What AI is good at with your money

There are a lot of posts floating around the internet titled something like "50 AI Prompts for Personal Finance." This is the other version: fewer prompts, and more about how to use them, including the parts where the tool will let you down. That last bit matters more with money than it does elsewhere, because a confidently wrong answer has a dollar cost attached to it.

Full transparency: you're reading this on the blog of a company that builds a banking app.¹ We'd rather you got good at asking questions about your own money than trusted any one tool to answer them for you, ours included.

What AI is for and what it isn't

It is not a financial advisor. It doesn't know your accounts or your balances, and it owes you no fiduciary duty, so anyone selling it as a replacement for professional advice on a complicated situation is selling something. (A fiduciary duty is a legal and ethical obligation to act solely in the best interests of another person or entity rather than for personal gain.)

What it is: an explainer with infinite patience that never makes you feel stupid for asking.

That's a narrower job, but it's also the exact thing missing from most people's financial lives.

The reason a lot of adults don't understand their own mortgage isn't that the information is unavailable. It's that asking can feel embarrassing, the person explaining it has a stake in the outcome, and there's never a good moment to say "wait, go back, what's escrow?"

You can ask an AI tool the same question four times in a row, in increasingly basic terms, at 11pm, and that willingness to be asked again is most of what it's good for here.

What separates a useful prompt from a useless one

Your numbers.

"Help me make a budget" gets you a generic explanation of the 50/30/20 rule. "Should I pay off debt or invest?" will generate a both-sides summary that doesn't resolve the situation you may be facing. These aren't bad prompts because of phrasing. They're bad because there's nothing in them to reason about.

Compare these two versions of the same question.

Weak

Should I pay off my car loan or invest the money?

Strong

I have $8,400 remaining on a car loan at 8.9% APR with 31 months left. I also have $6,000 in a savings account earning 3.8% APY. My emergency fund is separate and covers about three months of expenses. Walk me through the math of paying the loan off early versus leaving the money invested, show your calculations, and then tell me what you'd need to know about my situation that I haven't told you.

The second prompt can actually be answered.

And the last clause can make the biggest difference: "then tell me what you'd need to know about my situation that I haven't told you." This is a line you can add to almost any financial prompt.

It turns the tool from something that answers into something that asks. The questions it comes back with are usually the ones you should have been asking yourself.

One more habit: ask it to show its work. These models slip on arithmetic, and you can't check math you can't see.

Decoding documents: the best use and the least discussed

Financial documents are written to be legally complete, not easily comprehensible. You're usually reading them under time pressure, in a room, with someone waiting.

Try one of these prompts the next time you're face to face with money jargon you don't understand.

I'm pasting the Loan Estimate from a mortgage lender below. Go through it section by section and explain in plain language what each line item is, which fees are set by the lender versus by third parties, and which ones are typically negotiable. Flag anything that looks unusual in size.

Here are the terms from a credit card offer. Tell me the purchase APR and whether it's variable, what triggers the penalty APR, and any fee I'd be likely to trigger without realizing it.

This is the fund lineup available in my 401(k), with expense ratios. Explain what each fund actually holds, sort them by cost, and explain what the difference between the cheapest and most expensive option would mean over 25 years on a growing balance.

All of these give you a foundation to learn from, so that when you talk with a financial advisor or expert, you know what's going on with your money.

Building a budget around your actual life

A lot of budgets fail because they're built for an idealized version of the person who will follow them, so describe the real one instead. Here are a few prompt ideas to get you started.

I want to build a budget I'll actually follow. Here's my situation: [take-home pay and how often], [rent or mortgage], [other fixed bills], [debt minimums], [rough variable spending]. Two things about me: I consistently overspend on eating out, and my income varies month to month. Build me something that accounts for both instead of assuming they'll go away.

I've tried budgeting four times and quit within two months each time. Ask me questions to figure out why it keeps failing before suggesting anything.

That second one is the valuable one, because you can't fix a budgeting habit until you know which part of it keeps breaking.

Worth mentioning here, since you're on our blog: this is the problem Crew is built around. A budget that lives next to your bank account loses to the card in your pocket, so we put the budget inside the account. An AI can help you work out what the plan should be, but it can't hold you to it once you're standing at a card reader.

Comparing debt payoff strategies

Give it every balance and every rate. This is a math problem, and math problems are where being specific pays off immediately. Take a look at these prompts.

Here are all my debts: [balance, APR, and minimum payment for each]. I have $650 a month total to put toward them. Model two approaches: highest interest rate first, and smallest balance first. For each, show me the payoff order, the date I'd be debt-free, and the total interest I'd pay. Then tell me the actual dollar difference between them.

Using those same numbers, tell me what happens if I add an extra $100 a month, and what happens if I have to drop to $500 for three months.

The highest-rate-versus-smallest-balance argument usually gets made in the abstract. A dollar figure settles it.

Rehearsing conversations you're dreading

Underused, and more effective than it sounds. These are conversations where you get one shot, you're nervous, and the other party does this professionally.

Act as a finance manager at a car dealership. I have a pre-approval from my credit union at [rate]. Try the standard techniques on me, including payment-focused framing, add-ons, and extending the term, so I can practice recognizing them.

I need to talk to my partner about our spending, and the last two attempts turned into a fight. Help me plan how to open the conversation without it sounding like an accusation. Then role-play as someone who gets defensive about money so I can practice not escalating.

Arguing against you

When you ask an AI about a decision you've already made, it will tend to agree with you, so assign it the opposite side explicitly.

Here's a financial decision I've decided to make, and my reasoning: [decision and why]. Your job is to argue against it as persuasively as you can. What am I not considering? What's the strongest case that I'm wrong?

I'm about to [refinance / buy a house / consolidate debt / change my contribution rate]. List the ways this commonly goes badly for people, and tell me which of those risks apply to my situation specifically.

Translating jargon in the moment

This is the small, constant use that adds up. The trick is asking for the term in the context of your own document rather than in the abstract, so the answer comes back with your numbers in it.

My mortgage statement has a line called "escrow shortage" and my payment went up $180. Explain what happened, whether this is normal, and what my options are.

Explain what "utilization" means, then tell me, given that I have [limits and balances], what mine currently is and what would happen to it if I paid off [specific card].

If you want the full vocabulary in one place, that's a separate post: The Money Words Nobody Explained to You.

The bank account question, done correctly

You'll see a lot of prompts like "find me the best high-yield savings account for my family." Those can be tricky, and the best option is to look into the details and rates yourself.

An AI answering from memory will name banks and quote APYs from whenever its training data ended, which is a problem because deposit rates move constantly. It will sound sure of itself and specific, which is what makes it dangerous. You'd double-check a vague answer, but "Bank X is currently offering 4.35%" reads like a fact.

Live web search doesn't fix it either. Rate roundups are heavily monetized, and the "best" accounts on comparison sites are often paid placements.

So split the job in two. Let the AI build the framework you'll evaluate accounts against, and get the current numbers yourself, from primary sources.

I'm choosing a savings account for my family. Don't recommend specific banks or quote current rates. I'll research those myself. Instead, build me a comparison checklist: every feature and fee I should be evaluating, what a red flag looks like in each category, and the questions people typically forget to ask.

Here are three accounts I'm considering, with details I pulled from each bank's own disclosures: [paste]. Compare them against the checklist and tell me where each one is weak. Don't fill in anything I haven't given you. If a detail is missing, say so.

That second constraint is the one people don't think about. Ask a model to compare incomplete options and it will quietly fill the gaps with numbers that sound right and aren't, so tell it to flag what's missing instead.

For the actual numbers: the FDIC publishes national deposit rate averages, the NCUA does the same for credit unions, and individual banks publish their own rates and fee schedules. Go to those as your primary sources.

Some places where AI will let you down

Worth knowing before you rely on it.

• Current rates and prices. Anything that changes weekly is outside what a model knows unless it's actively looking it up, and even then it's worth checking against the source.
• Arithmetic. Models make multi-step errors that look fine. Ask to see the work, and re-run anything consequential in a spreadsheet.
• Anything requiring your actual accounts. It cannot see your balances, your transaction history, or your credit report. Everything it knows is what you typed, and it won't catch what you left out unless you ask it to.
• Tax specifics. Jurisdictional, revised annually, full of edge cases. Fine for understanding concepts: marginal versus effective rates, deduction versus credit. Not fine for "how much will I owe." Get contribution limits and bracket thresholds from the IRS directly.
• Personalized investment allocation. "What should my portfolio be" isn't an information question. It's an advice question, and the people qualified to answer it owe you legal duties a chatbot doesn't.
• Agreeing with you. These models are trained to be helpful, which in practice means they slide toward telling you your plan is sound. That's why the section on arguing against you exists.

Before you paste anything

Some plain warnings, because this post is telling you to put financial information into a text box.

• Redact identifiers. Account numbers, routing numbers, Social Security numbers, full names and addresses, card numbers. None of these are needed for any prompt above, so replace them with placeholders. The math works the same on "$8,400 at 8.9%" with or without the account number attached.
• Be careful with whole documents. A mortgage statement or brokerage summary usually carries identifiers alongside the numbers you want analyzed. Crop the screenshot or retype the relevant lines.
• Check your tool's data settings. Whether your inputs are used for training varies by product, plan, and setting, and it changes. Look it up for the specific tool you're using rather than assuming.
• Assume nothing is confidential. Not in the way a conversation with a professional is. There's no privilege here.

The habit that makes all of this work

One rule: for any number that will affect a decision, get it from the primary source.

Ask the AI	Get from the source
What this line item means	The fee amount on your Loan Estimate
Which numbers you should be comparing	The current APY, from the bank's own disclosure
Whether your reasoning holds up	The expense ratio, from the fund prospectus
How marginal and effective rates differ	Your bracket and contribution limits, from the IRS
A rehearsal before the phone call	What the rep actually agrees to, in writing

Use the AI to understand what a number means and to check your thinking. Then get the number itself from the bank's disclosure, the lender's written estimate, the prospectus, or the IRS.

Used that way it's a good tool, not because it knows more than you about your money, but because it'll explain the thing four times without sighing.

And once you know what your plan should be, the harder problem is sticking to it. That's the part we built Crew for.

Try Crew

---