Interpreting the Stock Market Depends on How Well You Interact with your GenAI and Learn from it

Generative AI is often described as a tool that can answer almost any question. In stock markets, however, the more interesting issue is not only what GenAI can answer, but how investors learn to ask.

In our paper, "How Stock Market Participants Use Generative Artificial Intelligence: Evidence from User-Platform Interaction Data," co-authored with Frank Ecker (Frankfurt School of Finance & Management), and Fan Wu (The Chinese University of Hong Kong), we study more than 1.7 million stock-related queries submitted to one of China's largest GenAI platforms. The data show how retail investors use GenAI to research listed firms, interpret corporate information, and think about investment decisions.

We find a gradual shift. Many users begin with broad questions about market conditions, firm prospects, or stocks worth watching. Over time, their queries often become more targeted, moving toward information search, comparison, cause and impact analysis, and financial statement analysis. GenAI is therefore not only a tool that provides answers; it may also shape how investors formulate the next question.

Retail investors face a familiar challenge: there is too much information, and much of it is difficult to process. Annual reports, earnings announcements, management forecasts, analyst reports, news articles, and online discussions all contain potentially useful signals. But locating the relevant information and understanding what it means can be costly, especially for individual investors with limited time or financial expertise.

This is where GenAI appears to play an important role. In our sample, users do not use it merely as a search engine or as a simple summarization tool. They often ask it to interpret information. The most common query topics include business outlook, production and operations, financial performance, product and technology, stock market performance, and competition.

The tasks users assign to GenAI are also revealing. Information integration appears in 77.6% of queries, information acquisition in 44.5%, and information awareness in 38.8%. Because a single query can involve more than one task, these categories are not mutually exclusive. Many investors ask GenAI to explain why something happened, assess its impact, compare firms, or evaluate financial statements. These are interpretive tasks. They require connecting facts, context, and judgment.

At the same time, some useful GenAI capabilities remain underexplored. Summarization accounts for only 3.3% of queries, even though summarizing long and complex disclosures is a widely expected use case for GenAI. This suggests that many retail investors may still be experimenting with the technology, may not know how to prompt it effectively, or may not fully trust AI-generated summaries of financial information. 

A second finding is that not all users extract value from GenAI in the same way.

More engaged users, and users who more frequently refer to financial metrics, tend to ask more analytical questions. They are more likely to use GenAI for financial statement analysis, information search, cause analysis, impact analysis, and firm comparisons. Less engaged users are more likely to remain at the awareness stage, asking broader and less structured questions about firms or market conditions.

This matters because GenAI is sometimes presented as a tool that democratizes access to financial analysis. Our evidence suggests a more nuanced picture. GenAI may lower information-processing costs, but it does not eliminate differences in users' ability to ask useful questions, evaluate answers, or connect the output to investment decisions.

Still, the story is not only about unequal ability. We also find evidence consistent with learning through interaction. When early GenAI answers contain more specific information or refer to financial metrics, users' later queries are more specific and more likely to include financial metrics. For example, when earlier answers mention Return on Equity (ROE), solvency, or accounts receivable, users are more likely to use those same metrics in later queries.

"AI does not remove differences in user capability. But because the interaction is iterative, the system may also help users learn what information to ask for and how to ask for it." Xitong Li

We cannot say from this evidence that users become better investors, or that GenAI improves their trading performance. But we can say that early interactions are reflected in later information requests. GenAI may therefore function as a learning interface, directing some users' attention toward the vocabulary and structure of financial analysis.

A third important finding is that users often turn to GenAI for context.

Corporate disclosures and news reports may tell investors what happened. GenAI is often used to ask what it means. Users ask about causes, consequences, business outlook, operations, technology, competition, and financial performance. In many cases, they are not simply looking for a fact. They are trying to place that fact into a broader interpretation.

This contextual role is visible around corporate information events. Query activity increases around major disclosures, but this variation tends to parallel media coverage. This suggests that traditional information channels still matter. Investors may first become aware of a corporate event through news or other intermediaries, and then turn to GenAI to understand the implications.

We also find that the positive association between favorable management forecasts and firm-specific GenAI queries is weaker when those forecasts are longer or cover more topics. This is consistent with a substitution mechanism: when firms provide richer context upfront, investors have less need to seek additional explanation from GenAI.

In an AI-assisted information environment, disclosures are not only read by investors. They may also be processed, summarized, and reinterpreted by AI tools. Firms therefore need to think about whether their disclosures provide enough context for both human readers and AI-assisted interpretation.

A Cautious Link to Market Activity

The link between GenAI use and market activity should be interpreted carefully. We do not show that GenAI causes stock prices to move. But we do find that higher query activity is associated with wider bid-ask spreads, higher abnormal trading volume, and stronger proxies for informed trading. The sentiment embedded in GenAI answers is also positively associated with same-day abnormal returns, especially when users react positively to those answers.

These patterns suggest that GenAI is becoming intertwined with market information flows. When investors use AI-generated answers to interpret firms, compare signals, or decide what to investigate next, those interactions may become part of the broader information environment in which trading takes place.

Preparing for AI-Assisted Investing

These findings create practical challenges for firms, platforms, and regulators.

For firms, the lesson is that disclosure quality is not only about releasing numbers on time. Investors also need explanations: what changed, why it changed, and how it connects to the firm's outlook, operations, and competitive position. Clearer, more contextual, and more consistently structured disclosures may reduce the need for investors to seek external interpretation through GenAI.

For GenAI platforms, the evidence points to a concrete design lesson: more information is not always better. Users are less likely to continue engaging with the platform when answers are longer or packed with references. Platforms may therefore need to pair depth with usability, offering concise answers, clear summaries, and suggested follow-up questions that help users move from broad curiosity to more effective analysis.

"If investors increasingly encounter corporate information through AI systems, then disclosure design and platform design become connected. Firms need to provide information that can be contextualized by AI, while platforms need to recognize that users value answers that are concise, useful, and not overly dense." Xitong Li

For regulators, the issue is not simply access to AI tools. The question is whether retail investors can assess the quality of AI-generated answers. GenAI may help some users process information more effectively, but it may also amplify differences between sophisticated and less sophisticated investors. There are also risks of hallucination, factual errors, biased framing, and overreliance on answers that sound confident.

For investors, the message is equally important. GenAI can be useful for organizing information, asking follow-up questions, and interpreting complex corporate events. But it should not replace judgment. Its value may be greatest when investors use it to sharpen their questions, not when they treat it as a source of final answers.