The Best of Both Worlds? Can AI Agents Close the Digital Divide in School Enrollment

Governments increasingly deliver public services through digital platforms embedding procedural frictions—passwords, document uploads, multi-step portals—that disproportionately burden vulnerable families. Arteaga, Kapor, Neilson, and Zimmerman (QJE 2022; AER:Insights 2026) show that addressing navigational frictions on Chile's school-choice platform produces test-score gains of 0.36–0.45 SD persisting five years, at $0.33 per student. In developing countries, where digital literacy gaps are larger, such frictions risk widening the inequalities digitization was meant to reduce. The researchers study post-assignment school enrollment, where families must navigate a digital portal to convert an admission offer into a seat. In Chile's most vulnerable public schools, 15–20 percent of admitted students never complete this step. The researchers conduct a three-arm randomized evaluation among ~1,000 offers, randomly assigning families to (i) status quo (email only), (ii) nudges (WhatsApp reminders plus robocall), or (iii) an AI agent (conversational voice AI and interactive WhatsApp that identifies missing steps, sends magic links, and guides enrollment). The design tests whether the binding constraint is salience—where nudges suffice—or procedural complexity, requiring support that only AI can provide at scale. The evaluation will generate causal evidence on whether AI can close access gaps in digital service delivery, with implications for platform design in developing countries.