Study: half of ChatGPT and Gemini stock picks land in just 10 names
In brief
- ChatGPT and Gemini: roughly 50% of stock picks fell on 10 stocks, per an unnamed study.
- Semiconductors averaged 41% of LLM-built portfolios, versus roughly 20% for the S&P 500, per the NBER paper.
- NVIDIA often got 18% to 20% of ChatGPT-built portfolios, according to the NBER paper.
- Recommended stocks drew nearly 10 times the media coverage of average firms.
What the NBER paper found
The NBER working paper tested models including ChatGPT 5.0/5.2 and Gemini 2.5 Flash. The models leaned heavily toward large-cap technology and semiconductor stocks. According to Crypto Briefing's summary, LLM-built portfolios held an average of 41% in semiconductor stocks, compared with roughly 20% for the S&P 500.
ChatGPT showed the strongest tilt. The paper found it often gave NVIDIA between 18% and 20% of a portfolio (and that share rose further in actively managed portfolios). Gemini's portfolios were narrower still, with a median of only 4.5 stock holdings and a preference for large firms and momentum stocks, meaning companies whose shares had already been rising.
Same picks, different models
The convergence went beyond one model. Parallel experiments across different models produced consistent recommendations, with top holdings frequently shared between them, Crypto Briefing reported. As of November 2026, NVIDIA, Alphabet, Adobe and Broadcom kept appearing near the top of the lists.
The research also examined media coverage. Stocks the models recommended received nearly 10 times the media coverage of average firms. The recommendations tracked the volume of that coverage more closely than they tracked underlying financial fundamentals, the research found.
Why it matters for markets
Crypto Briefing presented the findings as a crowding risk. In its account, AI-driven picks that pile into the same handful of names can lead to crowded trades. That's the outlet's framing, and the published account doesn't include its own measurements of trading behavior.
The research itself flagged a related concern. If models keep favoring hype over fundamentals, it warned, market performance and actual corporate health can drift apart. Crypto Briefing described the NBER paper as pointing in the same direction as the November study.
Crypto Briefing's account didn't name the authors of the November study, and the NBER paper's findings above come from the outlet's summary rather than the paper's own text.
Frequently asked questions
What did the NBER paper find about AI-built stock portfolios?
According to Crypto Briefing's summary, the NBER working paper found that LLM-built portfolios held an average of 41% in semiconductor stocks, compared with roughly 20% for the S&P 500. ChatGPT often gave NVIDIA between 18% and 20% of a portfolio. Gemini's portfolios had a median of only 4.5 stock holdings.
Why do ChatGPT and Gemini keep recommending the same stocks?
The research found that AI recommendations tracked the volume of media coverage more closely than underlying financial fundamentals. Recommended stocks received nearly 10 times the media coverage of average firms. Parallel experiments across different models also often produced the same top holdings.
Which stocks appeared most often in the AI recommendations?
As of November 2026, NVIDIA, Alphabet, Adobe and Broadcom kept appearing near the top of the recommendations, according to Crypto Briefing. The models also leaned toward large-cap technology, semiconductor and momentum stocks.
