MW What AI can - and can't - do for investors right now

By Gordon Gottsegen

Investors have plenty of options when it comes to using AI as a tool for investing, but these kinks still need to be ironed out

Artificial intelligence promises one of the biggest leaps forward in human productivity. So it shouldn't come as a surprise that traders want to use the technology to make money.

With its ability to take in and analyze a wide swath of data, AI has the potential to supercharge for investors the countless market-relevant news articles, earnings calls and social-media posts that it learns from. But as AI learns more, it also hits a wall of limitations. Experiments by financial-technology companies show that AI can synthesize information well, but also that AI currently struggles when it comes to telling an investor what to do with the information it provides.

"I think the fun thing about this space is that it feels like what [AI is good at] changes every three to four weeks. Something that wasn't good today might be really good in three to four weeks," Abhishek Fatehpuria, the vice president of product at brokerage firm Robinhood, told MarketWatch. "Generally, the stuff that we found [AI] to be really good at is summarization and synthesis of broad sources of complex information, which actually fits really nicely into the process of an investor."

Robinhood (HOOD), an investing platform popular among individual retail investors, is planning to launch its foray into AI, Robinhood Cortex, later this year. In a March demonstration of the tool, Robinhood showed how Cortex could pull together summaries of fundamental and technical analysis, or what the folks on Wall Street were saying, and then package it up in a quick summary that was easy for your average individual investor to understand.

The version of Cortex unveiled in the demonstration also went a step further by letting investors input their trading hypothesis - say, that a certain stock will go up by a certain amount by a certain time - and then suggest different trades to capitalize on that hypothesis, highlighting specific options strategies and so on.

The final product may differ from what executives previewed earlier this year. But the demonstration gave investors a hint of what's to come. Perhaps one day, the tool could empower investors to use AI to consistently beat the market.

Read more: A professor testing ChatGPT's, DeepSeek's and Grok's stock-picking skills suggests stockbrokers should worry

Combing through earnings calls

There have been several studies from Georgia State, the University of Chicago Booth School of Business and the University of Munster that used AI to comb through earnings calls and highlight issues relevant to investors. In those studies, large language models were able to pick up on subtle shifts in tone or evasiveness from business leaders, and use them to identify corporate risks and upcoming changes in capital expenditures.

But earnings calls are already highly scrutinized, and business executives are often prepped going into them. If investors increasingly use AI to interpret these calls, it's easy to imagine a future where business leaders talk in a way to appease AI. If an AI is trained to detect certain words that point to potential risks - as in the Chicago Booth study - executives may be trained to stop using those words.

AI isn't perfect and hardcore users of AI often point out that it can "hallucinate" or provide incorrect information. There have been examples of AI misinformation impacting financial markets, like in May 2023 when an AI-generated image of an explosion near the Pentagon briefly sent stocks tumbling. Although that image was designed to deceive, information that's inadvertently false could also have a similar effect.

MarketWatch ran an experiment where it fed ChatGPT a link to Robinhood's annual 2024 10-K financial filing and prompted the AI to summarize the finding and identify key points an investor should know. ChatGPT did a good job, pointing out financial metrics like earnings per share, business metrics like total funded customers and potential risks. It even included a stock chart at the bottom showing recent performance.

But when the same exact prompt was given to Google's Gemini, it got confused. It pulled data from Robinhood's 10-K filing for full-year 2023 instead of 2024. In this instance, Gemini prioritized searching Google for a recent 10-K filing, and landed on a 10-K report that was published in 2024, instead of using the one that covered 2024's financials - despite the fact that the correct report was linked to in the prompt. When corrected, it referenced a quarterly 10-Q financial report, again instead of the one included in the prompt.

Alphabet's $(GOOG)$ $(GOOGL)$ Google declined to comment.

When AI references the wrong source of information it can lead to mistakes and incorrect results. This is why many LLMs, like Gemini, have disclaimers at the bottom telling users to double-check responses. Gemini even has a double-check feature built into its interface. Google also tells users not to rely on Gemini for "medical, legal, financial or other professional advice" on a help page.

When AI is wrong, it's especially problematic if you're putting your money on the line. The bottom line: Humans still need to play a significant role when using AI to invest.

"The main issue right now with AI and investing is the amount of text you need to deal with. Even a single 10-K is enough to saturate the context window of an LLM. So you need to be smart about extracting just the relevant sections you need from tons of documents, which is the challenging part," investor Jeffrey Emanuel told MarketWatch.

Emanuel has worked as an investment analyst at different Wall Street firms, but he also has experience building his own crypto and AI projects, which have given him a deep technical background as well. In January, Emanuel wrote a blog post poking holes in Nvidia's $(NVDA)$ market outperformance, which helped spark a selloff of the stock related to DeepSeek's competing AI model.

Read more: The blogger who helped spark Nvidia's $600 billion stock collapse and a panic in Silicon Valley

But ultimately for AI to become a reliable investing tool, investors shouldn't need a technical background like Emanuel's in order to harness it, he said. Instead, AI needs to get better at presenting good information.

Emanuel said that in order for AI to succeed, it needs to progress toward more advanced searching and processing capabilities. He pointed to examples like multi-agent workflows, which allow multiple LLMs with their own specialties to work on the same problem side by side.

While some LLMs may struggle to analyze financial data, that could change if AI tools are designed specifically for this purpose.

"But that doesn't mean it will come up with good trade ideas, because the market is way more complex than that," Emanuel said. He pointed out that sometimes stocks move in the exact opposite direction than what the best information suggests will happen. That's because markets react to a wide range of primary and secondary variables. This makes it hard for AI to understand the direction of investments. But most humans have a hard time too, for the same reasons, Emanuel noted.

How AI can help investors

If AI has limitations when it comes to trading, it's important for investors to understand what it can and can't do before relying on it too heavily, experts say.

One thing that AI does pretty reliably is pattern recognition, and this can help with strategies involving technical analysis that try to determine where stocks are headed based on previous trading activity.

"Humans excel at pattern recognition, however when trading it is common for traders to find patterns that may not be fully formed to reinforce a market view or position," Neil McDonald, the U.S. chief executive of investing platform Moomoo, told MarketWatch.

The human brain sometimes convinces us that we see patterns that, in reality, don't exist, but it can miss patterns too, McDonald said. Since AI can be more methodical, it can provide a benefit to investors who use technical analysis to trade.

Moomoo $(FUTU)$ has built an AI-enabled Pattern Finder feature into its platform to hunt for 16 different technical-analysis signals across thousands of stocks and investments. According to McDonald, this helps keep technical traders "honest" when looking for patterns.

AI is cannibalizing social

Retail brokerage Public has been experimenting with using AI for investing over the past two years.

"We really found that the key for AI to work in investing is to have it admit when it doesn't know the answer, which is something that most other general-purpose AIs are not really good at." Jannick Malling, the co-chief executive of Public, told MarketWatch.

Investors don't want AI to be like the kid in class who raises their hand in response to every question, even if they don't know the answer. Public introduced Alpha, its AI investing "co-pilot" powered by ChatGPT-4, in May 2023. Alpha provides users with a language-based interface where investors can ask about specific assets, get analysis or do market screening. As Public has worked to improve Alpha, it found that AI could help motivate investors better than any other source of information.

"We've always tracked this thing called actionability rate, which is for any piece of content in the app, what's the correlation to making a trade that same day?" Malling said. "With news articles [the actionability rate] is like 2%. The social content was like 5%. With AI it's close to 40%. So it's been an order of magnitude more actionable."

MW What AI can - and can't - do for investors right now

By Gordon Gottsegen

Investors have plenty of options when it comes to using AI as a tool for investing, but these kinks still need to be ironed out

Artificial intelligence promises one of the biggest leaps forward in human productivity. So it shouldn't come as a surprise that traders want to use the technology to make money.

With its ability to take in and analyze a wide swath of data, AI has the potential to supercharge for investors the countless market-relevant news articles, earnings calls and social-media posts that it learns from. But as AI learns more, it also hits a wall of limitations. Experiments by financial-technology companies show that AI can synthesize information well, but also that AI currently struggles when it comes to telling an investor what to do with the information it provides.

"I think the fun thing about this space is that it feels like what [AI is good at] changes every three to four weeks. Something that wasn't good today might be really good in three to four weeks," Abhishek Fatehpuria, the vice president of product at brokerage firm Robinhood, told MarketWatch. "Generally, the stuff that we found [AI] to be really good at is summarization and synthesis of broad sources of complex information, which actually fits really nicely into the process of an investor."

Robinhood (HOOD), an investing platform popular among individual retail investors, is planning to launch its foray into AI, Robinhood Cortex, later this year. In a March demonstration of the tool, Robinhood showed how Cortex could pull together summaries of fundamental and technical analysis, or what the folks on Wall Street were saying, and then package it up in a quick summary that was easy for your average individual investor to understand.

The version of Cortex unveiled in the demonstration also went a step further by letting investors input their trading hypothesis - say, that a certain stock will go up by a certain amount by a certain time - and then suggest different trades to capitalize on that hypothesis, highlighting specific options strategies and so on.

The final product may differ from what executives previewed earlier this year. But the demonstration gave investors a hint of what's to come. Perhaps one day, the tool could empower investors to use AI to consistently beat the market.

Read more: A professor testing ChatGPT's, DeepSeek's and Grok's stock-picking skills suggests stockbrokers should worry

Combing through earnings calls

There have been several studies from Georgia State, the University of Chicago Booth School of Business and the University of Munster that used AI to comb through earnings calls and highlight issues relevant to investors. In those studies, large language models were able to pick up on subtle shifts in tone or evasiveness from business leaders, and use them to identify corporate risks and upcoming changes in capital expenditures.

But earnings calls are already highly scrutinized, and business executives are often prepped going into them. If investors increasingly use AI to interpret these calls, it's easy to imagine a future where business leaders talk in a way to appease AI. If an AI is trained to detect certain words that point to potential risks - as in the Chicago Booth study - executives may be trained to stop using those words.

AI isn't perfect and hardcore users of AI often point out that it can "hallucinate" or provide incorrect information. There have been examples of AI misinformation impacting financial markets, like in May 2023 when an AI-generated image of an explosion near the Pentagon briefly sent stocks tumbling. Although that image was designed to deceive, information that's inadvertently false could also have a similar effect.

MarketWatch ran an experiment where it fed ChatGPT a link to Robinhood's annual 2024 10-K financial filing and prompted the AI to summarize the finding and identify key points an investor should know. ChatGPT did a good job, pointing out financial metrics like earnings per share, business metrics like total funded customers and potential risks. It even included a stock chart at the bottom showing recent performance.

But when the same exact prompt was given to Google's Gemini, it got confused. It pulled data from Robinhood's 10-K filing for full-year 2023 instead of 2024. In this instance, Gemini prioritized searching Google for a recent 10-K filing, and landed on a 10-K report that was published in 2024, instead of using the one that covered 2024's financials - despite the fact that the correct report was linked to in the prompt. When corrected, it referenced a quarterly 10-Q financial report, again instead of the one included in the prompt.

Alphabet's $(GOOG.UK)$ (GOOGL) Google declined to comment.

When AI references the wrong source of information it can lead to mistakes and incorrect results. This is why many LLMs, like Gemini, have disclaimers at the bottom telling users to double-check responses. Gemini even has a double-check feature built into its interface. Google also tells users not to rely on Gemini for "medical, legal, financial or other professional advice" on a help page.

When AI is wrong, it's especially problematic if you're putting your money on the line. The bottom line: Humans still need to play a significant role when using AI to invest.

"The main issue right now with AI and investing is the amount of text you need to deal with. Even a single 10-K is enough to saturate the context window of an LLM. So you need to be smart about extracting just the relevant sections you need from tons of documents, which is the challenging part," investor Jeffrey Emanuel told MarketWatch.

Emanuel has worked as an investment analyst at different Wall Street firms, but he also has experience building his own crypto and AI projects, which have given him a deep technical background as well. In January, Emanuel wrote a blog post poking holes in Nvidia's (NVDA) market outperformance, which helped spark a selloff of the stock related to DeepSeek's competing AI model.

Read more: The blogger who helped spark Nvidia's $600 billion stock collapse and a panic in Silicon Valley

But ultimately for AI to become a reliable investing tool, investors shouldn't need a technical background like Emanuel's in order to harness it, he said. Instead, AI needs to get better at presenting good information.

Emanuel said that in order for AI to succeed, it needs to progress toward more advanced searching and processing capabilities. He pointed to examples like multi-agent workflows, which allow multiple LLMs with their own specialties to work on the same problem side by side.

While some LLMs may struggle to analyze financial data, that could change if AI tools are designed specifically for this purpose.

"But that doesn't mean it will come up with good trade ideas, because the market is way more complex than that," Emanuel said. He pointed out that sometimes stocks move in the exact opposite direction than what the best information suggests will happen. That's because markets react to a wide range of primary and secondary variables. This makes it hard for AI to understand the direction of investments. But most humans have a hard time too, for the same reasons, Emanuel noted.

How AI can help investors

If AI has limitations when it comes to trading, it's important for investors to understand what it can and can't do before relying on it too heavily, experts say.

One thing that AI does pretty reliably is pattern recognition, and this can help with strategies involving technical analysis that try to determine where stocks are headed based on previous trading activity.

"Humans excel at pattern recognition, however when trading it is common for traders to find patterns that may not be fully formed to reinforce a market view or position," Neil McDonald, the U.S. chief executive of investing platform Moomoo, told MarketWatch.

The human brain sometimes convinces us that we see patterns that, in reality, don't exist, but it can miss patterns too, McDonald said. Since AI can be more methodical, it can provide a benefit to investors who use technical analysis to trade.

Moomoo (FUTU) has built an AI-enabled Pattern Finder feature into its platform to hunt for 16 different technical-analysis signals across thousands of stocks and investments. According to McDonald, this helps keep technical traders "honest" when looking for patterns.

AI is cannibalizing social

Retail brokerage Public has been experimenting with using AI for investing over the past two years.

"We really found that the key for AI to work in investing is to have it admit when it doesn't know the answer, which is something that most other general-purpose AIs are not really good at." Jannick Malling, the co-chief executive of Public, told MarketWatch.

Investors don't want AI to be like the kid in class who raises their hand in response to every question, even if they don't know the answer. Public introduced Alpha, its AI investing "co-pilot" powered by ChatGPT-4, in May 2023. Alpha provides users with a language-based interface where investors can ask about specific assets, get analysis or do market screening. As Public has worked to improve Alpha, it found that AI could help motivate investors better than any other source of information.

"We've always tracked this thing called actionability rate, which is for any piece of content in the app, what's the correlation to making a trade that same day?" Malling said. "With news articles [the actionability rate] is like 2%. The social content was like 5%. With AI it's close to 40%. So it's been an order of magnitude more actionable."

(MORE TO FOLLOW) Dow Jones Newswires

June 11, 2025 07:30 ET (11:30 GMT)

MW What AI can - and can't - do for investors -2-

That finding led Public to do two things. First, it began building more AI products. This includes its Generated Assets tool launched in May, which allows investors to build custom indexes - like "publicly-traded companies led by chief executives younger than 40 years old" or "stocks unaffected by U.S. tariffs" - and benchmarks that index against the S&P 500 SPX. In the future, Public users will be able to invest in their generated indexes through fractional shares and direct indexing.

The second thing Public did was announce it would end its social feed. Public was known for its social-media-like feed that gave investors a place to talk about investing within its app. But after spending enough time experimenting with the new technology, Public decided to focus on AI instead.

"The core use-cases of the social feed used to be people posting recaps of earnings calls and bite-sized updates on what's happening in the markets on a given day. Now, this exact content is created by AI. Faster, more accurate, and woven throughout the product experience, where it's showing in the best context," the company's two co-founders wrote in a blog post.

"AI has cannibalized social," they added.

With this shift, Public is standing by the belief that in the future, AI will be more important to investors than social-media feeds. That's a notable shift from the retail investing boom in 2020 and 2021, when social media sparked a meme-stock mania and brought more people into the market.

That era saw a rise in the number of self-directed investors using platforms like Public, Moomoo and Robinhood. But Malling theorized that AI will change the way we think of self-directed investors.

"Historically, the world of investing has been very binary, split between self-directed and fully managed offerings. Do it yourself, or get someone else to do it. But with AI, there's a third path," Malling said.

-Gordon Gottsegen

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

June 11, 2025 07:30 ET (11:30 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.