Mingfeng Lin

As large language models (LLMs) become gateways to digital services, they may reshape how consumers discover sellers and how sellers compete. We study the economic impact of this shift by examining ChatGPT's February 2026 integration with OpenTable, an online restaurant reservation platform. Using daily reservation availability data from January 3 to April 18, 2026, we compare restaurants on OpenTable and Resy, a comparable non-integrated platform, in a difference-indifferences design. We find no significant effect on average demand for OpenTable restaurants, but substantial redistribution across restaurants. Restaurants with more cumulative reviews, typically a positive signal in human-driven discovery, experience demand losses relative to counterparts unaffected by the integration. In contrast, restaurants with textually richer reviews accessible to AI experience larger demand gains. Average ratings do not significantly moderate these effects. We further estimate restaurant-level heterogeneous treatment effects using a doubly robust estimator and find larger effects for restaurants with more distinctive characteristics and dining goals. Evidence on mechanisms reveals changes in both consumer search and restaurant recommendations: consumers express more free-form, intent-rich search criteria when interacting with AI, while AI surfaces restaurants differently from the platform's conventional recommendation algorithms. Our findings suggest that AI-mediated search can change the basis of competition among sellers by reshaping which information guides discovery and how demand is allocated.

Online labor platforms often introduce new features to help employers identify reliable workers. This study examines one such feature: a worker-offered guarantee, in which freelancers voluntarily deposit their own money with a bid, with forfeited funds donated to charity rather than paid to the employer. Although such guarantees are intended to signal confidence and commitment, our analysis of bid-level data from a major freelance platform shows that employers interpret them differently. Workers with medium reputations are most likely to offer guarantees, and employers are less likely to hire workers who use them. This skepticism appears economically justified: guarantee-offering workers receive lower post-contract ratings on average. The findings suggest that well-intentioned platform tools can backfire when users see them as signs of weakness or desperation rather than quality. For platform managers and policymakers interested in digital labor markets, the study highlights the need to design signals with credible costs, clear enforcement, and alignment with existing reputation systems. New marketplace features should be evaluated not only by whether users adopt them, but also by how other market participants interpret and act on them.

As entrepreneurship education has become increasingly prevalent in higher education, its broader consequences for students' career development have attracted growing attention. Supporters argue that entrepreneurship programs generate long-term value by cultivating transferable skills that benefit participants beyond venture creation. Critics, however, contend that such programs may crowd out other educational investments and potentially undermine academic and labor market outcomes. Despite ongoing debate over its net impact, existing research has focused primarily on whether entrepreneurship education increases entrepreneurial activity, leaving its effects on broader career trajectories underexplored. Drawing on ten years of longitudinal data from a university startup incubator, this study examines the impact of an entrepreneurship education program by comparing the post-graduation career trajectories of accepted and rejected applicants. The results show that exposure to entrepreneurship education does not necessarily increase entrepreneurial activity immediately after graduation. However, participation is associated with sustained career advantages, even among students who do not pursue entrepreneurship in the short term. Specifically, participants demonstrate better job matching, as reflected in higher earnings, longer tenure with their initial employers, and greater stability in occupational type. These findings provide empirical evidence that entrepreneurship education can generate enduring positive returns in the broader labor market. They also offer important implications for universities and organizations evaluating whether to invest in entrepreneurship programs, suggesting that the value of such programs extends beyond the creation of new ventures.

As AI providers increasingly explore advertising within generative AI systems, they face a fundamental tension between monetization and users' adoption. This research examines how, when, and why advertising disclosure in generative AI responses influences users' trust in the AI system. We argue that advertising disclosure can undermine users' trust by reducing perceived AI neutrality, and this negative effect depends on the disclosure format.When disclosure is attached to the target product that users explicitly seek in a generative AI query, it reduces both trust in the AI system and trust in the advertised product. In contrast, when disclosure is attached to a non-target product presented after an organic recommendation for the target product, it decreases trust in the advertised recommendation but does not significantly diminish trust in the AI model overall. Our claims are supported by results from three laboratory experiments. Taken together, our findings offer theoretical insights into trust formation in AI-mediated decisionmaking and provide practical guidance for the design and monetization using advertising of generative AI systems.

Our study examines how the value of skill certifications changes when an online labor platform eliminates testing fees. Although the policy led to a sharp increase in certification rates and faster hiring, it also reduced the signaling power of certifications. Employers became less likely to hire certified workers or offer them higher pay because their average performance declined. This drop in credibility disproportionately affected inexperienced workers who depended on certifications to establish trust. Importantly, we find that the negative effects of free certification can be reduced through better platform design. Showing more detailed certification information, such as test scores and how recently the certification was earned, helps employers interpret credentials more effectively. These design improvements can help preserve the value of certifications while expanding access.

Despite the popularity of the phrase “wisdom of the crowd,” not all crowds are wise because not everyone in them acts in an informed, rational manner. Identifying informative actions, therefore, can help to isolate the truly wise part of a crowd. Motivated by this idea, we evaluate the informational value of investors’ bids using data from online, debt-based crowdfunding, in which we were able to track both investment decisions and ultimate repayment statuses for individual loans. We propose several easily scalable variables derived from the heterogeneity of investors’ bids in terms of size and timing. We first show that loans funded with larger bids relative to the typical bid amount in the market or to the bidder’s historical baseline, particularly early in the bidding period, are less likely to default. More importantly, we perform theory-driven feature engineering and find that these variables improve the predictive performance of state-of-the-art models that have been proposed in this context. Even during the fundraising process, these variables improve predictions of both funding likelihood and loan quality. We discuss the implications of these variables, including loan pricing in secondary markets, crowd wisdom in different market mechanisms, and financial inclusion. Crowdfunding platforms can easily implement these variables to improve market efficiency without compromising investor privacy. This paper was accepted by David Simchi-Levi, information systems. Funding: We gratefully acknowledge the financial support of the Business Analytics Center in the Scheller College of Business and the Ewing Marion Kauffman Foundation. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2022.02575 .

On decentralized e-commerce platforms, search algorithms play a critical role in matching buyers and sellers. A typical search algorithm routinely refines and improves its catalog of data to increase search precision, but the effects of a more precise search are little known. We evaluate such effects via a 2019 quasiexperiment on a world-leading e-commerce platform in which the search algorithm refined some product categories into finer subgroups to allocate consumer queries to more relevant product listings. Our data cover millions of consumers’ search and purchase behaviors over six months across multiple search sessions and product categories, enabling us to investigate trade-offs over time and across categories. We find that a more precise search algorithm improves consumers’ click-through and purchase rates drastically and instantaneously, but it comes at the cost of a significant decrease in consumer engagement and unplanned purchases over a longer time horizon. On average, consumers who used to spend more time searching now conduct 5.5% fewer searches, spend 4.1% less time on the platform, and decrease their spending on related categories by 2.2% in the week after the search precision increases. Our examination of the mechanisms behind these consequences calls for more careful search algorithm designs that account for not only instant conversion based on search precision but also consumer engagement and sellers’ strategic responses in the longer horizon. This paper was accepted by Anindya Ghose, information systems. Funding: This work was supported by the Key Program of the National Natural Science Foundation of China [Grants 72141305 and 72192803] and the Ministry of Education, People’s Republic of China (Fundamental Research Funds for the Central Universities), both to L. Fang. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2023.00478 .

Online labor markets routinely employ auction mechanisms to assist the project providers (i.e., buyers or employers) in recruiting eligible participants (i.e., sellers or workers) who have placed bids in either sealed or open auctions. In a sealed auction, participants know only the number of competitors, but not their profiles and bidding amounts. Conversely, in an open auction, all this information is public. To understand which auction format results in more optimal allocation (i.e., effectively matching buyers and sellers) and greater time efficiency, we build a theoretical model to generate a series of hypotheses, then test these hypotheses using data from an abrupt change of auction format in an online labor market. Our findings indicate that in sealed auctions, where sellers cannot observe their competitors' bids, they tend to place lower bids and bid more quickly than sellers in open auctions. Additionally, the effect of auction format switching is less pronounced for experienced sellers compared to new sellers. The empirical results align with our theoretical explanation: In open-bid auctions, sellers have strong incentives to delay bidding or place initial bids higher than their valuation to learn from competitors' bids and adjust accordingly (if there is a chance). This strategic bidding behavior can lead to misallocation and time inefficiency. Specifically, while open auctions attract more participants, buyers take longer to make hiring decisions, and the likelihood of hiring decreases. Furthermore, sealed auctions result in better project completion and fewer seller exits.

Signaling mechanisms are prevalent in two-sided markets because they typically help reduce information asymmetry and improve efficiency. While a growing number of studies focus on the actions of signal-senders, there has been little empirical research on how signal-receivers may react to the signals, despite the importance of this question. A well-intentioned signal could backfire, or "boomerang," if the receiver does not react the way that its sender or designer intends. The introduction of a novel project-specific, voluntary guarantee-deposit ("guarantee" for short) in an online labor market provides an excellent opportunity to answer this question. Leveraging detailed transaction data, we study how employers (signal-receivers) react to the presence guarantees in the bids placed by workers (signal-senders). We find strong evidence that, despite the platform's advocacy for the benefits of providing a guarantee, employers are less likely to award contracts to workers who use this signaling mechanism than those who do not. This is particularly prominent when uncertainty about the project or worker is high. Only workers who already possess high customer ratings improve their chances of winning contracts by offering such a guarantee. These results indicate that employers do not just ignore this novel signal, but use it to "counter-screen" workers. In other words, this well-intentioned "signal" boomerangs (or backfires) for most workers who need them the most. Remarkably, such employers' reactions seem economically justified: Workers who offer guarantees are less likely to perform well on the job compared to those who do not. The level of sophistication the employers demonstrate regarding the signaling mechanism in online labor markets is striking. Our findings provide fresh nuances to the signaling literature in two-sided markets and have implications for platform designs.

Despite the growing popularity of online public funding for education (i.e., crowdfunding), controversy persists about whether teachers’ efforts to use this channel are justified and whether donations thus received can actually make a difference in students’ academic performance. We present the first empirical evidence of the positive influence of online education crowdfunding after teachers successfully raise funds. Using data from California public school teachers and their students, as well as the teachers’ fundraising activities on DonorsChoose.org , we find significant academic performance improvements among students after their teachers receive donations through this platform. Such improvements in students’ academic performance are stronger when more local donors participate in teachers’ fundraising campaigns. Notably, teachers who attempt to raise funds online but are unsuccessful are still able to improve their students’ performance. These findings not only document the positive impact of online education crowdfunding on students’ academic performance, but also demonstrate that such impacts are at least partially attributable to the nonfinancial motivational benefits on teachers from receiving such donations. Our study therefore contributes to the literature on crowdfunding, as well as the ongoing debates surrounding the financing of public education.