ChatGPT portfolio up 43% trading

Social posts on Tuesday circulated a screenshot claiming a “ChatGPT portfolio” was up 43% and had recently added Visa, reviving a familiar question about whether large language models can do more than summarize markets. The post did not appear to come from OpenAI, a registered investment adviser or a broker-dealer. Publicly available material instead points to a patchwork of retail experiments, academic back tests and newsletter-style tracking projects that use ChatGPT or related models to generate stock ideas. Those projects are public, but they are small, self-directed and heavily dependent on human-set rules. ### Where did the 43% claim come from? An X post shared on May 19 said a ChatGPT-run portfolio was up 43% and had added Visa, but the publicly visible post did not include a full methodology, brokerage statement or independently audited performance record. A separate June 7, 2025 Morningstar/MarketWatch report described University of Florida finance professor Alejandro Lopez-Lira’s work testing ChatGPT, DeepSeek and Grok for stock selection. One excerpt from that report said returns tied to ChatGPT client portfolios were 43.5% from September 2023 to May 30, 2025, though the available snippet did not provide the full portfolio construction, fees or turnover. (x.com) ### Is this OpenAI running money for investors? OpenAI’s public ChatGPT pages describe the product as a general-purpose assistant, and the material reviewed here did not show OpenAI offering a public, regulated stock-picking fund under the “ChatGPT portfolio” label. Nathan Smith, an Oklahoma high school student, has published one of the clearest public examples of a model-directed trading experiment. His GitHub page says ChatGPT managed a “real-money micro-cap stock portfolio,” with daily updates, weekly deep research and transparent logging. (morningstar.com) Smith wrote that the project was built for evaluation and documentation rather than “performance optimization” or monetization. ### How do these experiments actually work? (chatgpt.com) Nathan Smith’s repository says prompting consisted of daily updates and weekly deep research, with scripts handling price evaluation, holdings updates and performance tracking. A Futurism report on Smith’s project said he fed GPT-4o daily portfolio data and enforced strict stop-loss rules, meaning a human still defined the constraints and executed the process around the model. Futurism reported on July 31, 2025 that Smith’s $100 experiment was up about 25% in its first month. (github.com) Benzinga, in a report carried by Webull and other sites, described the same project as returning roughly 24% to 25% over four weeks while focusing on U.S. micro-cap stocks under $300 million in market value. ### What evidence is there beyond social media screenshots? Alejandro Lopez-Lira and Yuehua Tang published a paper titled “Can ChatGPT Forecast Stock Price Movements? (github.com) Return Predictability and Large Language Models,” first posted in 2023 and updated in 2025. The paper said large language models could predict stock reactions from news headlines in back tests using more than 134,000 headlines for over 4,000 companies. Morningstar’s June 2025 interview with Lopez-Lira said he had been impressed by what current AI chatbots could do in equity trading, while also acknowledging that AI systems can make mistakes. (futurism.com) The same report said his early work was based on simulated historical testing rather than live trading. (arxiv.org) ### Why does Visa show up in the latest version? A Morningstar excerpt listing one AI-generated portfolio included Visa at a 5% weight, describing it as a resilient payment network tied to global transaction growth. That does not by itself verify the May 19 social-media portfolio, but it shows Visa has appeared in at least one publicly described AI-generated stock basket. (morningstar.com) Visa has also been a frequent AI-related equity because of its role in digital payments infrastructure and its own public push into AI-enabled commerce tools. That makes it a plausible holding in model-generated portfolios looking for large-cap, cash-generative businesses. That is an inference from Visa’s public positioning and the portfolio excerpt, not a disclosed rationale from the May 19 poster. (morningstar.com) ### What comes next if readers want to follow this story? Nathan Smith’s Substack archive shows weekly updates running through late December 2025, including posts labeled “Week 24,” “Week 25” and “Week 26 Final Week Results.” His GitHub page remains public and says monthly blog updates document experiment progress. Alejandro Lopez-Lira’s research page and paper archive remain public, and the latest cited academic version of his ChatGPT stock-prediction paper was revised on October 28, 2025. (corporate.visa.com) Any new claim that a “ChatGPT portfolio” is up 43% would need a dated holdings list, a start date, and a documented benchmark to be checked against those earlier experiments. (alejandrolopezlira.site) (nathanbsmith729.substack.com)