When an investor asks a chatbot, 'I am 30 years old, willing to take some risks, and have $10,000 to invest—what should I do?', the response is often strikingly fluent. It provides a structured plan, specific stock lists, and professional-sounding advice about diversification, usually capped with a polite disclaimer. The tone is persuasive, but does that confidence actually safeguard your capital?
A study published in PLOS ONE in June 2025 took this question into the lab. Philipp Winder and Christian Hildebrand from the University of St. Gallen, alongside Jochen Hartmann from the Technical University of Munich, tasked ChatGPT, Gemini, and Copilot with creating investment portfolios based on varying ages and risk appetites. When compared against a broad, low-cost world index—the Vanguard Total World ETF—the results were clear: portfolios suggested by large language models were riskier than the benchmark across all five measured dimensions of risk.1
Why do AI models show a geographic bias in their investment suggestions?
The most striking finding of the study concerns geographic concentration. On average, the models allocated approximately 93 percent of portfolio value to U.S. stocks, whereas a broad world index typically carries only about 59 percent. While the chatbot may describe its plan as a 'world' portfolio, it is effectively tethered to the American market. This leaves the portfolio significantly more vulnerable to shocks within the U.S. economy compared to a truly diversified global approach.1
The study also analyzed sectoral distribution, noting that models assigned higher weights to technology and consumer cyclical stocks than the benchmark, while underweighting transportation and services. Furthermore, there was a clear tendency to prioritize currently trending stocks. Researchers identified this as 'trend chasing,' where a model interprets high past performance as a guarantee of future returns. Modern portfolio theory explicitly warns against this bias, yet because large language models are trained on data containing these patterns, they tend to amplify these common investment mistakes.1
Do AI-generated investment plans lead to higher hidden costs?
For most individual investors, a simple, broad index fund serves as an inexpensive, low-maintenance foundation. In contrast, the AI models leaned toward stock picking and active management. This approach inevitably leads to higher turnover, management fees, and transaction costs, often justified by the model's apparent confidence in its stock selections. Consequently, the total expense ratios of the suggested portfolios were higher than those of the benchmark. Over time, even small differences in expense ratios can significantly erode cumulative returns.1
Researchers describe this phenomenon as an 'echo of biases,' where the model interprets an investor's risk appetite while simultaneously reinforcing the geographic and sectoral prejudices present in its training data. Even when researchers attempted to 'fine-tune' prompts to avoid these issues, such as by explicitly asking the model to avoid management fees, risk was only partially mitigated. While broader instructions to 'avoid common investment errors' were more effective, hidden risks remained. Fine-tuning a prompt alone does not guarantee a safe or balanced portfolio.1
What are the regulatory risks of using AI in finance?
These warnings extend beyond academia. In January 2024, the U.S. Securities and Exchange Commission (SEC), FINRA, and state regulators issued a joint investor alert. Their message was blunt: do not trust promises of 'winning systems' labeled as AI. The information provided by chatbots can be inaccurate, incomplete, or entirely fabricated, and investors should never base critical financial decisions solely on AI-generated text.2
The SEC also took action against two investment firms in 2024 for 'AI washing,' a practice of making false or misleading claims about the use of artificial intelligence in their services. Both Delphia and Global Predictions settled with the commission, paying a combined $400,000 in penalties for misrepresenting their AI capabilities. The danger is not limited to the chatbot itself; it also includes companies that use the term 'AI' as a marketing veneer to attract investors.3
How should investors use AI tools in their financial planning?
Large language models can be useful for tasks like organizing a budget, explaining complex financial concepts, or identifying potential gaps in one's understanding. The danger lies in delegating the architecture of your entire portfolio to a chatbot. Linguistic analysis from the study confirms that these models communicate with moderate-to-high confidence, providing seemingly logical justifications that are almost always accompanied by a generic 'do your own research' disclaimer. For many, this disclaimer is not sufficient to offset the authoritative tone of the preceding advice.1
A prudent approach involves treating any list generated by a chatbot as a rough idea rather than a final decision. Investors should verify figures, scrutinize geographic and sectoral concentrations, and evaluate hidden costs. Unlike registered investment advisors who are bound by professional and legal standards, chatbots operate without such oversight, making individual verification essential.
Notes and Sources
- Winder, P., Hildebrand, C. & Hartmann, J. (2025). "Biased echoes: Large language models reinforce investment biases and increase portfolio risks of private investors". PLOS ONE, 20(6), e0325459.
- U.S. SEC Office of Investor Education and Advocacy, NASAA & FINRA (January 25, 2024). "Artificial Intelligence (AI) and Investment Fraud: Investor Alert".
- U.S. Securities and Exchange Commission (March 18, 2024). "SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence".
