A Comparative Study of AI-Generated (GPT-4) and Human-crafted MCQs in Programming Education
Open accessThe DOI matches its Crossref record; 82 of the 82 references it cites were found in a registry or an index; 11 of the 25 sources its claims cite were checked for support.
Abstract
There is a constant need for educators to develop and maintain effective up-to-date assessments. While there is a growing body of research in computing education on utilizing large language models (LLMs) in generation and engagement with coding exercises, the use of LLMs for generating programming MCQs has not been extensively explored. We analyzed the capability of GPT-4 to produce multiple-choice questions (MCQs) aligned with specific learning objectives (LOs) from Python programming classes in higher education. Specifically, we developed an LLM-powered (GPT-4) system for generation of MCQs from high-level course context and module-level LOs. We evaluated 651 LLM-generated and 449 human-crafted MCQs aligned to 246 LOs from 6 Python courses. We found that GPT-4 was capable of producing MCQs with clear language, a single correct choice, and high-quality distractors. We also observed that the generated MCQs appeared to be well-aligned with the LOs. Our findings can be leveraged by educators wishing to take advantage of the state-of-the-art generative models to support MCQ authoring efforts.
Overview
CoChat Summary
## TL;DR
The paper evaluates a GPT-4 pipeline for generating programming MCQs from course context and module-level learning objectives (LOs), comparing 651 AI-generated questions with 449 human-crafted ones across 246 LOs from 6 courses. It finds the generated MCQs are broadly comparable in quality to human-written ones, and better aligned with the intended LOs.
## Key findings
- The study evaluated 1,100 MCQs total: 651 generated by GPT-4 and 449 human-crafted, aligned to 246 module-level LOs from 6 courses.
- For MCQ quality (RQ1), the authors report the generated MCQs were comparable in quality to human-crafted ones, but had more problems with:
- multiple correct answers: 4.9% of generated vs 1.1% of human-crafted MCQs;
- distractors that gave away the answer: 4.0% of generated vs 0.9% of human-crafted MCQs.
- For LO alignment (RQ2), generated MCQs were noticeably better aligned with the LOs than human-crafted ones; Fisher’s exact test reported p < 10^-9.
- The paper reports 81.7% of all generated MCQs passed all evaluation criteria.
## Methods
- Built an LLM-powered MCQ generation pipeline using GPT-4 (gpt-4-0613), high-level course context, module information, and module-level LOs; the prompt also incorporated MCQ principles, Bloom’s taxonomy, question-type examples, and a JSON output format.
- Dataset: 246 LOs from four Python programming and two introductory data science courses; human MCQs were collected from 3 courses, manually matched to LOs, and reduced from 529 collected to 449 usable single-answer MCQs after exclusions.
- Evaluation: 7 students and 6 CS instructor annotators applied a 6-criterion rubric to the 651 generated and 449 human-crafted MCQs; the authors used Fisher’s exact test and reported inter-rater agreement with Gwet’s AC1 and Fleiss’ κ.
## Limitations
- The paper states that the human rater pool was limited, which may introduce bias.
- Because the human MCQs were created before the study, the instructors did not have the rubric, and the authors say the instructors might have written questions differently if they had.
- The authors note that their manual pairing of human MCQs to LOs may have caused some of the observed MCQ-LO misalignment.
- The study did not evaluate live classroom utility or pedagogical impact, did not compare difficulty of generated vs human MCQs, and did not assess alignment with the learning content rather than just the LO.
- Evident from the text: the work is limited to Python programming courses at the higher-education level, so generalizability to other languages, domains, or education levels is uncertain.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures



Verification
Claims the paper makes
6 Valid · 0 Not valid · 7 Not checked · 72 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“To address this issue we also compute Gwet's AC1 which is robust in such situations [58].”
What the sources say
- Supportedcites [58] [58] N. Wongpakaran, T. Wongpakaran, D. Wedding, and K. Gwet. 2013. A comparison of Cohen's Kappa and Gwet's AC1 when calculating inter-rater reliability coefficients: a study conducted with personality disorder samples. BMC Med Res Methodol (2013). https://doi.org/10.1186/1471-2288-13-61 - source
The Results (Chance agreement probability; page not provided) examine cases where “No” dominates the ratings and report that AC1 remained robust to the resulting marginal-probability problem. The Conclusions also favor AC1 over Cohen’s Kappa for that problem. (LLM, conf=0.92)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The 𝜅 statistic is known to severely underestimate the agreement in situations when one of the labels is clearly dominant [59].”
What the sources say
- Partly supportedcites [59] [59] S. Zec, N. Soriani, R. Comoretto, and I. Baldi. 2017. High Agreement and High Prevalence: The Paradox of Cohen's Kappa. The Open Nursing Journal (2017), 221-218. https://doi.org/10.2174/1874434601711010211 - sourceExperimental — accuracy not yet measured
The Results report that Cohen’s κ made agreement appear unsatisfactory despite high observed concordance when particular labels were prevalent; the Introduction calls this a prevalence-related paradox. The paper discusses Cohen’s κ, not Fleiss κ, so it does not establish the claim specifically for Fleiss κ. No page numbers are available in our copy. (LLM, conf=0.86)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We found that GPT-4 was capable of producing MCQs with clear language, a single correct choice, and high-quality distractors.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We found that GPT-4 was capable of producing MCQs with clear language, a single correct choice, and high-quality distractors.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We found that GPT-4 was capable of producing MCQs with clear language, a single correct choice, and high-quality distractors.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
32 more claims
Create a free account to see all 85 claims and to verify the 2 references still open.
- Probably not a substantial claim. 34
- It may state two separate claims as one. 20
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Nevertheless, MCQs are the most common type of knowledge assessment [5], e.g., in medical education [7].”
From Docimological Quality Analysis of LLM-Generated Multiple Choice Questions in Computer Science and Medicine (2024) · cites this paper as [5]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“The creation of high-quality MCQs is a time-consuming task [5, 6], even more considering that changes in the curriculum or students collecting questions from previous years requires teachers to create new sets of questions every once in a while [7].”
From Docimological Quality Analysis of LLM-Generated Multiple Choice Questions in Computer Science and Medicine (2024) · cites this paper as [5]
Citations
Cites
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404
Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318
Exploiting Generative AI to Scale up Intelligent Tutoring Systems
Jakubův, J., Chvalovský, K., Goertzel, Z., Kaliszyk, C., Olšák, M., Piotrowski, B., Schulz, S., Suda, M., & Urban, J. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.itp.2023.19 · Cited by 76,658
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173
A revision of bloom's taxonomy: an overview
Krathwohl, D. R. (2002) · Theory Into Practice · doi:10.1207/s15430421tip4104_2 · Cited by 8,132
Wongpakaran, N., Wongpakaran, T., Wedding, D., & Gwet, K. L. (2013) · BMC Medical Research Methodology · doi:10.1186/1471-2288-13-61 · Cited by 1,041 · cited 1×
“To address this issue we also compute Gwet’s AC1 which is robust in such situations [58].”
MethodLearning to Ask: Neural Question Generation for Reading Comprehension
Du, X., Shao, J., & Cardie, C. (2017) · Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p17-1123 · Cited by 696 · cited 1×
“Using the fine-tuned BERT classifier [24] (Section 4 has details), each LO was categorized into one of the six levels of the revised Bloom’s taxonomy [19].”
MethodA Systematic Review of Automatic Question Generation for Educational Purposes
Kurdi, G., Leo, J., Parsia, B., Sattler, U., & Al-Emari, S. (2020) · International Journal of Artificial Intelligence in Education · doi:10.1007/s40593-019-00186-y · Cited by 514
Automatic Generation of Programming Exercises and Code Explanations Using Large Language Models
Sarsa, S., Denny, P., Hellas, A., & Leinonen, J. (2022) · Proceedings of the 2022 ACM Conference on International Computing Education Research - Volume 1 · doi:10.1145/3501385.3543957 · Cited by 497
The Robots Are Here: Navigating the Generative AI Revolution in Computing Education
Prather, J., Denny, P., Leinonen, J., Becker, B. A., Albluwi, I., Craig, M., Keuning, H., Kiesler, N., Kohn, T., Luxton-Reilly, A., et al. (2023) · Proceedings of the 2023 Working Group Reports on Innovation and Technology in Computer Science Education · doi:10.1145/3623762.3633499 · Cited by 396 · cited 1×
“Note that some of the LOs extracted from the courses were not well-defined (e.g., no action verb) and, hence, it was not possible to categorize them in terms of Bloom’s taxonomy.”
Background
Research works printed but not linked: 1
- [2] Ali Mohamed Nabil Allam and Mohamed Hassan Haggag. 2012. The question answering systems: A survey. International Journal of Research and Reviews in Information Sciences (IJRRIS) 2, 3 (2012). -
Cited by
Items 101–125 of 125
Augmenting an LLM-Based Tutor Agent with a Large Action Model for Multimodal Interaction
Kluczyński, P., Picard, A., Mualla, Y., Tebourbi, H., Nouzri, S., Abbas‐Turki, A., & Gechter, F. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-20548-3_3
Generative AI-Based Automated Assessment Generation
Nyaaba, M., & Zhaı, X. (2026) · Springer eBooks · doi:10.1007/978-3-032-06565-0_3
AI-Driven Assessment in Early and Primary Education: Tools, Strategies, and Ethical Considerations
Even‐Simkin, E. (2026) · Springer eBooks · doi:10.1007/978-3-032-17190-0_17
Gao, H., & Zheng, J. (2026) · International Conference on Educational and Information Technology · doi:10.1109/iceit68991.2026.11521840
Two-Stage Fine-Tuning of Educational Large Language Models for Enhancing Appreciation Capabilities
Wu, Y., Wang, Z., Sun, H., & Xiang, Y. (2026) · International Conference on Big Data and Informatization Education (ICBDIE) · doi:10.1145/3806980.3806994
Large language models in software development learning
Hernández, S. C., Cruz, M. P. T., Sánchez, J. M. E., Luna-Gómez, A. A., Orozco, A. M. P., & Medina, R. H. (2026) · Discover Education · doi:10.1007/s44217-026-01750-0 · Cited by 3
Label Concreteness-Guided Diverse Data Synthesis for Low-Resource Relation Extraction
Liu, X., Hu, C., Zhang, R., Chen, J., & Xu, B. (2026) · IEEE Transactions on Audio Speech and Language Processing · doi:10.1109/taslpro.2026.3705560
IILAP+: Exploring AI-Assisted Dataset Creation for Critical Thinking Assessment
Nurbakova, D., & Магдановна, Е. Л. (2026) · ACM Conference on Learning @ Scale (L@S) · doi:10.1145/3774398.3811569
Supporting Data-Driven Mastery Learning at Scale
Wei, Y. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-29794-5_58
Dimitriadou, E. A., Stokke, S. A., Hegreberg, S., Nguyen, A., & Giannakos, M. N. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-29770-9_11 · Cited by 2
Generating Personalized Programming Exercises via Cognitive State Graphs and ZPD-Driven Prompting
Zhang, L., Zhu, J., & Ma, Y. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-29744-0_10
i-vip: A LLM-Driven Multi-agent System for Professional Development of Mathematics Teachers
Li, H., Yang, K., Chu, Y., Han, A., Meng, R., Copur‐Gencturk, Y., & Liu, H. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-29744-0_20
CODE-GEN: A RAG-Based Agentic AI System for Multiple-choice Question Generation
Duan, X., Nwanganga, F., & Wang, C. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-29760-0_10 · Cited by 1
Leveraging Human-AI Collaboration for a Passage-Based Question Authoring Tool
Demirtaş, M. A., Nam, S., & Griffin, G. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-29760-0_50
Using AI-powered multiple-choice question generation for self-regulated learning
Barra, E., Pilicita, A., Conde, J., Pozo, A., López‐Pernas, S., & Reviriego, P. (2026) · Revista de Educación a Distancia (RED) · doi:10.6018/red.711091
Proposal for an AI-Driven Learning Platform with Gamification for K-12 Students
Radovan, A., Klasić, A., Bele, D., & Mihaljević, B. (2026) · International Convention on Information and Communication Technology, Electronics and Microelectronics · doi:10.1109/mipro70003.2026.11591808
WEAVER: Weaving Learnersourced Questions into High-Quality Assessments with AI
Ni, L., Yin, S., Denny, P. C., & Liu, J. (2026) · Lecture notes in computer science · doi:10.1007/978-981-92-2864-5_24
Automatic Question Generation with Large Language Models: A Survey
de Oliveira, R. M., Hernández, A., Garbin, M. C., dos Reis, J. C., & Azevedo, R. (2026) · ACM Computing Surveys · doi:10.1145/3832773 · Cited by 1
Scaffolding Autocomplete: Improving Guidance for Learners using Generative Code Suggestions
Prather, J. E., MacNeil, S., Luxton-Reilly, A., Margulieux, L. E., Reeves, B. N., Denny, P. C., Leinonen, J., Homer, J., Nabid, R. A., & Rossetti, R. L. (2026) · ACM Conference on International Computing Education Research (ICER) · doi:10.1145/3765964.3811654
AI-Powered Automatic Quiz Generator with Explanations and Summarization for Educational Content
Yogendra, & Maralidhar, A. (2026) · International Conference on Power, Control and Computing Technologies (ICPC2T) · doi:10.1109/icpc2t68221.2026.11646354
Tang, X., Filonczuk, A., Zhang, X., & Cheng, Y. (2026) · Springer eBooks · doi:10.1007/978-3-032-24819-0_19
Yang, A. C. M., Kao, C., & Wu, Y. (2026) · IEEE Access · doi:10.1109/access.2026.3734218
Tselepatiotis, M., & Alepis, E. (2026) · International Conference on Information, Intelligence, Systems & Applications (IISA) · doi:10.1109/iisa71656.2026.11705534
Ibrahim, R. F., & Alnagi, E. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-39783-6_7
Sarhan, O. D. (2026) · Al-Mustansiriyah Journal of Sciences · doi:10.23851/mjs.v37i3.1857
Metadata
- Authors
- Doughty, J., Wan, Z., Bompelli, A., Qayum, J., Wang, T., Zhang, J., Zheng, Y., Doyle, A., Sridhar, P., Agarwal, A., Bogart, C., Keylor, E., Kultur, C., Savelka, J., & Sakr, M.
- Affiliations
- Carnegie Mellon University (US)
- Carnegie Mellon University, United States
- ORCID
- Venue
- Proceedings of the 26th Australasian Computing Education Conference
- Publisher
- ACM
- Published
- 2024-01-29
- Pages
- 114-123
- OpenAlex
- W4389471004
- Cited by
- 132
- References
- 59
- Citations by year
- 2026: 50 · 2025: 59 · 2024: 23
- FWCI
- 57.70
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- Australasian Computing Education Conference (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion, cc-by)
- arXiv (Cornell University) (cc-by)
- Topics
- Physical Sciences › Computer Science › Computer Science Applications › Teaching and Learning Programming
- Social Sciences › Education › Artificial Intelligence in Education
- Physical Sciences › Computer Science › Information Systems › Educational Technology and Assessment
- Keywords
- GPT-4, large language models, multiple-choice question generation, Python programming education, learning objectives, distractor quality, programming assessment
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.