Can ChatGPT Pick Stocks? Testing AI Picks Against the S&P 500

By Chanuka Geekiyanage Oct 16, 2026

Yes, ChatGPT can pick stocks that beat the S&P 500 over a few weeks or months, and several public tests show it. The best evidence suggests it does this by loading up on a narrow set of large, high-beta tech names rather than by showing real skill. A short win can push a reader into a portfolio that is far less diversified than the index it beat. This article compares the main public tests, explains why their results are hard to trust, and shows how to use ChatGPT for stock research without copying its picks blindly.

Panaprium is independent and reader supported. If you buy something through our link, we may earn a commission. If you can, please support us on a monthly basis. It takes less than a minute to set up, and you will be making a big impact every single month. Thank you!

What the Public Tests Show So Far

The results depend heavily on who ran the test, how long it lasted, and whether real money was at stake. Here are the four most useful tests, compared on the points that decide how much weight each deserves.

Test	Setup	Result	Main limit
NBER working paper (Carlin, Israelsen, Wazzan)	Daily prompts to ChatGPT, Claude, Gemini and Grok from August 2025, about 8 months	Raw returns beat the S&P 500, but abnormal returns were not statistically significant after matching for size, value, and momentum.	Claude, Gemini, and Grok samples are short.
Rallies AI Arena	$100,000 paper portfolio for ChatGPT from March 2026	About $159,250 by June 5, 2026, versus roughly 8% for the S&P 500	Paper trading, few holdings, one market regime
Alpha Arena Season 1	$10,000 per model, live crypto perpetual futures trading, Oct 18 to Nov 3, 2025	GPT-5 finished at $4,126, while Qwen3 Max led at $12,231	Crypto trading, not stock picking
Finder.com (2023)	ChatGPT picked stocks using debt and growth criteria, 8 weeks	Up 4.9% versus 3% for the S&P 500	Eight weeks and an old model

The NBER paper is the most rigorous of the four. It tracked live, forward-looking recommendations, so it avoids the problem of testing a model on data it may have already seen. Its verdict was that AI portfolios did not earn statistically significant abnormal returns.

Alpha Arena shows the opposite problem. The organizer himself said the models were given only numerical time-series data, which language models handle poorly. A bad result there says little about picking stocks, but it does warn against using a chatbot for short-term trading.

Why a Short Win Proves Little

The NBER team found that unconstrained AI portfolios are concentrated. ChatGPT 5.0 held roughly 18 to 19 stocks, and put 18 to 20 percent of the portfolio into Nvidia throughout the sample. Gemini’s portfolios held a median of just 4.5 names.

Semiconductors made up an average of 41 percent of the AI portfolios, against about 20 percent of the S&P 500. The average stock chosen had a beta of about 1.6, meaning it tends to swing roughly 60 percent more than the market. When researchers asked one ChatGPT version to hold beta between 0.9 and 1.1, it stayed near the top of that range and occasionally broke through it.

This is why raw outperformance misleads. If you buy large, high-momentum tech stocks during a period when those stocks win, you beat the index without any stock-picking skill. The researchers tested this with a peer-matched benchmark, and the apparent alpha mostly disappeared.

Two other problems make results hard to read. First, picks are driven by attention: the selected stocks received about ten times as many news articles as the average listed firm. Popular stocks are often already priced in.

Second, backtests can leak. A December 2025 arXiv study found that a measure of how likely a prompt was in a model’s training data explained about 37 percent of the standalone effect of the model’s return predictions. Any test that asks ChatGPT about past periods should be treated with caution.

Regime changes matter too. Money tracked monthly AI-selected picks starting in July 2025. In September, 8 of 9 outperformed the S&P 500, but in December, after a rotation out of tech, only 4 of 9 did, the worst month of the tracking period.

The Rallies Result: Signal or One Big Bet?

The Rallies AI Arena result is the most quoted headline for ChatGPT. By April 24, 2026, the $100,000 paper account had reached $139,005, while the S&P 500 equivalent stood near $106,129. By June 5, it was about $159,250, with the S&P 500 up roughly 8% over the same period.

The composition tells the story. On April 24, the largest holding was $66,500 in Credo Technology, nearly half the account. The rest sat in Alphabet, Nebius, and Amphenol, all tied to AI infrastructure, and ChatGPT sold nothing during that stretch.

That is a bet on one theme working, and it worked. The figures are paper trades over a few months, and the same portfolio would have looked very different if AI infrastructure stocks had sold off. By June, reporting noted the strategy had shifted toward a more balanced mix with cash, which is itself a sign that the early version carried more risk than it first looked.

The S&P 500 Is a Harder Benchmark Than It Looks

Beating the index is difficult even for professionals. S&P Dow Jones Indices reported that 79% of active large-cap U.S. equity funds underperformed the S&P 500 in 2025, a year the index gained 18%. Few funds repeat a good year, which is why a chatbot beating the index for eight weeks is weak evidence.

The index is also more concentrated than its name suggests. As of August 31, 2026, the ten largest constituents were 37.8% of the S&P 500 by S&P Dow Jones Indices data, with the largest stock alone near 8%. That means a buy-and-hold index investor already owns a heavy dose of the same AI leaders ChatGPT tends to pick, which is covered in more depth in this look at how AI stock concentration affects S&P 500 risk.

Image source: S&P Dow Jones

The practical consequence is overlap. If you hold an S&P 500 fund and then add ChatGPT’s Nvidia-heavy picks, you are doubling up on exposure you already own. You may feel diversified while your real risk rises.

What to Watch for in AI Stock Picks

These are the risks that matter most when you read a ChatGPT portfolio:

Single-name concentration: ChatGPT 5.0 held 18 to 20 percent of its NBER portfolio in Nvidia. One earnings miss would hit the whole account.
Sector tilt: Semiconductors averaged 41 percent of AI portfolios versus about 20 percent of the S&P 500.
High beta: An average beta near 1.6 means larger drops in a falling market.
Attention bias: Picks follow news volume, so you often buy what everyone already knows about.
Momentum chasing: Money’s December result shows how quickly a tech rotation can erase an AI edge.
Training leakage: Backtests of language models can overstate skill because the model may have seen the period.

Where ChatGPT Is Useful and Where It Is Not

The mistake is treating ChatGPT as a fund manager. It is better used as a research assistant whose work you check.

Situation	Best use	Why
Building a core long-term portfolio	A broad index fund, not ChatGPT picks	Most active large-cap funds lag the S&P 500, and AI picks add concentration.
Researching a company you already follow	Ask ChatGPT to summarize filings and list risks	It saves time, and you can verify every number
Wanting AI exposure	Check overlap with your index fund first	Mega-cap AI stocks are already a large part of the S&P 500
Short-term trading	Avoid it	Alpha Arena showed chatbots handle raw price data poorly
Curiosity about AI picks	Paper trade for 6 to 12 months	Samples of a few weeks are too short to separate luck from skill

If you want AI exposure without picking single names, the tradeoffs are laid out in this comparison of AI stocks versus AI ETFs on risk-adjusted returns. Funds reduce company-specific risk, but they can still be heavy in the same technology holdings.

Before you act on any AI-generated pick, run these checks:

Compare the pick’s return against the S&P 500 over the same dates, including dividends.
Find how much of the gain came from one stock, as with Credo in the Rallies account.
Look up the stock’s beta and sector weight, then compare them with what you already own.
Verify revenue, margin, and debt figures in the company’s own filing, not in the chatbot’s summary.
Size the position so a 40 percent drop would not change your plans.
Set a review date for the thesis instead of reacting to price moves.

My Take

ChatGPT can generate a plausible stock list, but I would not trust it to beat the market. The strongest study found that its outperformance came from tilts toward large, momentum-heavy tech stocks, not from skill, and the most dramatic winning runs rest on a few positions in one theme. The honest answer to the title question is that it can pick winners, but nobody has shown it can do so reliably after adjusting for risk.

I would use it to read filings, summarize earnings calls, and stress-test my own ideas. I would not use it for position sizing or timing, and I would not use it on a core portfolio. For most readers, a low-cost index fund remains the better baseline, and any AI-picked stocks should be a small satellite you can afford to lose.

The same logic applies to crypto. A chatbot that favors the most-covered coins will tilt your portfolio toward whatever is already crowded. Check concentration and overlap before you treat any AI list as a plan.

Conclusion

ChatGPT’s stock picks have beaten the S&P 500 in several short tests, but the most rigorous study found no significant abnormal returns once size, value, and momentum were accounted for. The wins came from concentrated, high-beta tech exposure, which can reverse quickly, as the tracked December results showed.

Use ChatGPT as a research tool and keep your core holdings in diversified funds. The practical next step is to take any AI pick and run the checks above: compare it against the index, measure concentration, and confirm the facts in primary filings before you put money in.

FAQs

1. Has ChatGPT ever beaten the S&P 500 in a real test?

Yes, in several short tests such as Rallies AI Arena, where a paper portfolio reached about $159,250 from $100,000 by June 5, 2026. The NBER study found that these gains were not statistically significant once portfolio tilts were adjusted for.

2. Should I buy stocks that ChatGPT recommends?

Not without checking them yourself, because AI picks tend to be concentrated in a few large, high-beta tech names. Compare each pick with your existing holdings and verify the company data in its filings first.

3. Is ChatGPT better than a financial advisor at picking stocks?

There is no evidence that it is, and 79% of active large-cap funds lagged the S&P 500 in 2025 even with professional managers. A chatbot also does not know your goals, taxes, or risk tolerance.

4. Can ChatGPT trade crypto profitably?

In Alpha Arena Season 1, GPT-5 lost most of its $10,000 while trading crypto perpetual futures over about two weeks. That test was short and narrow, but it gives little reason to trust a chatbot with active trading.

5. How long should I test AI stock picks before trusting them?

Paper trade for at least 6 to 12 months across different market conditions, and compare results with an index fund over the same dates. Even then, check how much of the return came from a single stock or sector.

References

NBER, AI Managed Household Portfolios: A Preliminary Report: https://www.nber.org/papers/w35153

Science of Money, Can ChatGPT beat the S&P 500? Eight months of daily picks suggest no: https://www.scienceofmoney.org/when-chatbots-play-stock-picker-what-ai-actually-recommends-for-your-portfolio-559/

Finbold, ChatGPT portfolio crushes stock market, gains 60%: https://finbold.com/chatgpt-portfolio-crushes-stock-market-gains-60/

Money, Under-the-Radar Stocks to Buy in Early 2026, as Picked by ChatGPT and Gemini: https://money.com/ai-picks-stocks-this-month/

S&P Dow Jones Indices, S&P 500: https://www.spglobal.com/spdji/en/indices/equity/sp-500/

Was this article helpful to you? Please tell us what you liked or didn't like in the comments below.

About the Author: Chanuka Geekiyanage

Chanuka is a content writer, editor and contributor at Panaprium, specializing in long-form editorial articles. He conducts in-depth research and produces well-structured, carefully edited content with a strong focus on clarity, SEO best practices, and consistent formatting. His work spans lifestyle, fashion, wellness, and technical topics, and all articles are reviewed and refined to meet Panaprium’s editorial standards.

What We're Up Against

Multinational corporations overproducing cheap products in the poorest countries.
Huge factories with sweatshop-like conditions underpaying workers.
Media conglomerates promoting unethical, unsustainable products.
Bad actors encouraging overconsumption through oblivious behavior.
- - - -
Thankfully, we've got our supporters, including you.
Panaprium is funded by readers like you who want to join us in our mission to make the world entirely sustainable.

If you can, please support us on a monthly basis. It takes less than a minute to set up, and you will be making a big impact every single month. Thank you.