AI tools influence 58% of investors’ decisions, but Rothschild test flags reliability gaps

A study by the Securities Authority released early this year found that 58% of independent investors rely on artificial-intelligence tools when making investment choices. Most use the tools as a supplement rather than as the sole source, indicating a perceptual shift from a trader-only experiment to a standard component of the average investor’s toolkit.

Rothschild Bank evaluated the actual value of the AI-generated recommendations. The analysis showed three fundamental problems. First, the answers vary dramatically between runs; the same question can produce a completely different portfolio mix without an apparent reason. Second, the output is not personalized to the investor’s risk profile or a defined strategy. Third, the model is disconnected from real-time data sources, causing it to miss recent earnings reports, interest-rate changes or events that occurred yesterday.

The shortcomings stem from the architecture of large language models (LLM). LLMs are trained to predict the next token that sounds plausible, not to verify data against live feeds. When asked “where to invest 100 thousand shekel”, they generate text that resembles investment advice because they have seen millions of similar pages during training, yet they lack access to Bloomberg, Reuters or trading platforms. The result is a polished answer that is detached from current market conditions, a form of misleading financial advice without an error flag.

These tools are useful for filling gaps: explaining terminology, providing historical fund comparisons, or building “what-if” scenarios. They are not a substitute for professional advice or independent research. Entrusting ChatGPT or the cloud to manage a pension or stock portfolio is akin to playing roulette with a convenient interface. A two-day data lag is not a bug; it is a deal-breaker.