Can ChatGPT Be Your Financial Advisor? What the Tests Show and Where It Breaks
In one Vanguard survey, 37% of investors who used AI for money questions said it gave them wrong or misleading information, and another 44% couldn't tell. Here's what chatbots handle well, where the published tests caught them slipping, and how to ask so you get a better answer.
Danielle Harrison, a financial planner in Columbia, Missouri, asked an AI model a question she already knew a fair amount about. Her husband was joining her firm, and she wanted to know how they should structure the business. The chatbot was sure of itself: S corporation. She kept going, adding details about their situation. By the end, as she told NPR, the same chatbot had reversed itself and was telling her to form an LLC. Without her own background, she said, she'd have left with the wrong information.
Most people asking a chatbot about money don't have her background, which is the reason they're asking. When Vanguard surveyed 6,686 of its own investors in June 2026, 37% of those who had used AI for financial questions said it had given them incorrect or misleading information. Another 44% said they couldn't tell. Only 19% were confident what they got was sound.
That isn't a reason to stop using these tools. The same research shows they get a lot right, and they cost nothing at two in the morning. It's a reason to know which questions they're good at.
How Many People Are Actually Doing This
It depends on who's asking and how. A Gallup survey conducted with Edward Jones in spring 2026 found that about 1 in 5 Americans who sought financial advice in the past year turned to AI. J.D. Power asked 4,000 people and found 40% had used AI in the prior three months to help manage their finances. TD Bank's survey of about 2,500 consumers put it at 55%, up from 10% a year earlier.
Those numbers aren't measuring the same thing. Gallup asked where people went for financial advice. TD asked whether people use AI to aid their financial management decisions, which is a much wider net. What the surveys agree on is trust. In Gallup's poll, only about 3 in 10 adults had at least some confidence in AI's expertise with money, and 3% had a great deal. In TD's, 18% said they'd trust AI to make financial recommendations on its own.
So people are using it and not quite believing it. That's a reasonable place to be.
What Chatbots Get Right
The broad strokes, mostly.
Taha Choukhmane, an associate professor at MIT's Sloan School of Management, co-authored a study, not yet published, in which 1,000 adults wrote their own prompts asking an AI model for financial advice. The researchers then simulated what would have happened over a lifetime if people followed it. As NPR reported, following the advice would have left people with more savings. The model nudged them toward saving more, investing in the stock market, and taking less risk as they aged.
LendingTree ran a smaller test in August 2025, putting 20 personal finance prompts to ChatGPT and comparing ten of them across ChatGPT, Claude, and Gemini. The answers were similar across the three and mostly correct. On a question about whether to save a down payment or invest it, ChatGPT laid out the tradeoffs and steered away from the riskiest choice. The loan and interest calculations matched LendingTree's own calculators, which surprised the researchers.
Chatbots are also good at plain explanation. What's the difference between a mutual fund and an index fund? How does a Roth IRA differ from a traditional one? What does "vesting" mean in the offer letter you just got? Choukhmane's advice to the AP was to use AI for explaining and defining terms. For that job it's patient and fast, and it won't make you feel dumb for asking.
Where It Breaks
The failures in the published tests fall into a few groups.
It fills in what you didn't say. LendingTree asked about retirement planning for a 50-year-old couple with $100,000 saved. ChatGPT told them to save more than $30,000 a year to catch up, without asking what they earn or why they were behind. LendingTree's researchers pointed out that a proper answer needs the couple's income, what they're saving now, and what kinds of accounts they have. The chatbot skipped all of it and answered anyway.
Its numbers can be a year old. In that same test, ChatGPT gave the 401(k) contribution limit as $23,000 and labeled it the 2025 rule. The 2025 limit was $23,500. For 2026 it's $24,500, with an $8,000 catch-up for people 50 and over. Tax thresholds change every year, and a model's training doesn't always keep up.
It struggles with messy situations. In the MIT study, the model's advice on a job loss called for spending cuts the researchers considered too harsh, and it didn't suggest using savings that exist for exactly that purpose. They weren't impressed with its guidance on rebalancing a portfolio either, and they found it suggested riskier moves for men than for women.
It agrees with you. NPR profiled an IT manager in Ohio who uses ChatGPT constantly for money questions. When it starts praising every idea he has, he tells it to stop. LendingTree's researchers described the models as quick to agree and quick to assume.
It can invent things. A joint notice from the SEC, state securities regulators, and FINRA warns that AI output can be faulty or completely made up even when the information it started with was accurate, and it tells investors not to rely on it alone for investment decisions.
NPR's summary of what experts told them is useful here. AI advice works best at the two ends: basic questions, and detailed prompts from people willing to hand over a lot of information. The trouble is in the middle, where most real questions live.
How to Ask So the Answer Is Better
A few habits close much of that gap.
- Give it the whole picture. Ages, income, filing status, state, and balances broken out by account type (401(k), Roth, taxable, cash). "Should I do a Roth conversion?" gets a generic answer. The same question with your numbers gets a much better one.
- Ask what it assumed. End with "list every assumption you made about my situation." You'll often spot the wrong one right away.
- Ask what it's missing. "What would you need to know to answer this properly?" turns the chatbot into the one asking questions, which is what the LendingTree test showed it doesn't do by default.
- Tell it to push back. LendingTree suggests instructing it to act as a skeptical expert who asks questions and says so when it doesn't know.
- Ask for sources, then open them. Models sometimes cite pages that don't exist.
- Change one detail and ask again. If the answer flips the way it did for Harrison, you've learned the first answer wasn't solid.
What to Check Yourself, and What to Keep Out of the Chat
Verify anything with a dollar limit or a date attached. Contribution limits, income phaseouts, and tax brackets should be checked against IRS.gov for the current year. The same goes for deadlines, and for anything that depends on your particular employer plan, since a chatbot hasn't read your plan documents unless you gave them to it.
As for what to share: the model needs your numbers, not your identity. Leave out account numbers, your Social Security number, logins, and full statements with identifying details. Round figures work fine. LendingTree's guidance is not to share anything sensitive, like banking information. The Ohio man in NPR's story doesn't give the chatbot direct access to his accounts.
When to Put It in Front of a Person
Accuracy aside, a chatbot and a planner differ in who answers for the advice. As Bobbi Rebell, a CFP® professional, put it to the AP, "There's no AI that is a fiduciary." The certification's standards of conduct require that a CFP® professional act as a fiduciary at all times when providing financial advice, which means in the client's best interest. A chatbot has no such duty and no license to lose. If its answer costs you money, that's yours to absorb.
That matters most for decisions that are large, hard to undo, or tied to a tax year: a Roth conversion, exercising stock options, handling an inheritance, picking a retirement date. Our guide to whether you need an advisor at all covers the wider list.
It doesn't have to be one or the other. Vanguard found its investors were about three times as likely to use AI to prepare for or review a meeting with a financial professional as to replace one. That's a sensible split. Do the learning with the chatbot, then bring a sharper list of questions to a person who can see your whole situation and is accountable for the answer.
One more finding from that survey deserves a mention. About four in ten AI users said chatbots beat human advisors at avoiding conflicts of interest. That worry comes from how some advisors are paid, and it's worth understanding the different fee models before you hire anyone. An advice-only planner charges a flat or hourly fee that you pay directly and earns no commissions. They give specific recommendations, including on investments, and you keep the accounts at your own brokerage and place the trades yourself.
If a one-time check is what you're after, hourly planning fits. Based on data from 97 advice-only planners on the Advice-Only Network (March 2026), hourly work runs a median of $300 per hour, a one-time comprehensive plan runs a median of $3,000, and monthly ongoing relationships run a median of $250 per month.
The man in NPR's story, the one who uses ChatGPT for money questions so often he's given it a nickname, still sits down with a human adviser once a year.
Ready to find an advice-only financial advisor?
Every advisor in our network charges transparent fees with zero product sales or AUM charges.
Browse Advisors