AI summaries in the financial market were met with skepticism for a pretty simple reason: a lot of people believed that condensing a complex news story into three bullet points would be the fastest path for an uninformed reader to make bad decisions with their own money.
That line of thinking makes sense, right?
But a study from Harvard Business School turned that logic completely upside down.
Joseph Pacelli, associate professor at the institution and Gerald Schuster Professor of Business Administration, investigated what actually happens when readers get access to AI-generated summaries before diving into a financial news article, and the result was quite different from what critics expected.
Instead of creating shortcuts that replace a full read, the summaries worked like a trail of breadcrumbs guiding the reader deeper and deeper into the story, boosting both comprehension and engagement with the complex information in the articles.
And the effects went beyond reader behavior: the financial market also responded differently to articles that had AI summaries, with faster trades and prices adjusting more efficiently. 📊
Let’s break down how this research was conducted, what it found, and what it changes about the way we think about AI and financial information.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
How the research was designed
The study led by Pacelli was no surface-level data analysis or opinion survey. The work combined two approaches that complement each other in a really interesting way. The first was an analysis of real market data, looking at a full year of articles published by The Wall Street Journal. The second was a controlled lab experiment with real readers, distributing financial articles with and without AI-generated summaries to separate groups of participants.
The goal was to understand, empirically, whether these summaries changed the way people consumed and interpreted information, not just how much they thought they understood. That distinction matters a lot, because self-perceived comprehension and actual comprehension are very different things, and confusing the two would be a serious methodological mistake.
The Wall Street Journal started attaching AI summaries to some of its articles in 2024, which created near-perfect conditions for a natural experiment. The researchers identified 1,734 articles containing financial analyses of 158 companies, published between July 2024 and June 2025. From there, they examined the effect of each article on each company’s stock in the 30 minutes following its online publication.
The paper was co-authored with Tony Cho and Allen Huang, professors at the Hong Kong University of Science and Technology, and Kristina Rennekamp, a professor at Cornell University. The scientific paper, titled Generative AI and Investor Processing of Financial Media, was published as a working paper from Harvard Business School.
The most interesting aspect of the experimental design was precisely what the researchers did not assume from the start. Rather than presupposing that summaries would be harmful or beneficial, the team let the data speak. As Pacelli himself admits, he went into the research thinking AI summaries might not be all that useful, leading to quick, shallow reads. And the data said something that contradicts the conventional wisdom held by a good chunk of the financial publishing world: readers who had access to AI summaries read more, understood more, and engaged in a more meaningful way with the full articles than those who jumped straight into the text with no prior framing.
Financial summaries boost market efficiency
When investigating stock behavior after publication, the researchers found a pretty concrete result: companies mentioned in articles that had AI summaries saw a 3.5% increase in trading volume. That alone is a strong sign that readers were using these summaries to make buy and sell decisions more quickly.
But the team went further. They developed their own generative AI algorithm to analyze the tone of different paragraphs, separating those that appeared in the summaries from those that were only in the body of the article. What they discovered was revealing:
- The market reacted with stronger and more immediate trading volume to information highlighted in the AI summaries, suggesting investors were using those summaries to make decisions.
- Articles with summaries also caused prices to adjust more quickly than articles without this feature.
According to Pacelli, this indicates that traders are being informed faster, which creates more efficient markets. And there is an important detail: the effects were larger when articles were more complex or when there were many news items and earnings announcements competing for attention. In other words, summaries are especially valuable at exactly the moments when people are overwhelmed with information and need to keep up with a hectic routine. 📈
Like an appetizer before the main course
To better understand how readers process these summaries, Pacelli and his colleagues ran a lab experiment with 124 graduate students. Each participant received two identical Wall Street Journal articles, one with an AI summary and one without, and then answered questions to test how much they remembered of the content.
The numbers were clear enough to tear down the narrative that AI summaries act as lazy substitutes for reading:
- Readers with the summary remembered more of the article overall. They correctly answered 4.72 out of 7 questions about general content and 2.49 out of 4 questions about information that appeared only in the body of the article.
- Readers without the summary remembered less. They correctly answered 3.95 out of 7 questions overall and 2.05 out of 4 on information that was not in the summary.
Notice a fascinating detail in those numbers: the summaries improved recall not only of the information they contained, but also of information that appeared only in the body of the article. This shows the summary acted as a cognitive map that helped readers know what to expect from the text, which points deserved more attention, and how to connect information throughout the read. The mental friction many people feel when facing dense financial news, packed with jargon and numbers, decreased noticeably.
And there is more: the summaries frequently encouraged investors to go deeper. Google searches for information about a company increased for articles with AI summaries, suggesting people were seeking additional information from other sources. As Pacelli puts it perfectly, the summaries do not lead to less engagement but rather to more. It is like an appetizer that whets your appetite so you stick around for the main course. 🧠
An important caveat
Despite all the optimism in the results, Pacelli himself makes a point of raising a flag. Generally speaking, generative AI summaries seem to work as they should, helping readers manage cognitive load while filtering through the daily bombardment of information to focus on what matters. But it is essential to consider the context and the type of content the AI is summarizing.
The thing is, financial news readers have a huge incentive to seek quality information, after all, real money is on the line in the companies they invest in. It is not clear whether the same would happen with news on topics where the reader has no direct stake, which could lead to more superficial analysis.
As the professor puts it very directly, it takes a lot more than three bullet points to truly understand something like inflation. If things keep moving in this direction without care, there is a risk that readers will lose the ability to appreciate the richness and complexity of certain subjects. That is a warning that extends well beyond the financial market.
How to use AI summaries in practice
Thinking about what media companies and other businesses should take from this research, Pacelli offers three very practical recommendations:
- Consider using generative AI to summarize information released to the public. Everyone is already using their own AI to summarize content. If your company does not do it, the risk is that someone else will do it for you, and maybe in a way that carries their own biases.
- Evaluate the use of AI summaries for complicated materials. Summaries prove most valuable precisely when the audience faces information overload or particularly complex content.
- Put people in charge of reviewing AI summaries. Since AI-generated summaries can increase the risk of misinformation due to biases or hallucinations, they require rigorous fact-checking and review.
What this changes about how we think about AI and information
Pacelli’s study is not just another piece of research about artificial intelligence. It represents an important turning point in the conversation about the role of AI in mediating complex information. For a long time, the debate was stuck between two extremes: on one side, the enthusiasts who believed AI would solve any communication problem with enough automation; on the other, the skeptics who saw every algorithm-generated summary as a threat to the depth and nuance that financial journalism demands. The research showed that neither side was entirely right.
AI, when applied thoughtfully and in context, can be an amplifier of human understanding, not a replacement for it. The difference lies in how the feature is positioned within the reading experience. A summary that competes with the article, suggesting the reader does not need to go beyond those three lines, has a potentially negative effect. But a summary that works as preparation, as an introduction that creates context and curiosity, has the opposite effect: it increases engagement and improves the quality of information consumption as a whole.
For the media industry, especially outlets covering finance, investing, and economics, this distinction is critical. The question shifts from whether to use AI summaries to how to use those summaries in a way that enhances the reading experience rather than ending it. And that is a much more interesting question, because it puts technology in service of the reader’s journey instead of as a shortcut that compromises it. The Harvard Business School research delivered a solid, evidence-based argument for anyone who wants to have that conversation seriously. 🤖
