ⓘ Educational information only — not financial advice. Nothing on this page is a recommendation to buy or sell anything. Investing involves risk, including loss of principal.
Guide · Prediction & rules

AI Stock Trading & Prediction: What the Research Shows

Academic studies find real statistical signals in LLM stock analysis — and the regulators are busy punishing firms that oversell them. Here is the evidence, the caveats, and the rules.

TMBy Tomás Marin, CFA, Investment Analyst · Reviewed by Amara Reed, CFP · Fact-checked Sep 2026 · Last updated Sep 16, 2026

Can AI predict stock prices? The headline study

The landmark paper is Lopez-Lira & Tang's "Can ChatGPT Forecast Stock Price Movements?" — now published in the Journal of Financial Economics[1]. They asked ChatGPT whether news headlines were good or bad for a company, converted the answers to sentiment scores, and tested whether the scores predicted next-day returns out-of-sample. Result: ChatGPT-4's scores did predict daily returns, beating traditional sentiment methods — a long-short portfolio earned an average of 38 basis points per day before transaction costs[1]. Now read the caveats, because they are the whole story: results are pre-cost and daily-rebalanced (round-trip costs cut the cumulative return dramatically), predictability is concentrated in smaller stocks and negative news, and the authors' own theory predicts the edge shrinks as more people use LLMs.

The intellectual backdrop is half a century old: Fama's efficient-market hypothesis says prices already reflect public information[2], and SPIVA scorecards keep showing that most active professional managers underperform their benchmarks over time[3]. If professionals with teams and data struggle, an app promising easy predictions deserves deep skepticism.

The honest summary

AI finds weak, shrinking, cost-sensitive statistical edges in research settings. That is fascinating — and it is not a reason to trade your savings on a bot's say-so.

Can ChatGPT pick stocks?

ChatGPT can summarize filings, explain concepts, model a budget and brainstorm research questions. What it cannot do: act as your adviser. OpenAI maintains Financial Services Terms restricting professional financial-advice use, and a general-purpose chatbot has no registration, no fiduciary duty and no obligation to be right[4]. Its training data has a cutoff, it can hallucinate figures and sources, and live tools change its answers. Treat its output as a research assistant's first draft — then do the verification yourself.

How people "AI trade" today

The realistic tools are less dramatic than the ads. Screeners and scores: Danelfin rates stocks 0–10 by modeled odds of beating a benchmark, and its own FAQ is admirably blunt — "a highly rated asset can still lose money" — with featured results labeled as backtests[5]. Strategy builders: Composer (now part of SoFi) lets you assemble no-code automated strategies from $10/month[6]. Research terminals: Fiscal.ai (formerly FinChat) answers questions over S&P Global data[7]. All of them are inputs — none of them replaces your judgment, and none of them owes you a profit.

The rules: AI-washing is an enforcement priority

The SEC has already charged firms for fake AI claims — and the cases are a plain-English guide to what "too good to be true" looks like. Delphia claimed it used client data to train AI that predicted "companies and trends about to make it big"; the SEC found the system never existed — $225,000 penalty[8]. Global Predictions (PortfolioPilot) claimed to be the "first regulated AI financial advisor" — $175,000 penalty[8]. Rimar Capital claimed an AI platform managing up to $20 million when real assets were under $2 million — $310,000 in penalties plus an industry bar[9]. The SEC, FINRA and NASAA also issued a joint investor alert in January 2024 warning that "AI trading bots" promising outsized returns are a fraud pattern[10].

Red-line checklist

Walk away from: guaranteed returns, "cannot lose" claims, unregistered bots or advisers, pressure to deposit quickly, and anyone who won't show a registration number you can verify on IAPD[11]. Promising certainty is the oldest fraud signal there is — AI just gives it a new costume.

Frequently asked questions

Can AI predict stock prices?+
Not reliably for individuals. Academic research finds statistical predictability from LLM sentiment analysis — pre-cost, concentrated in small stocks, and expected to shrink with adoption[1]. That is research, not a retail strategy.
Is AI stock trading legal?+
Using AI tools to research and trade is legal. Firms offering automated advice or management must be SEC-registered, and the SEC has charged firms for false AI claims[8][9]. Unregistered bots promising guaranteed returns are a classic fraud pattern[10].
Can ChatGPT pick stocks?+
It can summarize filings and explain concepts, and academic studies show its analysis carries statistical return information[1] — but it is not a licensed adviser, has no fiduciary duty, and OpenAI restricts professional financial-advice use[4].
Why do most AI trading bots lose money?+
Three structural reasons: trading costs eat small statistical edges[1], short-term prediction is extraordinarily hard (efficient-market evidence[2]; most active managers lag benchmarks[3]), and many "bots" are outright scams[10].
What is SEC "AI-washing"?+
Making false or unsubstantiated AI claims to sell investment products. The SEC has settled cases against Delphia, Global Predictions and Rimar for exactly that[8][9] — the same Marketing Rule that governs adviser advertising applies to AI claims.

Sources

All claims verified September 2026. Academic results are reported with their stated caveats.

  1. Lopez-Lira & Tang, "Can ChatGPT Forecast Stock Price Movements?", Journal of Financial Economics (2026): ar5iv.labs.arxiv.org · sciencedirect.com
  2. Fama, "Efficient Capital Markets", Journal of Finance 25(2), 1970: jstor.org
  3. S&P Dow Jones Indices — SPIVA US scorecards: spglobal.com
  4. OpenAI Financial Services Terms: openai.com; The Verge (Nov 3, 2025): theverge.com
  5. Danelfin review and pricing (Aug 30, 2026): therundown.ai
  6. Composer pricing (official): composer.trade
  7. Fiscal.ai (formerly FinChat): finchat.io
  8. SEC — AI-washing settlements (Delphia, Global Predictions), Mar 2024: sec.gov · mofo.com
  9. SEC — Rimar Capital charges (Oct 2024): sec.gov · mofo.com
  10. SEC/FINRA/NASAA investor alert — AI and investment fraud (Jan 2024): investor.gov
  11. SEC IAPD — verify any adviser: adviserinfo.sec.gov