Abstract
<title>Abstract</title> <p>Artificial intelligence increasingly mediates how young people enter the workforce, with recommender systems guiding students’ educational and career decisions at the start of their working lives. Such systems are evaluated largely on predictive accuracy and user acceptance, leaving open a question central to an inclusive future of work: who benefits, across socioeconomic positions, when AI recommends careers? Drawing on Social Cognitive Career Theory and the Capability Approach, this two-study investigation examines feasibility-adjusted career fit—alignment between a recommendation and a student’s interests and abilities, adjusted for their practical capacity to pursue it. Study 1, a transparent policy simulation (20,000 profiles; five guidance policies; three labour-market scenarios), shows that accuracy-optimised recommendation maximises unadjusted fit while widening socioeconomic gaps and assigning infeasible pathways to half of bottom-quintile students; a demographic-parity constraint equalises access but not outcomes; feasibility-informed selection—by constraint-aware AI or counsellor review—produces the highest simulated feasibility-adjusted fit and fewest infeasible recommendations, but sharply reduces poorer students’ exposure to high-opportunity pathways unless paired with capability-expanding support. Study 2, a randomised experiment with 987 Indian secondary and university students, finds counsellor-reviewed AI guidance produced the highest perceived feasibility-adjusted fit (d = 0.52 versus AI-only) and the smallest socioeconomic disparity, while AI-only guidance depressed reported agency and uncertainty awareness and elevated algorithmic deference (d = 1.20); explanation partially restored calibration. AI-assisted guidance should be evaluated as infrastructure for a just school-to-work transition: by feasibility, equity, calibrated uncertainty and preserved agency, not predictive fit alone.</p>