Breadcrumb
- Home
- AI Hub For Education
- Research Study Repository
- Outcomes – Differentiation
Search and Filter
The Repository
Outcomes – Differentiation
Research synthesis is AI-generated, human reviewed. Updated 11/2026.
Displaying 1 - 30 of 822
Calibrating LLM Tutor Initiation to Learner Expertise: Proactive and Reactive Support for Computational Thinking
Arun Rai, Liwei Chen, Balasubramaniam Ramesh, Cynthia Breazeal, Yuan Long, Benjamin M. Ampel, Andrea Aria. (11/2026). SSRN.
Overall, there was no average treatment effect on computational thinking across all 90 students.
Not Faster, But Better: Generative AI And Teacher Skill In Peru
Carla Z. Glave, Carolina Lopez, Ezequiel Molina, Roman Andres Zarate. (10/2026). World Bank.
Providing AI licenses, training, and follow-up sharply increased AI adoption but saved teachers no time on any of their core tasks.
Teacher-Mediated Multimodal Platform And Math Achievement: Evidence From Goblins' Yearlong District Implementation
Anika Alam, Nicolas Dominguez-Mel, Jing Liu. (10/2026). EdWorking Papers.
Using the Goblins platform is associated with a 0.11 standard deviation increase in MAP math scores within the district.
The Effects Of Course-Integrated AI Tutoring On Student Performance And Engagement: A Randomized University Trial
Jing Liu, Tracy Sweet, Marty Haoyuan Chen, Jonathan Engelberg, Megan C. Masters, Martyn Clark, Axel Persaud, Alia Lancaster, Jeffrey Hollingsworth, Jennifer King Rice. (10/2026). EdWorking Papers.
Among sections of the same course, access to the AI tutor reduced final grades by 0.37 standard deviations and learning management system participation by 0.90 standard deviations.
How University Disability Services Professionals Write Image Descriptions For Hci Figures Using Generative AI
Muhammad Raees, Yugo Iwamoto, Konstantinos Papangelis, Jamison Heard, Garreth W. Tigwell. (09/2026). arXiv.
HCI experts rated AI-assisted figure descriptions significantly higher than human-only descriptions across accuracy, completeness, clarity, effectiveness, and descriptiveness.
When Evaluators Cry Wolf: Lessons From Production LLM-As-Judge Evaluation In Educational AI
Chris Rohlfs, Rodrigo Vergara Bosse, Priscilla Rain Hopper, Keanon O'Keefe, Patrick Russell. (09/2026). arXiv.
Across 21 deployed evaluators, the calibration strategy cut confirmed false-positive flags by 99% and raised per-flag precision from 0.6% to 48.9%.
Fairness Theatre: Evaluating Post-Hoc Fairness Interventions In Vendor-Controlled Early Warning Systems
Kelly McConvey, Angelina Zhai, Rebecca Li, Shion Guha. (09/2026). arXiv.
Post-hoc fairness interventions redistributed disparities without consistently reducing them, leaving error burdens for marginalized groups intact or worse while dashboard metrics appeared to improve.
Critical Thinking With Generative AI: A Constraint-First Design Pilot Of A Thinking-Partner Intervention
Fatima Tuz Zahra, Jiangen He, David Matthew Bowers, Wei Wang. (09/2026). arXiv.
Undergraduate students showed significant gains on all ten statistics learning dimensions and on eight of nine AI literacy dimensions after the semester-long intervention.
Characterizing LLM-Based Family Education Through The Lens Of Activity Theory: A Scoping Review Of The Hci Literature
Lan Luo, Yuqi Liang, Jie Cai, Anqi Wang, Dongyijie Primo Pan, Muzhi Zhou, Chun Yu, Pan Hui. (09/2026). arXiv.
Most reviewed studies focused on child–parent interaction (83%) and centered on language, narrative, AI literacy, and relational development rather than mathematical or scientific inquiry.
Judgment-Centred Software Engineering Education: A Post-Hype Review And Framework For AI-Augmented Learning
Qusay H. Mahmoud. (09/2026). arXiv.
Generative AI can improve access to explanations and practice, but effects on assessed learning are highly uneven and depend on scaffolding, prior knowledge, and task design.
Studentbench: AI And Human Tutoring Yield Equivalent Gre Learning Gains
Curtis Northcutt, Inaara Hasmani, Kevin Feng, Trevor Khangi, Andreas Plesner, Jonas Mueller. (09/2026). arXiv.
AI tutoring produced learning gains statistically equivalent to expert human tutoring on the combined GRE assessment (p=.015).
"We'Ll Fix It Later": Education, AI, And The Deferral Of Privacy In EdTech
Meghna Manoj Nair, Rachel Greenstadt. (09/2026). arXiv.
One-third of audited platforms made no meaningful AI disclosure despite visible AI features, and 73% provided only generic accountability and breach-response language.
AI-Assisted Social Story Intervention For Special Education: The Design Of Adapted Stories
Buyankhishig Enkhjargal, Himanshi Lalwani, Hanan Salam. (09/2026). arXiv.
AI assistance suggests it can lower the burden of preparing Social Stories while supporting more individualized story creation.
From Content Generation To Learning Support: Pedagogy-Guided Generative Video Tutors For STEM Learning
Xinchen Ma, Shuimu Wang, Gaole He, Yanbin Zhang, Chunyang Wang, Yunshi Lan, Weining Qian. (09/2026). arXiv.
The PIVOT framework generates educational videos with pedagogically aligned content, clear and engaging presentation, and coherent instructional flow across four STEM domains.
When AI Tutors Speak: Evidence From A Randomized Field Experiment
Shihao Yang, Marshall Van Alstyne, Chrysanthos Dellarocas. (09/2026). arXiv.
Students assigned to the structured AI tutor gained 6.63 more points on a 55-point assessment than peers in the holdout group.
A Course Intelligence Platform For Higher Education: Lessons From AI-Assisted Course Evaluation
Bo Yuan, Jiazi Hu, Haimei Zhao. (09/2026). arXiv.
A domain-adapted model, Llama-UKP, achieved substantial agreement with expert judgments, reaching a Pearson correlation of 0.754 and a Spearman correlation of 0.843.
Instructional Objectives Completed And Skill-Specific Early Reading Growth In A Digital Reading Tutor
Nicholas A. Gage. (09/2026). EdArXiv.
Each additional 10 objectives completed predicted a 3.15 percentile point gain on composite reading scores (g = 0.10).
Dose, Not Access: Digital Reading Tutor Use And Early Reading Growth
Nicholas A. Gage. (09/2026). EdArXiv.
High-dose users averaging at least 15 minutes weekly outperformed near-zero users on end-of-year DIBELS reading composite scores by a medium effect size.
Artificial Intelligence Driven Physics Assignments Using Context Prompts
Miguel Rodriguez, Peter Wulff. (09/2026). arXiv.
Context-prompted assignments helped students move beyond answer-seeking to connect introductory physics concepts directly to their jobs, hobbies, and everyday lives.
AI Literacy Over Tool Design: A Mixed-Methods Study Of Scaffolded Versus Unrestricted Generative AI In Programming Education
Sepinoud Azimi. (09/2026). arXiv.
Assignment scores and concept inventory gains were equivalent between students using the scaffolded AI coach and those using unrestricted AI tools.
The Role Of Implicit And Explicit Demographic Signals In Large Language Model-Based Student Assessment
Donya Rooein, Luca Benedetto, Dirk Hovy. (09/2026). arXiv.
In essay scoring, implicit conversation history caused Llama-70B scores to inflate by 1.57 points relative to default, while explicit demographics caused small shifts.
Enhancing Procedural Writing Through Personalized Example Retrieval: A Case Study On Cooking Recipes
Paola Mejia-Domenzain, Jibril Frej, Seyed Parsa Neshaei, Luca Mouchel, Tanya Nazaretsky, Thiemo Wambsganss, Antoine Bosselut, Tanja Kaser. (09/2026). arXiv.
Participants receiving adaptive feedback improved their recipe writing quality and completeness significantly more than those without adaptive feedback.
Evaluating AI Tutoring At The Speed Of Innovation: Practitioner-Led Micro-Randomised Trials Of An AI Tutoring Platform In Gcse Science
Wayne Harrison, Rahil Khowaja, Emma Dobson, Germaine Uwimpuhwe, Steve Higgins. (09/2026). arXiv.
Students allocated to the AI tutoring platform scored significantly higher on science post-tests than those doing business-as-usual revision, with an effect size of 0.33.
Generative AI Assisted Workflows In Architectural Conceptual Design: Performance, Creative Self-Efficacy, And Cognitive Load
Yao Xiao, Han Jiang, Rachel Hurley, Shichao Liu. (09/2026). arXiv.
Generative AI and precedent search showed no significant overall differences in conceptual design performance, cognitive workload, or task-specific creative self-efficacy.
An Emancipatory Vision For Designing (Generative) AI For Learner Flourishing
Luis Pablo Prieto, Yannis Dimitriadis. (09/2026). arXiv.
Current AI in education often relies on problematic assumptions, including treating agency as a zero-sum game, overestimating predictability in complex systems, and presuming infinite resources for permanent scaffolding.
Flawed But Memorable: Student Critical Reception Of Interest-Personalized Genai Analogies In Computing Education
Seth Bernstein, Naaz Sibia. (09/2026). arXiv.
Most participants found interest-personalized analogies more engaging and memorable than generic technical explanations, though reports of trust were mixed.
From Misconceptions To Evidence: What Science Teachers Make Visible When Co-Designing Agentic Learning Apps
Nizam Kadir, Wei Ting Liow, Sumbul Khan, Lay Kee Ang. (09/2026). arXiv.
All four design proposals focused AI on eliciting, scaffolding, or diagnosing scientific reasoning rather than generating answers.
A Prompt-Engineering Approach To Develop Scalable, Flexible, And Real-Time Hybrid Micro-Level Personalization In A General Purpose AI Teaching Assistant
Saptarshi Basu, Sandeep Kakar, Ashok Goel. (09/2026). arXiv.
Prompt-based configurations produced high semantic similarity but lower lexical similarity across 2,910 responses, showing that answers remained course-grounded while varying in phrasing and structure.
Beyond Helpfulness: A Teaching-Over-Solving Diagnostic For Measuring Educational Impact In LLM Tutors
Junyi Yao, Zihao Zheng, Baichuan Li. (08/2026). arXiv.
Problem-solving and pedagogical scores were only moderately correlated across eight AI models (Pearson correlation of 0.421, Spearman rank correlation of 0.690).
Learning To Prompt: Improving Student Engagement With Adaptive LLM-Based High-School Tutoring
Po-Chin Chang, Nicholas Hogan, Aske Plaat, Michiel T. van der Meer. (08/2026). arXiv.
In simulations, the adaptive prompt router outperformed static baselines (0.694 versus 0.647 and 0.64) and matched optimal strategies in 8 of 13 evaluated subjects.
