Pedagogical Approach
Career Readiness Training uses an AI tutor that follows a method called scaffolded Socratic tutoring. The tutor guides learners with questions and hints and gives direct explanations only as a last resort. This page describes the method and the research it draws on.
Three Principles
Section titled “Three Principles”| Principle | What it means |
|---|---|
| Scaffolded Socratic tutoring | The tutor helps learners build understanding through questions, hints, and worked examples. |
| Continuous formative assessment | Every exchange is also a check of understanding. |
| Mastery-based progression | The quiz opens only after the learner has covered all topics of the lesson. |
How the Tutor Teaches
Section titled “How the Tutor Teaches”The tutor gives graduated assistance in five steps:
- Assess. It begins each topic by asking what the learner already knows.
- Guide. It asks leading questions that help the learner reason through the material.
- Hint. If the learner struggles, it gives partial information or a worked example.
- Explain. It gives a direct explanation only as a last resort.
- Fade. As the learner shows understanding, it reduces support and encourages independent reasoning.
The design responds to evidence that unrestricted access to an AI assistant can harm learning. In a randomized controlled trial with nearly 1,000 high school students, unrestricted access to a chat assistant raised practice scores by 48 percent and lowered later exam scores by 17 percent once access was removed. A tutor version with safeguards, which gave hints and withheld answers, raised practice scores by 127 percent and largely avoided the later decline (Bastani et al. 2025).
In a randomized controlled trial in a university physics course, students who used an AI tutor built on research-based teaching principles achieved learning gains more than double those of students in an active-learning class, in less time (Kestin et al. 2025). A review of recent trials concludes that AI tutoring is effective when it is pedagogically well designed, for example when it guides with questions (Burns 2026). Beale (2025) translates established learning theory into design recommendations for AI tutors, including graduated assistance and fading support.
Checking Understanding
Section titled “Checking Understanding”The tutor checks understanding throughout the conversation. It uses four techniques:
| Technique | What the tutor asks the learner to do |
|---|---|
| Explain back | Summarize a concept in their own words |
| Apply | Apply a concept to their own situation |
| Predict | Predict an outcome before the tutor gives the answer |
| Retrieve | Recall earlier topics |
The tutor moves on when the learner’s answers indicate understanding.
This practice builds on research showing that one-to-one tutoring combined with mastery learning improves achievement by about two standard deviations (Bloom 1984). Recent work shows that a learner’s knowledge can be estimated from the turns of a tutoring dialogue (Scarlatos, Baker, and Lan 2025) and recommends that AI tutors respond differently to slips, gaps in mastery, and misconceptions (Vanacore et al. 2026).
The Quiz
Section titled “The Quiz”The quiz opens once the tutor has covered every topic of the lesson and the learner has answered the checks satisfactorily. It is delivered inside the chat. The questions are fixed for each lesson, and the platform scores the answers by a fixed rule. The AI model does not grade the quiz.
The standard quizzes have 10 multiple-choice questions. The default pass mark is 70 percent and can be set for each lesson. A learner who does not pass stays in the lesson and can try again.
This is a simplified model of mastery. Bayesian knowledge tracing, which research-grade tutoring systems use, is not implemented.
Two findings inform this design. Tutoring systems that track mastery step by step are nearly as effective as human tutors, with effect sizes of 0.76 and 0.79 (VanLehn 2011). In a study of a conversational quiz format, users recalled about 20 percent more correct answers than users of a flashcard application (Ruan et al. 2019).
After the Quiz
Section titled “After the Quiz”After a learner passes, the lesson stays open in support mode. The tutor answers follow-up questions on the lesson’s topics, using the lesson content as its basis. It does not restart the lesson or repeat the quiz. Returning to the material later supports retention.
AI in Technical and Vocational Education
Section titled “AI in Technical and Vocational Education”The approach also draws on reports by international organizations on AI in technical and vocational education and training: UNESCO-UNEVOC (2021), World Bank, ILO, and UNESCO (2023), World Bank (2024), OECD (2025), and Tegelbeckers et al. (2025).
References
Section titled “References”| Authors | Year | Title | Source |
|---|---|---|---|
| Bastani, H., O. Bastani, A. Sungu, H. Ge, Ö. Kabakcı, and R. Mariman | 2025 | Generative AI without Guardrails Can Harm Learning: Evidence from High School Mathematics | Proceedings of the National Academy of Sciences 122 |
| Beale, R. | 2025 | Dialogic Pedagogy for Large Language Models | arXiv:2506.19484 |
| Bloom, B. S. | 1984 | The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring | Educational Researcher 13 (6): 4–16 |
| Burns, M. | 2026 | What the Research Shows about Generative AI in Tutoring | Brookings Institution |
| Kestin, G., K. Miller, et al. | 2025 | AI Tutoring Outperforms In-Class Active Learning: An RCT Introducing a Novel Research-Based Design in an Authentic Educational Setting | Scientific Reports 15 |
| OECD | 2025 | How Can Innovative Technologies Transform Vocational Education and Training? | OECD |
| Ruan, S., L. Jiang, J. Xu, et al. | 2019 | QuizBot: A Dialogue-Based Adaptive Learning System for Factual Knowledge | Proceedings of CHI ’19 |
| Scarlatos, A., R. S. Baker, and A. Lan | 2025 | Exploring Knowledge Tracing in Tutor-Student Dialogues | Proceedings of LAK ’25; arXiv:2409.16490 |
| Tegelbeckers, H., A. Dietrich, F. Bünning, and S. Zug | 2025 | European Insights: AI Integration in TVET | UNESCO-UNEVOC |
| UNESCO-UNEVOC | 2021 | Understanding the Impact of Artificial Intelligence on Skills Development | UNESCO-UNEVOC |
| Vanacore, K., R. S. Baker, A. H. Closser, and J. Roschelle | 2026 | The Path to Conversational AI Tutors | arXiv:2602.19303 |
| VanLehn, K. | 2011 | The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems | Educational Psychologist 46 (4): 197–221 |
| World Bank | 2024 | AI Revolution in Education: What You Need to Know | World Bank |
| World Bank, ILO, and UNESCO | 2023 | Building Better Formal TVET Systems: Principles and Practice in Low- and Middle-Income Countries | World Bank |

