How Much Are AI Chatbots Worth to Students? A Cross-Country Discrete Choice Experiment on Preferences and Future Expectations

How to cite

Blasco, Andrea; Benedetti, Margherita. 2026. "How Much Are AI Chatbots Worth to Students? A Cross-Country Discrete Choice Experiment on Preferences and Future Expectations." AEA RCT Registry. AEARCTR-0018605. https://doi.org/10.1257/rct.18605-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Andrea Blasco* , Margherita Benedetti

* Primary investigator

Registry ID
AEARCTR-0018605
DOI
https://doi.org/10.1257/rct.18605-1.0
First registered
Last updated
Status
In development
Trial dates
2026-05-02 – 2026-05-22
Countries
Bulgaria (), Germany (), France (), Italy (), Portugal (), Sweden ()
Registry record
http://www.socialscienceregistry.org/trials/18605

Abstract

This study examines how information about the labour-market relevance of artificial intelligence affects students’ preferences for AI chatbot services, intended use of AI for studying, and expectations about future careers. The study is conducted as an online randomised survey experiment with an embedded discrete choice experiment among 2,310 students aged 18-25 in six European countries. Participants receive information on either the labour-market value of AI skills or the extent to which occupations can be performed or supported by generative AI. The discrete choice experiment elicits preferences for AI chatbot subscription plans that vary in price, accuracy, privacy, usage limits, task domain, and image-generation features. Primary outcomes include chatbot choices, intended study use, perceptions of AI tools, career expectations, employment anxiety, and interest in AI-skills training. The study estimates the causal effects of information provision on these outcomes and on the valuation of chatbot attributes.

Intervention

The survey includes two information provisions. The first provides information on the labor-market value of AI skills; the second provides information on the extent to which tasks in future occupations can be performed or supported by generative AI. The DCE elicits preferences over AI chatbot subscription plans that vary in monthly price, usage limits, accuracy, data storage/privacy, task domain, and image-generation limits.

Experimental design

Eligible respondents will be randomly assigned at the individual level to two independent information provisions (no blocks or clusters). The first provides information on the labour-market value of AI skills; the second provides information on the exposure of occupational tasks to generative AI. Each provision is assigned with probability 0.5, creating four experimental conditions: no information, AI-skills information only, AI-task-exposure information only, and both information provisions.

The DCE attributes were randomly combined into paired A/B alternatives using the Efficient Design Tool (EDT), a Python-based framework for constructing D-efficient discrete choice experiment designs with prior parameter estimates (Hernández 2024). EDT applies a random-swapping optimisation algorithm (Quan et al. 2011) to generate statistically efficient choice sets while minimising the variance of the estimated coefficients. The resulting design enabled the construction of balanced and efficient experimental alternatives for the study.

Design details

Randomization method
Completely at random indpendently for each treatment; random number generator.
Randomization unit
individual respondent
Clusters
no clusters
Observations
Expected respondents 2,310 (equally split across countries)
Arms
approximately 1,155
Primary outcomes
For the first intervention, we consider the following primary oucomes: the eight DCE subscription-plan choices (question: Q8). In the respondent-level file, each task records whether the respondent chooses option A, option B, or Neither. The key outcomes are (1) the willingness to pay for the AI chatbot plans and (2) the preference weights of the randomized plan attributes: price, usage limits, accuracy, data storage/privacy, task domain, and image-generation limits. We will also test the effect of the intervention on intention to use AI chatbots for studying in the future (Q12); employment anxiety (Q21) and interest in AI training (Q23) For the second intervention, we consider the following primary outcomes: Career-related measures of the perceived impact of AI on their ideal job collected after the second information provision (Q19); Confidence that their skills will match future demand of labour (Q20); employment anxiety (Q21); and interest in AI training (Q23).

Cited by

No citation data available for this registration yet.

Keywords