Generating Solutions for Society
Prompt Adaptation as a Dynamic Complement in Generative AI Systems
As generative AI systems rapidly improve, a key question emerges: How do users keep up—and what happens if they fail to do so. Drawing on theories of dynamic capabilities and IT complements, we examine prompt adaptation—the adjustments users make to their inputs in response to evolving model behavior—as a mechanism that helps determine whether technical advances translate into realized economic value. In a preregistered online experiment with 1,893 participants, who submitted over 18,000 prompts and generated more than 300,000 images, users attempted to replicate a target image in 10 tries using one of three randomly assigned models: DALL-E 2, DALL-E 3, or DALL-E 3 with automated prompt rewriting. We find that users
with access to DALL-E 3 achieved higher image similarity than those with DALL-E 2—but only about half of this gain (51%) came from the model itself. The other half (49%) resulted from users adapting their prompts in response to the model’s capabilities. This adaptation emerged across the skill distribution, was driven by trial-and-error, and could not be replicated by automated prompt rewriting, which erased 58% of the performance improvement associated with DALL-E 3. Our findings position prompt adaptation as a dynamic complement to generative AI—and suggest that without it, a substantial share of the economic value created when models advance may go unrealized.
Eaman Jahani, UMD
Benjamin Manning, MIT
Hong-Yi TuYe, MIT
Mohammed Alsobay, MIT
Christos Nicolaides, University of Cyprus
Siddharth Suri, Microsoft Research
David Holtz, Columbia
Tracking-Based Advertising After Apple's App Tracking Transparency: Firm-Level Evidence and Policy Implications
TechREG CHRONICLE, November 2025
We discuss the impact of Apple’s App Tracking Transparency's (“ATT”) on targeted, online advertising. We overview the empirical results of Aridor, Che, Hollenbeck, Kaiser & McCarthy (2025) that measured the impact of ATT on e-commerce firms. The results point to a large reduction in the efficacy of targeted advertising and subsequently large revenue losses, borne primarily by smaller firms. We discuss the competition policy implications of this by highlighting the potentially anticompetitive implications of privacy measures implemented by private firms and the lack of substitutability between advertising networks, despite a large exogenous shock in the efficacy of Meta advertising.
Aridor, Guy: Assistant Professor of Marketing, Northwestern University
Hollenbeck, Brett: Associate Professor of Marketing, UCLA
McCarthy, Daniel: Associate Professor of Marketing, University of Maryland
AdGazer: Improving Contextual Advertising with Theory-Informed Machine Learning
Journal of Marketing
Contextual advertising involves matching features of ads to features of the media context where they appear. We propose AdGazer, a new machine learning procedure to support contextual advertising. It comprises a theoretical framework organizing high and low-level features of ads and contexts, feature engineering models grounded in this framework, an XGBoost model predicting ad and brand attention, and an algorithm optimally assigning ads to contexts. AdGazer includes a Multimodal Large Language Model to extract high-level topics predicting the ad-context match. Our research uses a unique eye-tracking database containing 3531 digital display ads and their contexts, and aggregate ad and brand gaze times. We compare AdGazer’s predictive performance to two feature learning models, VGG16 and ResNet50. AdGazer predicts highly accurately with hold-out correlations of 0.83 for ad gaze and 0.80 for brand gaze, outperforming both feature learning models and generalizing better to out-of-distribution ads. Context features jointly contributed at least 33% to predicted ad gaze and about 20% to predicted brand gaze, good news for managers practicing or considering contextual advertising. We demonstrate that the theory-informed AdGazer effectively matches ads to advertising vehicles and their contexts, optimizing ad gaze more than current practice and alternatives like text-based and native contextual advertising.
Michel Wede (UMD Smith) Jianping Ye (UMD PhD student); and Rik Pieters (Tilburg University, the Netherlands)