Prompting Opportunity: AI Guidance, Student-Aid Beliefs, and Educational Aspirations

We study a two-arm, individually randomized online experiment. All participants receive access to the same general AI career chatbot and see a common, persistent reminder (i.e., nudge) at the top of the bot window emphasizing the importance of informing oneself about educational pathways to promote general usage. The treatment arm additionally receives proactive, salient information about BAfoeG (i.e., student aid) and a route to a BAfoeG calculation module. The control arm receives no proactive finance messages or treatment-specific call to action.

2026-09-21

2026-10-31

1. BAfoeG beliefs (value and weighted index of all components; first stage).
2. Intention to enter higher education (downstream aspiration outcome).

We elicit all at baseline and 14 days after their initial sign-up and usage of the chatbot.

1. Components of the BAfoeG belief: perceived eligibility, expected monthly amount, and expected repayment.
2. Subjective educational information and preparation effort during the 14 days after baseline.
3. Intention to move out of the current parental household for the next educational pathway.
4. Preferred destination for the next educational pathway and distance from the current or last school to that destination.
5. Chatbot engagement and treatment fidelity.
6. Follow-up response and attrition.

Two-arm individual randomized online experiment with a common active-control chatbot. Eligible participants complete the same baseline questionnaire before treatment-specific content is activated. Participants are assigned approximately equally to control or proactive study-finance treatment (pure randomization). The primary follow-up is administered 14 days after baseline, with up to seven days to respond. We have a rolling sign-up in the treatment window, i.e. after the summer break, such that the exact date might vary. As we will randomize on the individual levels to ensure power, we will report diagnostics regarding spillovers.

Not available

Randomization will be conducted in blocks of two based on order of enrollment. For each new block, the first participant will be independently assigned to the treatment condition with probability 0.5, and the second participant will receive the opposite assignment. This ensures a 1:1 allocation within complete blocks prior to attrition. The purpose of this procedure is to ensure balanced group sizes at assignment and to allow for potential efficiency gains.

Participants with an observed outcome will not be excluded from the primary analysis solely because the outcome of their paired participant is missing or because their randomization block is incomplete at the end of recruitment. All other preregistered inclusion and exclusion criteria remain unchanged.

If, within each role, the probability of observing the outcome is the same in both treatment arms, expected group sizes will also remain equal after attrition. However, exact 1:1 balance after attrition is not guaranteed.

Individual - To be precise: Individuals before their career decision (we further allow the usage for parents of individuals for career decisions; this is not our main unit of interest, but rather a pilot for future studies.. We will report potential findings transparently in the appendix, though not consider them for the main analysis and, thus, power calculation or randomization).

No

NA.

2,000 pupils (baseline) - 1000 blocks of size 2

Given LLN, we will converge to a 50/50 share, hence 1,000 pupils for both treatment arms

With 2,000 analyzed individuals, equal assignment, a two-sided alpha of 0.05, and 80 percent power, the approximate unadjusted minimum detectable effect is 0.125 standard deviations for a continuous outcome and 6.3 percentage points for a binary outcome with a control mean of 0.50. If 80 percent of baseline completers provide the day-14 outcome, analyzed N is approximately 1,600 and the corresponding unadjusted continuous-outcome MDE is approximately 0.140 standard deviations. If only 80 percent belong to the confirmatory population and 80 percent of that group respond, analyzed N is approximately 1,280 and the MDE is approximately 0.157 standard deviations, or 7.8 percentage points for a binary outcome with a 0.50 control mean.