evidoria

← Back to browse

Good practice Imported

Universitas Terbuka's GenAI Tutoring Trial Reaches 37,743 Distance-Learning Students

Indonesia · South Tangerang · See the Indonesia profile

Evidence: Quasi-experimental Top 28% 60/100 · Ask Evidence Copilot about this practice

A quasi-experimental study at Indonesia's 550,000-student Open University compared 18,713 students in ChatGPT-assisted tutorials against 19,030 in human-tutored classes, finding small but statistically significant gains in assignment scores and forum engagement.

18,713
Students, AI-assisted classes
19,030
Students, human-tutored classes
7.42
Forum posts per student, AI-assisted
7.33 p=0.046
Forum posts per student, human-tutored
206.16
Assignment score, AI-assisted
202.89 p=0.003
Assignment score, human-tutored
37,743
Total students in study

Details

Maturity
Pilot
Promoter
Universitas Terbuka (Indonesia Open University)
Period
2024-2025 (published July 2025)
Keywords
higher education, distance learning, generative AI tutoring

Context

Universitas Terbuka (Indonesia's Open University), a distance-education institution enrolling more than 550,000 students nationwide, piloted a generative-AI tutoring model built on OpenAI's ChatGPT (versions 3.5 and 4.0) integrated into its Moodle learning-management system across four courses.

Objectives

Extend low-cost, round-the-clock academic support to geographically dispersed and lower-income students who cannot access private tutoring, while testing whether GenAI-assisted tutoring matches human-tutored outcomes.

Activities

Researchers Tian Belawati and Dimas Prasetyo compared 500 GenAI-assisted tutorial classes (18,713 students) against 500 human-tutored classes (19,030 students) in a quasi-experimental design. The programme pairs AI access with explicit instruction in prompt engineering and critical evaluation of AI-generated answers.

Results

Students in AI-assisted classes posted slightly more often in discussion forums (7.42 vs 7.33 posts per student, p=0.046) and scored marginally higher on assignments (206.16 vs 202.89 points, p=0.003); both differences were statistically significant but practically small (Cliff's Delta of 0.01 and 0.015 respectively).

Conclusions

The study, published in Open Praxis in July 2025, is among the largest quasi-experimental evaluations of generative AI tutoring conducted at a single institution to date, though the measured effects were statistically significant yet practically small.

Implementation
Indicative cost
Medium (€50k–€500k) — Built on a commercial LLM API (ChatGPT 3.5/4.0) integrated into the existing Moodle LMS across four courses, rather than a bespoke AI system.
Time to results
Short (< 1 year) — Piloted 2024-2025; study published July 2025.
Staffing & skills
Universitas Terbuka tutors/faculty, researchers Tian Belawati and Dimas Prasetyo

Conditions for success

  • AI access paired with explicit instruction in prompt engineering and critical evaluation of AI-generated answers
  • Integration into the existing Moodle LMS rather than a standalone tool

Common failure modes

  • No data-protection, bias-auditing or disclosure framework for the ChatGPT integration is reported
  • Effect sizes were statistically significant but practically small (Cliff's Delta 0.01-0.015)

Where it fits

Governance type
public distance-education university
Scale
institutional, across a national archipelago student body
Income level
lower-middle income

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful