
Users of an AI-powered US wealth app increased their monthly share of index fund purchases by 7.6 percentage points (ppts) after adopting the tool, and decreased their monthly share of single-stock purchases by 5.9 ppts.
When retail investors asked an AI assistant what to buy, one fund kept coming up.
The Vanguard Total Stock Market ETF (VTI) was the most recommended ticker for six of the seven ChatGPT and Gemini models studied in a new NYU Stern working paper. Within the Lipper Multi-Cap Core category, it took 53.9% of investor ticker recommendations, against 15.8% of the category’s assets.
The paper, “Generative AI and Retail Investment” by finance PhD student Valerie Baldinger, finds that investors who started using an AI assistant shifted their new purchases away from single stocks and towards index funds. Decades of investor education and traditional advice have struggled to achieve that.
Baldinger used de-identified data from Mezzi, a US wealth management app and SEC-registered investment advisor, which added an assistant built on OpenAI’s ChatGPT and Google’s Gemini models in November 2024. Users link their brokerage and retirement accounts, so the assistant can see their holdings when it answers. The dataset covers about 1,900 investors and 69,598 prompts, matched to transaction and holdings records. The assistant does not execute trades: investors decide whether to act.
Baldinger served as an unpaid contractor for Mezzi and has no financial interest in the company, according to the paper. Mezzi has promoted an earlier version of the research on its website. The paper is her job market paper and has not been peer-reviewed.
Investors Ask About Stocks, the AI Suggests Funds
In just 3.5% of prompts submitted to chatbots did investors express interest in buying an ETF or mutual fund. However, the assistant recommended a fund in 18% of its answers, around five times as often. When investors asked questions that named no ticker and did not mention stocks, 89.7% of its buy recommendations were for ETFs or mutual funds.
Using a staggered difference-in-differences design, which compares investors who adopted the assistant at different times, Baldinger estimates that adoption raised index funds’ share of monthly purchases by 7.6 percentage points, from a base of about 33%, and cut individual stocks’ share by 5.9 points.
The shift reduced what economists call ‘return loss’. That is the expected return an investor gives up by holding diversifiable risk the market does not reward. Across the sample, annual return loss fell by 43 basis points (bps) from a pre-adoption average of about 407 bps. For the less-diversified half of investors, it fell by 101 bps. For the more diversified half, the effect was not statistically significant.
Investors also bought cheaper funds after adoption, with expense ratios on purchased funds about 3.6 bps lower.
Why Personalisation May Matter
Earlier attempts to change behaviour through advice have struggled. Baldinger cites a 2012 study by Utpal Bhattacharya and co-authors which found that only about 5% of investors offered free, unbiased advice took it up. Those who did saw no improvement in diversification.
However, the advice tends to have been followed more widely here. Of the answers to contain buy advice, investors bought at least one recommended security within 10 trading days in 36.2% of cases.
Using changes over time in which AI model was deployed, Baldinger estimates that a one-standard-deviation increase in how personalised an answer was raised the chance it was acted on by about 12 percentage points. She notes the estimate is imprecise and that other explanations — including model quality, trust, and on-demand access — cannot be ruled out.
The Instructions Matter
The platform’s own settings also shaped behaviour. In mid-December 2025, the assistant’s instructions were changed to encourage it to name specific tickers. The share of its answers recommending them jumped from under a fifth to about a third. Over the following four weeks, the platform’s heaviest existing users raised the fund share of their new investment by 8.1 percentage points relative to non-users.
In January 2026 the instructions were changed again, telling the assistant to avoid recommending tickers. Ticker recommendations fell to about 5% of answers. And the fund share of heavy users’ new investment fell by 7.7 points. That left it close to where it was before December.
Concentrated recommendations
Vanguard’s VTI was not an outlier. Across models, the 10 most recommended tickers accounted for half of all ticker recommendations. After adoption, the share of investors’ fund purchases going to the 20 most widely recommended funds rose by 7.2 percentage points.
Baldinger notes that this concentration may have implications for financial markets if retail investors herd into the same securities as AI adoption spreads.
Caveats
However, it must be noted that the sample is not representative. Mezzi users are wealthier and more active than typical US equity investors, with a median portfolio of about $556,000 against $155,000 for equity holders in the Federal Reserve’s Survey of Consumer Finances. While the sharpest diversification gains accrued to the least diversified portfolios, it remains unclear how similar tools would perform among lower-wealth or less financially engaged households.
Furthermore, adoption is voluntary and investors choose what to ask, which complicates causal claims. Mezzi is also a paid advisory service, and its assistant runs on Mezzi’s own instructions and account data. So the results may not carry over to people asking ChatGPT or Gemini directly.
All investors in the study are US-based, and European retail investors cannot generally buy US-domiciled ETFs such as VTI due to EU PRIIPs key information document rules. Even so, as European wealth platforms and neo-brokers embed conversational AI into their apps, the findings raise a pressing question for UCITS ETF issuers: what determines which funds the models recommend, and whether that can be shaped.
Originally published on ETF Stream
For more news information and analysis, visit the Equity ETF Content Hub.
A message from Advisor Perspectives and VettaFi: Discover something new! Click here to register for our upcoming webcasts.
© ETF Stream
More Fixed Income Topics >