Adapting to AI: How Information on Occupational AI Exposure Affects Educational Decisions
How to cite
Chuan, Amanda; Reyes, German; Serrano, Joaquin; Zhang, Hanzhe. 2026. "Adapting to AI: How Information on Occupational AI Exposure Affects Educational Decisions." AEA RCT Registry. AEARCTR-0018261. https://doi.org/10.1257/rct.18261-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
Are educational decisions affected by information about how artificial intelligence (AI) impacts future job prospects? We will run a field experiment on undergraduate students at a large state university. The treatment group will receive information about how generative artificial intelligence (AGI) has changed various occupations, while the control group will receive placebo information. Our survey will elicit self-reported choice of major, desired occupations, and beliefs regarding future job market outcomes both before and after treatment. We will link these data to administrative student record data on course selection, major declarations, and graduation outcomes. We will also link survey results to post-graduation employment data for students to examine whether treatment assignment influences employment outcomes and occupational selection.
Intervention
We will present information about AI's impact on occupations from Brynjolfsson et al. (2025) and the Anthropic Economic Index (Handa et al., 2025). This information is presented in an accessible, interactive visualization that presents detailed information about each occupation. We also provide the opportunity to converse with a chatbot trained with data and research regarding the impact of AI on various occupations.
Experimental design
The treatment condition provides information about the exposure of different occupations to generative AI through two components. First, participants interact with an interactive visualization that ranks occupations by their exposure to generative AI. Second, participants interact with an AI chatbot powered by Claude (Anthropic).
The control condition mirrors the treatment in format and interaction burden but strips out all AI-related content. The control visualization displays the same occupations as treatment, but provides information about their alphabetical grouping based on the first letter. The control chatbot acts as a generic career advisor providing education requirements and fields of study.
Design details
Cited by
No citation data available for this registration yet.