Which country you are taxed in.

What you are actually trying to do, and by when.

Then ask your question. The difference is not subtle. The same model that told you to cancel subscriptions will now tell you that your subscriptions are a rounding error next to the commission structure you have not renegotiated in three years.

The structure that works

Three steps, and the order matters.

1. Build the profile once

Put everything above into one message and ask the model to hold it and repeat it back as a summary. Check the summary. Models mishear numbers.

Save it somewhere you can paste it again. Sessions expire, and rebuilding this from memory every time is how people give up.

2. Ask it to rank, not to advise

The useful question is almost never what should I do. It is which of these is worth the most, in money, over the next year.

Ranking forces arithmetic. Advice invites a list. Ask for the euro impact of each option over one, three and five years and you will usually find the thing you were about to work on is fourth.

3. Make it produce something you can act on tomorrow

A plan with no date is a wish. End the session by asking for the single next action, with a date, and what you will measure to know it worked.

If the answer is not something you could start on Monday, the prompt was still too vague.

Now the part that is missing everywhere else

Everything above asks you to type your financial life into somebody else's computer. So here is the list nobody selling AI prompts prints.

Never paste these

Account numbers, IBANs, sort codes or card details.

Passport, ID card or national insurance numbers.

Logins, passwords or recovery codes.

Full signed contracts, offer letters or payslips as files.

Anything that identifies your colleagues by name alongside their pay.

Not one of those makes an answer better. Every one of them is worth something to somebody else. A model does not need your IBAN to tell you your savings rate.

These are usually fine

Salary and how it is structured. Spending. What you have invested and roughly in what.

Your country, your age, your goal and your timeline.

The numbers off a payslip, typed out, rather than the payslip. Same information to the model, far less of a document sitting in a chat history.

Two habits worth having

Check the data controls of whatever tool you use, and check them again occasionally, because they change. Do not trust an article for this, including this one.

Assume anything you paste could be read by a person one day. Then paste accordingly. That single assumption does more than any setting.

What this is not

A model is not a licensed adviser and neither am I. On tax in particular it will be confident and it will sometimes be wrong, and the confidence does not drop when the accuracy does.

Use it for the arithmetic, the ranking and the rehearsal. Take anything that turns into a decision about your money to somebody qualified where you live.

The eleven prompts in the book work exactly this way. The first builds the profile, the rest run against it, and none of them ask you for an account number. The full contents are published.

Common questions