More than half of Americans who ask chatbots for money advice say they would act on it without checking it first. That is not a guess, it is the headline finding from a PensionBee survey of 1,000 U.S. adults released on September 17, 2026, and it lands right in the middle of a much bigger shift: people are quietly replacing Google searches, financial forums, and even human advisors with a text box. The question worth asking is not whether people are doing this, they clearly are, but whether the advice they are getting back is actually any good.
How many people are actually doing this
The scale here is bigger than most people realize. A Harris Poll survey conducted for NerdWallet in June 2026 found that 26% of Americans have used an AI chatbot to answer a personal finance question, and among those users, nearly half had done it within the past week. LendingTree's research puts chatbot usage even higher among younger adults, with AI shaping financial decisions for roughly half of the users it surveyed, and 61% of that group asking specifically for help managing day-to-day money and 42% asking about the stock market.
Separate data cited by fiduciary industry researchers suggests that around half of adults under 50 are using tools like ChatGPT for some form of financial planning, a number Forrester had projected would cross that threshold in 2026. This is not a niche behavior anymore. It is closer to a default first step for a large share of younger consumers before they ever talk to a bank, a broker, or a financial planner.
Is it safe to use ChatGPT for investing decisions
The honest answer, based on the available testing, is: not on its own. A September 2026 study by fintech research firm Saturn tested 18 popular AI models, including ChatGPT, Claude, Copilot, Grok, and Gemini, against 121 real financial questions repeated five times each, producing more than 10,000 responses. The overall error rate was 57%, and it climbed to 88% on the hardest, most complex questions. Free versions of the models performed worse than paid ones, with a 63% error rate compared to 49%.
The mistakes were not typos or rounding errors. Researchers found calculation errors, missing risk warnings, outdated tax information, and in some cases outright invented rules that do not exist in any real financial framework. A separate academic study found that when models were asked to cite sources for financial claims, ChatGPT fabricated references about 20% of the time, while Gemini did so in over 76% of cases. If a human advisor made up a rule or a citation at that rate, they would not keep their license for long.
The PensionBee findings on trust versus accuracy
What makes the PensionBee survey notable is that it measures the gap between confidence and actual accuracy. Among the 1,000 U.S. adults surveyed who already use chatbots for money questions, 23% said a chatbot had given them information about their finances that turned out to be wrong, and 5% said they only discovered the mistake after they had already acted on it. Despite that, 57% said they would act on a chatbot's guidance about something as significant as investing or retirement timing without double-checking it.
One detail stands out: Baby Boomers who use these tools were the least likely of any generation to say they had ever been given wrong information, at 82%. That could mean older users are asking simpler, more conservative questions, or it could mean they are less equipped to catch subtle mistakes in the answers they get. Either way, it points to a real gap between how accurate people believe these tools are and how accurate the testing shows them to be.
Chatbots are not the same thing as robo-advisors
It helps to separate two different technologies that get lumped together in this conversation. Robo-advisors like Betterment and Wealthfront have used automated, rules-based portfolio management for over a decade. They rebalance portfolios, harvest tax losses, and adjust allocations based on a risk questionnaire, all built on structured financial logic rather than a language model generating text on the fly.
General-purpose chatbots like ChatGPT, Claude, and Gemini are a different animal. They generate plausible-sounding answers based on patterns in training data, not verified financial rules tied to your actual accounts. That distinction matters because a robo-advisor cannot invent a tax rule that does not exist, while a chatbot demonstrably can and does, according to the Saturn testing.
Fintech companies are leaning into the trend anyway
Rather than warning people away from chatbot-based financial research, some fintech companies are building directly into it. LendingTree launched a ChatGPT plug-in on August 4, 2026, that lets users share their estimated credit score, state, and loan amount to see how those factors affect mortgage and refinance rates, then routes them to LendingTree's marketplace to compare real offers. CEO Scott Peyree framed it plainly: consumers are already researching financial decisions this way, so the company built a product to meet them there.
This is a useful signal. Companies with a direct financial stake in getting this right are not trying to stop people from asking chatbots about money, they are trying to plug their own verified data and offers into the conversation instead of leaving users with an unverified, generic answer. That is a tacit admission that the raw chatbot answer alone is not good enough to act on.
What this actually means for your money
The realistic middle ground looks like this: chatbots are genuinely useful for the things they are good at, explaining what a term means, helping you understand a concept you are embarrassed to ask a person about, or drafting questions to bring to an actual advisor. They are not reliable for the things people are increasingly using them for anyway, like specific tax guidance, retirement timing, or investment allocation, where the Saturn testing showed error rates climbing above 80% on complex questions.
- Use AI chatbots to learn concepts, not to make final decisions on retirement, taxes, or large investments.
- Never treat a single chatbot answer as verified, cross-check anything with a real number attached to it against a primary source or a professional.
- Be cautious about what personal financial details you type into a general chatbot, since these tools are not built with the same data handling standards as a bank or licensed advisor.
The takeaway
The real problem is not that people are using AI for financial questions, that shift is already permanent and not inherently bad. The problem is the mismatch between how confident these tools sound and how often independent testing shows they get the specifics wrong. A chatbot that is wrong 57% of the time on complex questions but delivers every answer with the same calm, authoritative tone is arguably more dangerous than a tool that flags its own uncertainty. Until that changes, the safest way to use AI for money decisions is as a starting point for questions, never as the final word.
