Advised investors ask AI about money more, and verify it less

A new survey of 1,981 US adults suggests the people most likely to ask a chatbot about money are also the likeliest to act on the answer without a second opinion. That is the opposite of what the advice industry assumes about clients who already have a human adviser, and it is the kind of behavior gap that never shows up in a product demo.

What the study found

The Evergreen.ai Financial Decisions Study surveyed US adults aged 25 to 75 who make or share household financial decisions. Across the full sample, 48.4% had asked a chatbot about their finances.² Among AI users who answered the follow-up, 77.8% acted on what they were told, and 28.7% did so without verifying it.³

Advised investors verify less

Among respondents with an adviser, 63.8% had asked an AI tool about their money in the past 12 months, against 38.5% of those without one.²

The advised rate is nearly triple the unadvised one. Combining the two advised figures, roughly 27% of all advised respondents asked an AI money question and then acted without checking it. That 27% is a calculation from the reported rates, not a number the study states.⁴

Why the gap may not be what it looks like

The report's authors say the data shows an association, not proof that advice drives the behavior. One possible reading is that people used to seeking professional guidance treat a chatbot as one more adviser in the room. Another is that advised respondents skew wealthier and more engaged. The report itself notes that AI users were less likely than nonusers to say they had made a major decision without knowing its tax impact, 15.4% versus 41.6%, and attributes that gap partly to AI users skewing wealthier and more likely to have an adviser.²

The accuracy problem sits underneath

Separate benchmarking by adviser fintech Saturn, cited in the same coverage, found that popular chatbots give wrong financial answers more often than right ones. The best-performing free model still returned incorrect or incomplete responses 56% of the time.⁵ We have not seen Saturn's methodology, so treat that figure as a reported claim, not a verified ranking. It also means the tool a person trusts most is, on this benchmark, likelier to be wrong than right, which is why the verification gap in the survey matters.

This lines up with academic work on confident errors: models can sound sure while being wrong, which is exactly when a person is least likely to double-check.

Read the sponsor line

Evergreen.ai is a service of Evergreen Wealth Advisors, a Miami registered investment adviser, so the study's sponsor sells in this category. Its founder says Americans are seeking guidance wherever they can find it and asks whether the tools were built to deserve their trust. That is a reasonable question, and also a marketing position. Separately, a Security Benefit and Greenwald Research survey of 200 US financial professionals found 45% say clients often or sometimes bring AI-sourced information to them.⁶

A simple habit worth testing

Whatever the group, the useful behavior is cheap: ask a second model or a primary source before acting. A second answer takes under a minute, while an unchecked wrong one can cost real money. Our own side-by-side tests of the same finance question across several models are the next step here, and we want to know which questions you would like us to run.

Which money question should we put to five AI models next? Tell us what to test.