The Effect of Beliefs on Generative-AI Adoption by Entrepreneurs

How to cite

Ronconi, Juan Pedro; Balmaceda, Marcos. 2025. "The Effect of Beliefs on Generative-AI Adoption by Entrepreneurs." AEA RCT Registry. AEARCTR-0016544. https://doi.org/10.1257/rct.16544-1.0

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

Juan Pedro Ronconi* , Marcos Balmaceda

* Primary investigator

Registry ID
AEARCTR-0016544
DOI
https://doi.org/10.1257/rct.16544-1.0
First registered
Last updated
Status
In development
Trial dates
2025-08-11 – 2026-04-01
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/16544

Abstract

Generative AI could help vulnerable entrepreneurs overcome information disadvantages, yet adoption remains minimal. Among Chilean microentrepreneurs—most without higher education or formal business training—AI tools remain largely unused despite their potential to democratize access to business expertise. We test whether beliefs about AI benefits explain this adoption gap through a randomized controlled trial. We randomize entrepreneurs into receiving information on the potential of GenAI as a business consultant or as a content creator, with a control group receiving neutral technical information. Through incentivized willingness-to-pay elicitation, we measure how information affects beliefs about GenAI's business value. We evaluate the impact of exogenously changing beliefs on GenAI take-up and on firm performance. Results inform whether low-cost information interventions can help reduce inequality in entrepreneurial success.

Intervention

We implement a randomized information experiment with two treatment arms and one control group to test how different framings of AI capabilities affect entrepreneurial beliefs and adoption intentions.

The intervention consists of brief (2-minute) information treatments delivered through an online survey. Each treatment provides information about generative AI applications for business, while the control receives neutral technical information without actionable business applications.

Experimental design

We partner with ChileConverge, a platform that offers online training for entrepreneurs in Chile. Entrepreneurs in their database are invited to complete a survey, with a raffle to incentivize participation. We stratify the sample into thirds based on baseline characteristics available in our partner’s database (e.g., gender, education level, and business sector) to improve balance across treatment arms. Random assignment occurs within strata, with equal probability allocation (1/3 to T1, 1/3 to T2, 1/3 to Control). Treatments are administered immediately after baseline measures, within the same survey session. Follow-up surveys are conducted 2 and 6 months after baseline.

Design details

Randomization method
Stratified randomization in Stata.
Randomization unit
Randomization is done at the individual level.
Clusters
n/a
Observations
Minimum of 900 observations. We will continue collecting observations until one of the following conditions is satisfied: (i) We collect 10,000 observations; or (ii) Four weeks of data collection have passed. Therefore, final N will be between 900 and 10,000.
Arms
One third assigned to T1, one third to T2, one third to control.
Primary outcomes
We consider different measures of GenAI adoption depending on the survey round. At baseline (immediately post-treatment): Interest in adopting GenAI measured through (i) behavioral outcome: signing up to receive information about the AI online course offered by our partner; and (ii) self-reported likelihood of using or learning more about ChatGPT for their business in the next two weeks on a 1-7 Likert scale. At midline (2 months) and endline (6 months): (i) behavioral outcome: indicator for signing up for the AI online course offered by our partner; (ii) self-reported use of GenAI, including extensive margin (any use) and intensive margin (frequency of use); and (iii) types of business applications implemented.
Secondary outcomes
The first stage in our empirical model corresponds to the impact of the treatments on entrepreneurs' beliefs about the benefits of using GenAI. We will measure this through an index that combines questions on how entrepreneurs expect their income and profits to change if they adopted GenAI. We will also elicit willingness to pay (WTP) for a tutoring session on how to use GenAI to improve business operations. This measure is also related to the belief about the benefits of adopting GenAI and has the advantage of being incentivized. Moreover, besides GenAI take-up, as a secondary outcome we will also examine the impact of our treatments on revenue and profits at midline and endline.

Cited by

No citation data available for this registration yet.

Keywords