id,authors,year,title,venue,status,country,participants,participants_note,design_type,design,area,measured_when,what_was_measured,main_finding,short_result,result,limitations,link,checked_from,notes,checked_on
burelli-2026,"Paolo Burelli, Fabio Calefato, Daniela Grassi, Mihaela Yurieva Hristova, Nicole Novielli, Alberto Antonio Romano, Paolo Tell",2026,Using Biometrics to Understand AI-Assisted Coding Performance and its Perception,arXiv preprint 2606.20598 (May 2026),Preprint,Italy and Denmark,60,"60 computer science students (34 in Bari, 26 in Copenhagen)",Brain study,"Crossover experiment with EEG, eye tracking and other sensors: coding with and without an AI assistant",Brain activity,Brain activity recorded,"EEG theta/alpha ratio, blink rate, skin conductance, heart rate, code quality and self-rated workload","With the AI assistant, EEG and eye measures pointed to lower mental engagement, and the link between bodily arousal and code quality seen without AI disappeared.",EEG and eye measures showed lower engagement when coding with AI,Worse,"Not peer reviewed; students, not professional developers; signals are indirect markers of engagement.",https://arxiv.org/abs/2606.20598,full text,,2026-10-16
clin-2026,"Alexandre Clin Deffarges, Nataliya Kosmyna, Pattie Maes",2026,Socrates went Nuclear: Comparing Interaction Strategies for AI systems in a Learning Context using Brain Sensing,arXiv preprint 2609.00584 (September 2026),Preprint,Not stated,50,50 participants in three groups,Brain study,Randomized experiment with an EEG headband: an unrestricted chatbot vs a hint-only Socratic chatbot vs a brain-adaptive tutor,Brain activity,Brain activity recorded,Learning gain on an immediate post-test and EEG-based engagement,"The unrestricted chatbot produced the biggest immediate test gains, the brain-adaptive tutor the highest EEG engagement, and most unrestricted users simply looked up answers, which the authors say is not evidence of deeper learning.","an unrestricted chatbot gave the biggest immediate gains, mostly by giving answers",Mixed,Not peer reviewed; small groups; only an immediate test; a consumer EEG headband.,https://arxiv.org/abs/2609.00584,full text,,2026-10-16
gao-2026,"Yu-Yan Gao, Jingyi Zhang, Shen Xu, Wanying Zheng, Yafeng Pan, Xiaojing Li",2026,Prefrontal multiscale entropy and state transitions distinguish large language model-assisted from search-assisted learning,"NeuroImage 338, 122123",Peer reviewed,China*,,Sample size not given in the abstract,Brain study,"Experiment with fNIRS: the same test taken with an LLM, with search, or unassisted",Brain activity,Brain activity recorded,Test performance and the complexity of prefrontal signals (fNIRS),"Both the LLM and search raised test scores over working unaided, but with the LLM the prefrontal signal was less complex and switched between states less often than with search.","same benefit as search, but less complex prefrontal activity",Mixed,Sample size not stated in the abstract; signal complexity is hard to interpret as better or worse thinking.,https://doi.org/10.1016/j.neuroimage.2026.122123,abstract,,2026-10-16
jin-2026,"Zheyu Jin, Junting Yin, Qiuyu Tian, Xinyan Zhou, Yangzhuo Li, Ting Yang, Junlong Luo",2026,"Can large language models teach creativity? Cognitive gains, affective gaps, and distinct neural patterns","Computers in Human Behavior: Artificial Humans 8, 100288",Peer reviewed,China*,95,95 participants in three groups,Brain study,Randomized experiment with fNIRS: creativity training by an LLM vs by a human instructor vs no training,Brain activity,Brain activity recorded,Creativity (Alternative Uses Task) and prefrontal and temporo-parietal activity (fNIRS),"Training from an AI instructor raised creativity scores, especially originality and flexibility, but the AI and human instructors engaged different brain systems, and the human instructor produced stronger prefrontal connectivity and a closer sense of rapport.","an AI creativity instructor raised creativity scores, with different brain activity",Better,Small groups; short training; fNIRS covers only part of the brain surface.,https://doi.org/10.1016/j.chbah.2026.100288,abstract,,2026-10-16
shrivastava-2026,"Rhea S. Shrivastava, Aditi Sharma, Komal Dabas, Deepayan Mukherjee, Gaetano Cascini, Sonal Keshwani",2026,Neurocognitive assessment of generative AI on designers’ creative cognition: evidence from biologically inspired design tasks,"Proceedings of the Design Society 6, 1111-1120",Peer reviewed,India*,30,30 designers,Brain study,"Experiment with EEG: design ideas from nature examples only, ChatGPT only, or both",Brain activity,Brain activity recorded,EEG markers of convergent and divergent thinking across four design phases,How ChatGPT changed EEG markers of convergent and divergent thinking depended on how it was combined with nature examples and on the design phase.,EEG effects of ChatGPT depended on the design phase,Mixed,Small sample; conference paper; EEG power is an indirect marker of creative thinking.,https://doi.org/10.1017/pds.2026.10469,abstract,,2026-10-16
xia-2026,"Liangliang Xia, Yan Dong, Wei-Peng Teo",2026,Cognitive and neural mechanisms of improving informal reasoning in human-GenAI interactive learning contexts: An fNIRS study,"NeuroImage 328, 121796",Peer reviewed,China*,78,78 participants,Brain study,Experiment with fNIRS: learn with a generative AI chatbot vs with a search engine,Brain activity,Brain activity recorded,"Improvement in informal reasoning, and prefrontal and temporo-parietal activity (fNIRS)","Learners using generative AI improved their reasoning no more and no less than learners using a search engine, but different patterns of brain activity went with improvement in each group.",reasoning improved no differently with AI or search,No clear difference,Modest sample; one learning task; fNIRS covers only part of the brain surface.,https://doi.org/10.1016/j.neuroimage.2026.121796,abstract,,2026-10-16
zhang-2026,"Jiayue Zhang, Bochen Xu, Yiheng Liu, Wenqi Cai, Lanlan Wu, Senqing Qi, Yali Peng, Jingjing Yu, Taotao Long, Bao Ge",2026,Exploring ChatGPT's potential in dialogic teaching: A comparative neurocognitive study of AI and human instruction,"Acta Psychologica 268, 107268 (2026); an earlier version appeared at an IEEE conference in 2024",Peer reviewed,China*,34,34 undergraduates,Brain study,Randomized experiment with EEG: a lesson taught through dialogue with ChatGPT vs with a human instructor,Brain activity,Brain activity recorded,"Retention and transfer tests, and EEG during and after the lesson","Students taught through ChatGPT remembered the lesson as well as students taught by a person but did worse at applying it to new problems, and their EEG suggested more visually driven effort.","same retention, weaker transfer after learning through ChatGPT",Mixed,"Very small sample; one lesson; compared with a human tutor, not with no help.",https://doi.org/10.1016/j.actpsy.2026.107268,abstract,,2026-10-16
baradari-2025,"Dünya Baradari, Nataliya Kosmyna, Oscar Petrov, Rebecah Kaplun, Pattie Maes",2025,NeuroChat: A Neuroadaptive AI Chatbot for Customizing Learning Experiences,CUI 2025 (ACM conference on conversational user interfaces),Peer reviewed,United States*,24,"24 participants, each tried both chatbots",Brain study,Within-person experiment with EEG: a chatbot that adapts to EEG-measured engagement vs a standard chatbot,Brain activity,Brain activity recorded,"EEG and self-reported engagement, and short-term learning","A chatbot that adapted to each learner's EEG-measured engagement raised engagement compared with a standard chatbot, but short-term learning did not differ.",a brain-adaptive chatbot raised engagement but not learning,Mixed,"Small sample; a short session; compares two chatbots, not chatbot vs no chatbot.",https://doi.org/10.1145/3719160.3736623,full text,,2026-10-16
ge-2025,"Shijun Ge, Shaoshuai Huang, Xinheng Song, Yuanbo Sun, Nanyi Wang, Kaize Qi, Xuandong Zhao, Jiahua Li, Yin Cui",2025,Comparing TRIZ and brainstorming in human–agent design collaboration: effects on cognitive processes and performance,"Artificial Intelligence for Engineering Design, Analysis and Manufacturing 39, e21",Peer reviewed,China*,32,32 designers,Brain study,Randomized experiment with fNIRS: designing with LLM-based agents vs with experienced human designers,Brain activity,Brain activity recorded,"Prefrontal activity (fNIRS), cognitive load, and expert ratings of the design ideas","Designers working with LLM agents showed lower mental load and more fluent and flexible ideas than designers working with human partners, but the human pairs came up with more original ideas.",LLM partners lowered mental load; human partners gave more original ideas,Mixed,"Small sample; compared with human collaborators, not with working alone.",https://doi.org/10.1017/s0890060425100097,full text,,2026-10-16
kosmyna-2025,"Nataliya Kosmyna, Eugene Hauptmann, Ye Tong Yuan, Jessica Situ, Xian-Hao Liao, Ashly Vivian Beresnitzky, Iris Braunstein, Pattie Maes",2025,Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task,"arXiv preprint 2506.08872 (version 2, 31 Dec 2025)",Preprint,United States (Boston area),54,54 students in sessions 1-3 (18 per group); 18 returned for session 4,Brain study,"Lab experiment with EEG: ChatGPT vs search engine vs no tool, 4 sessions over 4 months",Brain activity,Brain activity recorded,EEG brain connectivity while writing SAT-style essays; ability to quote one's own essay; sense of ownership,"Brain connectivity was weakest in the ChatGPT group and strongest in the no-tool group, and 15 of 18 ChatGPT users (83%) could not correctly quote their own essay minutes after writing it, against 2 of 18 in each of the other groups.",weakest brain connectivity with ChatGPT; 83% could not quote their own essay,Worse,Not peer reviewed; 18 people per group and only 18 in the crossover session; one essay task. A published commentary questions the EEG analysis and the reporting of results.,https://arxiv.org/abs/2506.08872,full text,,2026-10-16
wang-2025,"Shuxin Wang, Xin Tao, Hongbo Ma, Fanglian Li, Chuanqi Wu",2025,EEG assessment of artificial intelligence-generated content impact on student creative performance and neurophysiological states in product design,"Frontiers in Psychology 16, 1508383",Peer reviewed,China,64,64 design students (32 per group),Brain study,"Randomized experiment with a consumer EEG headband: a 3-hour design task with ChatGPT, Midjourney and Stable Diffusion vs traditional software",Brain activity,Brain activity recorded,"Creative performance of the designs, and concentration and relaxation from EEG",Students using generative AI tools produced designs rated more creative (115.1 vs 110.7 points) and showed higher EEG concentration levels than students using traditional software.,more creative designs and higher EEG concentration with AI tools,Better,A consumer EEG headband; one design task; the AI group also had image generators.,https://doi.org/10.3389/fpsyg.2025.1508383,full text,,2026-10-16
tanveer-2026,Muhammad Tanveer,2026,Artificial intelligence literacy moderating the effects of artificial intelligence use on cognitive outcomes including memory retention decision making and cognitive laziness,"Discover Psychology 6, 182",Peer reviewed,Saudi Arabia,518,518 university students,Survey,Survey (cross-sectional),Memory,What people reported,"AI use, self-rated memory retention, decision making and cognitive laziness","Students who used AI more rated both their memory and their decision making higher, and also reported more cognitive laziness.",more AI use went with better self-rated memory and more laziness,Mixed,Self-reported measures; path coefficients near 0.97 are unusually high for survey data; single author.,https://doi.org/10.1007/s44202-026-00697-5,full text,,2026-10-16
barcaui-2025,André Barcaui,2025,ChatGPT as a cognitive crutch: Evidence from a randomized controlled trial on knowledge retention,"Social Sciences & Humanities Open 12, 102287",Peer reviewed,Brazil*,120,120 undergraduates,Experiment,"Randomized trial: study with ChatGPT vs traditional methods, then a surprise test 45 days later",Memory,Tested later without AI,"A surprise knowledge-retention test 45 days after learning, without AI","On a surprise test 45 days later, students who had studied with ChatGPT answered 57.5% correctly, against 68.5% for students who studied without it.",57.5% vs 68.5% correct on a surprise test 45 days later,Worse,One course; single author; country taken from the author's university.,https://doi.org/10.1016/j.ssaho.2025.102287,abstract,,2026-10-16
kreijkes-2025,"Pia Kreijkes, Viktor Kewenig, Martina Kuvalja, Mina Lee, Jake M. Hofman, Sylvia Vitello, Abigail Sellen, Sean Rintel, Daniel G. Goldstein, David Rothschild, Lev Tankelevitch, Tim Oates",2025,Effects of LLM use and note-taking on reading comprehension and memory: A randomised experiment in secondary schools,"Computers & Education 243, 105514",Peer reviewed,United Kingdom (England),405,405 students aged 14-15,Experiment,"Pre-registered randomized experiment in schools: study texts with an LLM, by taking notes, or both",Memory,Tested later without AI,"Comprehension and recall tests three days later, without AI","Three days later, students who took notes, with or without the LLM, understood and remembered more than students who used the LLM alone, although most students preferred the LLM.",note-taking beat using an LLM alone for recall three days later,Worse,Two text passages; one study session; England only.,https://doi.org/10.1016/j.compedu.2025.105514,abstract,,2026-10-16
abbas-2024,"Muhammad Abbas, Farooq Ahmed Jam, Tariq Iqbal Khan",2024,Is it harmful or helpful? Examining the causes and consequences of generative AI usage among university students,"International Journal of Educational Technology in Higher Education 21, 10",Peer reviewed,Pakistan,494,494 university students surveyed three times (study 2),Survey,Survey in three waves (time-lagged),Memory,What people reported,"ChatGPT use, procrastination, self-rated memory problems and self-reported grade point average","Students who used ChatGPT more reported more procrastination and more memory problems, and reported lower grades.",more ChatGPT use went with self-reported memory problems and lower grades,Worse,"Self-reported memory and grades; a survey cannot show cause, even across three waves.",https://doi.org/10.1186/s41239-024-00444-7,full text,,2026-10-16
chan-2024,"Samantha Chan, Pat Pataranutaporn, Aditya Suri, Wazeer Zulfikar, Pattie Maes, Elizabeth F. Loftus",2024,Conversational AI Powered by Large Language Models Amplifies False Memories in Witness Interviews,arXiv preprint 2408.04681; CrimRxiv,Preprint,Online panel (Prolific),200,200 adults in four groups,Experiment,"Randomized experiment: watch a crime video, then answer questions from an LLM chatbot, a scripted chatbot, a survey, or no interview",Memory,Tested later without AI,"False memories of the video, immediately and one week later, and confidence in them","An LLM chatbot that asked misleading questions created over three times as many false memories as the control and 1.7 times as many as a survey with the same questions, and those false memories were still there a week later.",a misleading chatbot planted three times as many false memories,Worse,"Not peer reviewed; the chatbot was designed to mislead, as in a suggestive police interview; one video.",https://arxiv.org/abs/2408.04681,full text,,2026-10-16
bassner-2026,"Patrick Bassner, Ben Lenk-Ostendorf, Ramona Beinstingel, Tobias Wasner, Stephan Krusche",2026,"Less stress, better scores, same learning: The dissociation of performance and learning in AI-supported programming education","Computers and Education: Artificial Intelligence 10, 100537",Peer reviewed,Germany,275,275 students in an introductory programming course,Experiment,"Three-arm randomized trial: a hint-giving AI tutor vs unrestricted ChatGPT vs no AI, in a 90-minute exercise",Learning,Tested later without AI,Exercise scores with AI; knowledge tests and code comprehension without AI,"Both AI groups scored much higher on the exercise, but neither learned more than the no-AI group on tests taken afterwards.","higher exercise scores with AI, no extra learning",No clear difference,One 90-minute exercise in one course.,https://doi.org/10.1016/j.caeai.2025.100537,abstract,,2026-10-16
contractor-2026,"Zara Contractor, Germán Reyes",2026,Experimental Evidence on the Learning Impact of Generative AI,IZA Discussion Paper 18792 (July 2026),Working paper,United States,211,211 undergraduates (204 came back a week later),Experiment,Randomized experiment in proctored sessions: learn a new topic and write an essay with or without off-the-shelf AI,Learning,Tested later without AI,"Knowledge tests and essays written without AI, right after and one week later","Students allowed to use AI while studying scored 0.27 standard deviations higher on tests taken without it, and the gain held a week later; students who used AI to explain concepts gained the most.",studying with AI raised unaided test scores by 0.27 SD,Better,Not peer reviewed; one college; one session of study.,https://docs.iza.org/dp18792.pdf,full text,,2026-10-16
hou-2026,"Xiaoyu Hou, Bo Xiao, Hexu Liu, Shane Mueller",2026,The Role of Instructional Guidance in Generative AI-Assisted Learning: Empirical Evidence from Construction Engineering Education,arXiv preprint 2606.05509 (June 2026),Preprint,United States*,95,"95 students (33 slides, 29 guided AI, 33 unguided AI)",Experiment,"Randomized experiment: review with slides, with an AI assistant, or with the AI plus a five-step prompting guide",Learning,Tested later without AI,Multiple-choice and open-ended test questions without AI,"Students who used the AI with a structured prompting guide scored about 2 to 3 points higher (out of 18) on open-ended questions, while students who used the AI without guidance did no better than students who reviewed slides.",AI with a prompting guide helped; AI without one did not,Mixed,Not peer reviewed; small groups; one topic.,https://arxiv.org/abs/2606.05509,full text,,2026-10-16
learnlm-2026,LearnLM Team (Google) and Fab AI,2026,Teaching with Gemini: Measuring the impact of Guided Learning on student mathematics progress in Sierra Leone,"Technical report, Google DeepMind (May 2026)",Working paper,Sierra Leone,1763,"1,763 students in 48 classrooms at 12 schools",Experiment,Pre-registered cluster-randomized trial: Gemini Guided Learning in half of weekly maths lessons for eight weeks vs usual teaching,Learning,Tested later without AI,"A maths assessment at the end, without AI",Classes that used Gemini's tutoring mode in half their maths lessons for eight weeks gained 0.26 standard deviations more on the end-of-trial maths test.,Gemini Guided Learning raised maths scores by 0.26 SD,Better,Google's own evaluation of its own product; not peer reviewed; teachers led the AI sessions.,https://storage.googleapis.com/deepmind-media/LearnLM/learnLM_sierraleone_may26.pdf,full text,,2026-10-16
liu-2026,"Grace Liu, Brian Christian, Tsvetomira Dumbalska, Michiel A. Bakker, Rachit Dubey",2026,AI Assistance Reduces Persistence and Hurts Independent Performance,COLM 2026 (Conference on Language Modeling); arXiv 2604.04721,Peer reviewed,United States (online panel),1222,"1,222 adults across three experiments",Experiment,"Three randomized experiments: AI assistant for 10-15 minutes, then problems without it",Learning,Tested later without AI,"Fraction problems and SAT reading questions solved, and questions skipped, after the AI was removed",People who had an AI assistant solved fewer problems once it was removed (reading: 76% vs 89% of the control group; fractions: 71% vs 77%) and gave up on more of them.,solved fewer problems and gave up more once the AI was removed,Worse,Short online sessions of about 15 minutes; simple tasks; effects on long-term learning were not measured.,https://arxiv.org/abs/2604.04721,full text,,2026-10-16
shen-2026,"Judy Hanwen Shen, Alex Tamkin",2026,How AI Impacts Skill Formation,"arXiv preprint 2601.20245 (version 2, Feb 2026)",Preprint,Not stated (online developers),52,"52 software developers, 26 per group",Experiment,Pre-registered randomized experiment: learn a new coding library with or without an AI assistant,Learning,Tested later without AI,"Quiz on the library's concepts, code reading and debugging, taken without AI; time to finish",Developers who learned with the AI assistant scored 17% lower on the follow-up quiz (about two grade points) and were not significantly faster.,learning a new code library with AI left quiz scores 17% lower,Worse,Small sample; one library and one session; written by researchers at an AI company.,https://arxiv.org/abs/2601.20245,full text,,2026-10-16
bastani-2025,"Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakcı, Rei Mariman",2025,Generative AI without guardrails can harm learning: Evidence from high school mathematics,"PNAS 122(26), e2422633122",Peer reviewed,Turkey,1000,"nearly 1,000 students in about 50 classes",Experiment,"Randomized field experiment: GPT-4 chat vs GPT-4 tutor with safeguards vs no AI, four 90-minute sessions",Learning,Tested later without AI,"Math practice scores with AI, then exam scores with no AI","Plain GPT-4 raised practice scores by 48% but, once it was taken away, those students scored 17% lower on the exam than students who never had it; a tutor version that gave hints instead of answers removed the drop.",exam scores 17% lower after plain GPT-4 was taken away,Worse,"One school, one subject, autumn 2023; the tutor version did not raise exam scores either.",https://doi.org/10.1073/pnas.2422633122,full text,A correction (August 2025) fixed one author's affiliation; the results are unchanged.,2026-10-16
dai-2025,"Xusheng Dai, Zhaochun Wen, Jianxiao Jiang, Huiqin Liu, Yu Zhang",2025,How Students Use AI Feedback Matters: Experimental Evidence on Physics Achievement and Autonomy,"arXiv preprint 2505.08672 (version 2, May 2025)",Preprint,China,387,387 high school students in two five-week trials (121 and 266),Experiment,Two randomized trials: AI-generated hints on homework (compulsory) and on-demand AI help vs usual workbook answers,Learning,Tested later without AI,Physics achievement and self-regulated learning,"Effects depended on the student: required AI hints helped low achievers, on-demand AI help helped high achievers, but high achievers given required hints and lower achievers given on-demand help became less self-directed.",AI feedback helped some students and made others less self-directed,Mixed,Not peer reviewed; one school system; subgroup effects in modest samples.,https://arxiv.org/abs/2505.08672,full text,,2026-10-16
desimone-2025,"Martin De Simone, Federico Tiberti, Maria Barron Rodriguez, Federico Manolio, Wuraola Mosuro, Eliot Jolomi Dikoru",2025,From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria,World Bank Policy Research Working Paper 11125,Working paper,Nigeria,759,"759 students completed the final test (1,328 were assigned)",Experiment,Randomized trial: six weeks of after-school English tutoring with Microsoft Copilot (GPT-4) vs regular schooling,Learning,Tested later without AI,"A test of English, AI knowledge and digital skills, and the end-of-term school exam",Six weeks of after-school tutoring with Microsoft Copilot raised test scores by 0.31 standard deviations (0.23 in English) and end-of-term exam results by 0.21.,six weeks of Copilot tutoring raised scores by 0.31 SD,Better,"Not peer reviewed; many students did not finish (759 of 1,328); teachers guided the AI sessions.",https://doi.org/10.1596/1813-9450-11125,full text,,2026-10-16
fan-2025,"Yizhou Fan, Luzhen Tang, Huixiao Le, Kejie Shen, Shufang Tan, Yueying Zhao, Yuan Shen, Xinyu Li, Dragan Gašević",2025,"Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance","British Journal of Educational Technology 56(2), 489-530",Peer reviewed,China,117,117 university students in four groups,Experiment,Randomized lab experiment: ChatGPT-4 vs human expert vs writing checklist vs no help,Learning,Tested later without AI,"Essay scores, knowledge gain and transfer tests, motivation and self-regulated learning steps","Students helped by ChatGPT improved their essays the most, but their knowledge gain and transfer were no better than the other groups', and they spent less time on the planning and self-checking steps the authors call metacognitive.","better essays with ChatGPT, no more knowledge gained",Mixed,Small groups (25 to 35 students); one lab writing task; 70% women.,https://doi.org/10.1111/bjet.13544,full text,,2026-10-16
fischer-2025,"Mira Fischer, Holger A. Rau, Rainer Michael Rilke",2025,AI Tutoring Enhances Student Learning Without Crowding Out Reading Effort,IZA Discussion Paper 18338 (December 2025),Working paper,Germany,334,334 university students,Experiment,Randomized experiment: textbook only vs an AI tutor after 10 minutes of reading vs the AI tutor throughout,Learning,Tested later without AI,An exam taken without AI,"Access to an AI tutor grounded in the readings raised exam scores by 0.23 standard deviations, and unrestricted access beat access only after 10 minutes of independent reading by 0.21.",an AI tutor raised exam scores by 0.23 SD,Better,"Not peer reviewed; a tutor built on the course readings, not an open chatbot; one study session.",https://docs.iza.org/dp18338.pdf,full text,,2026-10-16
huang-2025,"Siyu Huang, Chang Wen, Xueying Bai, Sihong Li, Shuining Wang, Xiaoxuan Wang, Dong Yang",2025,Exploring the Application Capability of ChatGPT as an Instructor in Skills Education for Dental Medical Students: Randomized Controlled Trial,"Journal of Medical Internet Research 27, e68538",Peer reviewed,China,187,187 dental students,Experiment,Randomized trial: one week of skills training with videos plus ChatGPT vs videos alone,Learning,Tested later without AI,"A virtual-reality test of operative skill, mental effort (eye tracking) and self-efficacy","Students who trained with ChatGPT scored higher on the skills test (73.1 vs 65.5) with less mental effort, and students with weaker spatial ability gained the most.",higher skill-test scores with less mental effort after ChatGPT training,Better,One university; one week; ChatGPT was added on top of the same videos.,https://doi.org/10.2196/68538,full text,,2026-10-16
iqbal-2025,"Javed Iqbal, Zarqa Farooq Hashmi, Muhammad Zaheer Asghar, Muhammad Naseem Abid",2025,Generative AI tool use enhances academic achievement in sustainable education through shared metacognition and cognitive offloading among preservice teachers,"Scientific Reports 15, 16610",Peer reviewed,China,465,465 students at five universities,Survey,Survey in two waves (time-lagged),Learning,What people reported,"Generative AI use, shared metacognition, cognitive offloading and academic achievement (7-point scales)","Student teachers who said they used generative AI more, and expected more from it, reported higher academic achievement.",more AI use went with higher self-reported achievement,Better,Self-reported measures; survey design cannot show cause.,https://doi.org/10.1038/s41598-025-01676-x,full text,,2026-10-16
jia-2025,"Wenxiu Jia, Li Pan, Siobhan Neary",2025,Effect of GenAI Dependency on University Students' Academic Achievement: The Mediating Role of Self-Efficacy and Moderating Role of Perceived Teacher Caring,"Behavioral Sciences 15(10), 1348",Peer reviewed,China,418,418 university students,Survey,Survey (cross-sectional),Learning,What people reported,"Dependence on generative AI, self-efficacy, and academic achievement (survey indicators and grade point average)","Students who depended more on generative AI felt more capable but had lower academic achievement, which the authors describe as a Dunning-Kruger effect.",more AI dependence went with lower achievement,Worse,One snapshot; achievement partly self-reported.,https://doi.org/10.3390/bs15101348,full text,,2026-10-16
kestin-2025,"Greg Kestin, Kelly Miller, Anna Klales, Timothy Milbourne, Gregorio Ponti",2025,AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting,"Scientific Reports 15, 17458",Peer reviewed,United States,194,194 students in one physics course,Experiment,"Randomized crossover trial: a purpose-built AI tutor at home vs in-class active learning, two topics",Learning,Tested later without AI,Physics test scores before and after each lesson,Students learned more than twice as much (median gain) from a tutor built on GPT-4 with teaching rules than from the same lesson taught by in-class active learning.,a purpose-built AI tutor doubled learning gains over active learning,Better,"One course at one university; a tutor designed to guide, not a general chatbot; short-term tests.",https://doi.org/10.1038/s41598-025-97652-6,full text,,2026-10-16
kumar-2025,"Harsh Kumar, David M. Rothschild, Daniel G. Goldstein, Jake M. Hofman",2025,Math Education With Large Language Models: Peril or Promise?,"AIED 2025 (Lecture Notes in Computer Science 15880, pp. 60-75)",Peer reviewed,United States (online),1818,"1,818 US adults in two experiments",Experiment,Two pre-registered randomized experiments: math practice with LLM explanations vs correct answers only,Learning,Tested later without AI,New test questions answered without any help,"Practising with LLM explanations led to more learning than seeing only the correct answers, most of all when people tried the problem first, and even explanations with arithmetic errors beat answers alone.",LLM explanations taught more than correct answers alone,Better,Short online sessions; high-school-level math; explanations shown rather than chat.,https://doi.org/10.1007/978-3-031-98459-4_5,abstract,,2026-10-16
lehmann-2025,"Matthias Lehmann, Philipp B. Cornelius, Fabian J. Sting",2025,AI Meets the Classroom: When Do Large Language Models Harm Learning?,"arXiv preprint 2409.09047 (version 2, March 2025); SSRN 4941259",Preprint,Germany (lab experiments); Netherlands (field study),176,"176 students in two lab experiments (107 and 69), plus a field study of 113 students",Experiment,Two pre-registered randomized lab experiments (ChatGPT during practice vs none) and a field study,Learning,Tested later without AI,A coding post-test taken without AI,"Access to ChatGPT made no overall difference to what students learned; in further analyses, students who used it to produce solutions covered more topics but understood each less well, while those who asked it for explanations understood more.",no overall effect on learning; it depended on how students used it,No clear difference,Not peer reviewed; small lab samples; the usage-pattern findings are exploratory.,https://arxiv.org/abs/2409.09047,full text,,2026-10-16
melumad-2025,"Shiri Melumad, Jin Ho Yun",2025,Experimental evidence of the effects of large language models versus web search on depth of learning,"PNAS Nexus 4(10), pgaf316",Peer reviewed,United States (online panels and a university lab),10462,"10,462 people in seven experiments (the abstract's figure; the text also gives 10,426)",Experiment,"Seven experiments, most pre-registered: learn a topic from ChatGPT or Google AI summaries vs ordinary web links",Learning,Tested later without AI,"How deep people felt their knowledge was, and the length, originality and persuasiveness of advice they then wrote","People who learned a topic from an AI summary instead of web links said their knowledge was shallower and wrote shorter, less original advice that other readers were less likely to follow.",learning from AI summaries left shallower knowledge than web links,Worse,"Depth of knowledge was mostly self-rated; short, one-off learning tasks on everyday topics.",https://doi.org/10.1093/pnasnexus/pgaf316,full text,,2026-10-16
ba-2024,"Hongjun Ba, Lili Zhang, Zizheng Yi",2024,Enhancing clinical skills in pediatric trainees: a comparative study of ChatGPT-assisted and traditional teaching methods,"BMC Medical Education 24, 558",Peer reviewed,China,77,77 paediatric trainees,Experiment,Randomized trial: two weeks of ChatGPT-assisted teaching vs conventional teaching,Learning,Tested later without AI,"A theory exam and an observed clinical-skills assessment (Mini-CEX), both without AI","Trainees taught with ChatGPT did the same as the others on the theory exam but scored higher on the observed clinical-skills assessment, especially patient communication and clinical judgement.","same theory scores, better clinical skills after ChatGPT-assisted teaching",Mixed,Small trial at one hospital; two weeks.,https://doi.org/10.1186/s12909-024-05565-1,full text,,2026-10-16
gan-2024,"Wenyi Gan, Jianfeng Ouyang, Hua Li, Zhaowen Xue, Yiming Zhang, Qiu Dong, Jiadong Huang, Xiaofei Zheng, Yiyi Zhang",2024,Integrating ChatGPT in Orthopedic Education for Medical Undergraduates: Randomized Controlled Trial,"Journal of Medical Internet Research 26, e57037",Peer reviewed,China,110,110 medical students finished (129 started),Experiment,Randomized trial: two weeks of studying orthopaedics with ChatGPT vs with no AI,Learning,Tested later without AI,"An orthopaedics test and end-of-semester exams in other subjects, taken without AI",Students who studied with ChatGPT scored higher on the orthopaedics test (141.2 vs 130.8 points) and on end-of-semester surgery and obstetrics exams.,higher orthopaedics and end-of-term exam scores after studying with ChatGPT,Better,One university; 19 of 129 students dropped out.,https://doi.org/10.2196/57037,full text,,2026-10-16
kavadella-2024,"Argyro Kavadella, Marco Antonio Dias da Silva, Eleftherios G. Kaklamanos, Vasileios Stamatopoulos, Kostis Giannakopoulos",2024,Evaluation of ChatGPT's Real-Life Implementation in Undergraduate Dental Education: Mixed Methods Study,"JMIR Medical Education 10, e51344",Peer reviewed,Cyprus,77,"77 second-year dental students (39 ChatGPT, 38 literature search)",Experiment,Randomized classroom study: prepare an assignment with ChatGPT vs with a literature search,Learning,Tested later without AI,A multiple-choice knowledge exam taken without AI,Students who prepared their assignment with ChatGPT did better on the later knowledge exam than students who searched the literature.,better knowledge-exam results after an assignment done with ChatGPT,Better,Small groups in one course; 70 of 77 sat the exam.,https://doi.org/10.2196/51344,full text,,2026-10-16
pardos-2024,"Zachary A. Pardos, Shreya Bhandari",2024,ChatGPT-generated help produces learning gains equivalent to human tutor-authored help on mathematics skills,"PLOS ONE 19(5), e0304013",Peer reviewed,Online (Mechanical Turk),274,274 adult learners,Experiment,Randomized online experiment: algebra and statistics practice with ChatGPT-written hints vs tutor-written hints vs no hints,Learning,Tested later without AI,"Learning gain from pre-test to post-test, and time on task","Hints written by ChatGPT produced learning gains no different from hints written by human tutors, and only the ChatGPT hints beat no help, though 32% of the ChatGPT hints failed the authors' quality checks before fixes.",ChatGPT-written hints taught as well as tutor-written hints,Better,"Short online sessions; hints were pre-written and checked, not live chat.",https://doi.org/10.1371/journal.pone.0304013,full text,,2026-10-16
kazemitabaar-2023,"Majeed Kazemitabaar, Justin Chow, Carl Ka To Ma, Barbara J. Ericson, David Weintrop, Tovi Grossman",2023,Studying the effect of AI Code Generators on Supporting Novice Learners in Introductory Programming,"CHI 2023 (ACM conference), doi 10.1145/3544548.3580919",Peer reviewed,Canada,69,"69 children and teenagers aged 10-17 (33 with AI, 36 without)",Experiment,Randomized experiment (matched pairs): learn Python with or without the Codex code generator over 10 sessions,Learning,Tested later without AI,"Coding tasks with AI; code-modification tasks and tests without AI, including one a week later","Learners with the AI code generator finished more coding tasks and scored 1.8 times higher while using it, and did no worse on coding without it, including on a test a week later.",coding with AI helped and did not hurt skills a week later,No clear difference,Small groups of young volunteers; an early code model (Codex); the one-week difference was not statistically significant.,https://doi.org/10.1145/3544548.3580919,full text,,2026-10-16
dellacqua-2026,"Fabrizio Dell’Acqua, Edward McFowland, Ethan Mollick, Hila Lifshitz, Katherine C. Kellogg, Saran Rajendran, Lisa Krayer, François Candelon, Karim R. Lakhani",2026,Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality,"Organization Science 37(2), 403-423 (2026); first circulated as HBS Working Paper 24-013 (2023)",Peer reviewed,Not stated (Boston Consulting Group consultants),758,758 consultants,Experiment,"Pre-registered randomized experiment: realistic consulting tasks with no AI, GPT-4, or GPT-4 plus prompting training",Critical thinking,Work done with AI,"Speed, quality and correctness of answers to consulting tasks","On tasks GPT-4 could handle, consultants using it completed 12.2% more tasks, 25.1% faster and at higher quality, but on a task chosen to be beyond its abilities they were 19% less likely to reach the correct answer.","GPT-4 made consultants faster and better, but 19% less often right on a task beyond it",Mixed,"One firm's consultants; one session; measures work done with AI, not skill afterwards.",https://doi.org/10.1287/orsc.2025.21838,abstract,,2026-10-16
kircaburun-2026,Kağan Kırcaburun,2026,Generative AI Use and Critical Thinking Dispositions in Higher Education: A Cross-Sample Study of the Sequential Role of Metacognitive Weakness and Epistemic Laziness,"Journal of Intelligence 14(7), 147",Peer reviewed,Turkey and online (English-speaking),659,"659 students (441 Turkish-speaking, 218 English-speaking)",Survey,Survey in two samples (cross-sectional),Critical thinking,What people reported,"How much students use generative AI, critical thinking dispositions when using it, metacognitive weakness and epistemic laziness","In both samples, students who used generative AI more reported weaker critical-thinking habits when using it, linked to weaker self-monitoring and less willingness to think things through.",more AI use went with weaker critical-thinking habits,Worse,Self-reported measures at one point in time; the link runs through indirect paths in a model.,https://doi.org/10.3390/jintelligence14070147,full text,,2026-10-16
yan-2026,"Zhen Yan, Zhuotao Fang, Weixing Zou",2026,The association between generative AI use and university students' critical thinking: a moderated mediation model of GenAI feedback literacy and reflective thinking,"Frontiers in Psychology 17, 1939286",Peer reviewed,China,421,421 undergraduates,Survey,Survey (cross-sectional),Critical thinking,What people reported,"Generative AI use, feedback literacy, reflective thinking and self-reported critical thinking","Students who used generative AI more reported stronger critical thinking, almost entirely through being better at judging and using AI feedback.",more AI use went with stronger self-reported critical thinking,Better,Self-reported measures at one point in time.,https://doi.org/10.3389/fpsyg.2026.1939286,full text,,2026-10-16
zhi-2026,"Jiayin Zhi, Harsh Kumar, Mina Lee",2026,Investigating the Effects of LLM Use on Critical Thinking Under Time Constraints: Access Timing and Time Availability,"CHI 2026 (ACM conference), doi 10.1145/3772318.3791796",Peer reviewed,United States (online panel),393,393 adults,Experiment,"Pre-registered randomized experiment: chatbot available early, throughout, late or never, with 10 or 30 minutes",Critical thinking,Work done with AI,A standard critical-thinking task: a reasoned written decision based on documents of mixed quality,Using the chatbot from the start helped under time pressure but made the written reasoning worse when people had enough time; starting alone and using the chatbot late showed the opposite pattern.,"AI from the start helped under time pressure, hurt reasoning when time was ample",Mixed,One task and one session with online participants.,https://doi.org/10.1145/3772318.3791796,full text,,2026-10-16
chang-2025,"Li-Chun Chang, Li-Ling Hung, Ting-Wei Liu, Chun-Hsia Huang, Hui-Ling Lin, Li-Ling Liao",2025,Relationships between ChatGPT use with self-directed learning and critical thinking among school and university nurses in Taiwan,"BMC Nursing 24, 1426",Peer reviewed,Taiwan,413,413 nurses (from a random sample of 600),Survey,Survey of a stratified random sample (cross-sectional),Critical thinking,What people reported,"ChatGPT use, self-directed learning and a critical thinking inventory","Nurses who used ChatGPT more often scored higher on critical thinking, although self-directed learning had the strongest link.",more ChatGPT use went with higher critical thinking in nurses,Better,Self-reported measures at one point in time; one profession.,https://doi.org/10.1186/s12912-025-04069-7,full text,,2026-10-16
gerlich-2025,Michael Gerlich,2025,AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking,"Societies 15(1), 6",Peer reviewed,United Kingdom,666,"666 valid responses (669 took part), plus interviews",Survey,Survey with interviews (cross-sectional),Critical thinking,What people reported,"How often people use AI tools, how much they offload thinking to devices, and a 23-question critical-thinking questionnaire","Heavier AI-tool users scored lower on critical thinking (correlation r = -0.68), largely through more cognitive offloading, and younger participants relied on AI most and scored lowest.",heavier AI-tool use went with lower critical thinking (r = -0.68),Worse,"One snapshot, so it shows a link, not a cause; questionnaire measures; a convenience sample recruited on social media.",https://doi.org/10.3390/soc15010006,full text,"A correction (September 2025) replaced Table 4, which had been printed as a copy of Table 3; the author states the conclusions are unaffected.",2026-10-16
kulal-2025,Abhinandan Kulal,2025,"Cognitive Risks of AI: Literacy, Trust, and Critical Thinking","Journal of Computer Information Systems (online 2025), 1-13",Peer reviewed,Not stated in the abstract,625,"625 postgraduate students, teachers and research scholars",Survey,Survey (cross-sectional),Critical thinking,What people reported,"AI use, trust in AI, AI literacy and critical thinking (questionnaires)","More AI use went with lower critical thinking, more so among people who trusted AI more, and AI literacy softened the effect of trust.",more AI use went with lower critical thinking,Worse,Self-reported measures at one point in time; single author.,https://doi.org/10.1080/08874417.2025.2582050,abstract,,2026-10-16
lee-2025,"Hao-Ping (Hank) Lee, Advait Sarkar, Lev Tankelevitch, Ian Drosos, Sean Rintel, Richard Banks, Nicholas Wilson",2025,The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers,"CHI 2025 (ACM conference), doi 10.1145/3706598.3713778",Peer reviewed,"Several countries (online panel; most often UK, Canada, US)",319,319 knowledge workers describing 936 real work tasks,Survey,Survey of real work tasks (cross-sectional),Critical thinking,What people reported,"Whether people said they used critical thinking on each task, and how much effort it took with AI","Workers reported doing critical thinking on 555 of 936 AI-assisted tasks (59%); the more they trusted the AI, the less critical thinking they reported, and the more they trusted themselves, the more they reported.",more trust in AI went with less critical thinking at work,Worse,"Self-reports of thinking, not tests of it; one snapshot in time, so it shows a link, not a cause.",https://doi.org/10.1145/3706598.3713778,full text,,2026-10-16
tian-2025,"Jinrui Tian, Ronghua Zhang",2025,Learners' AI dependence and critical thinking: The psychological mechanism of fatigue and the social buffering role of AI literacy,"Acta Psychologica 260, 105725",Peer reviewed,China,580,580 university students,Survey,Survey (cross-sectional),Critical thinking,What people reported,"AI dependence, critical thinking, mental fatigue and information literacy (questionnaires)","Students who depended more on AI reported lower critical thinking, partly through more mental fatigue.",more AI dependence went with lower critical thinking,Worse,Self-reported measures at one point in time; a corrigendum was published in 2026 that we could not read.,https://doi.org/10.1016/j.actpsy.2025.105725,abstract,"Corrigendum: Acta Psychologica 267, 107179 (2026). Its text is behind a site check we did not pass, so we could not see what it changes.",2026-10-16
stadler-2024,"Matthias Stadler, Maria Bannert, Michael Sailer",2024,Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry,"Computers in Human Behavior 160, 108386",Peer reviewed,Germany,91,"91 university students (44 ChatGPT, 47 Google)",Experiment,Randomized lab experiment: research a science question with ChatGPT-3.5 or with Google,Critical thinking,Work done with AI,Mental effort (cognitive load) and the quality of students' written recommendations and justifications,Students who researched with ChatGPT reported less mental effort but wrote lower-quality reasoning and justifications than students who used Google.,less mental effort but weaker reasoning than with Google,Worse,One topic (nanoparticles in sunscreen); one session; ChatGPT-3.5.,https://doi.org/10.1016/j.chb.2024.108386,full text,,2026-10-16
fernandes-2026,"Daniela Fernandes, Steeven Villa, Salla Nicholls, Otso Haavisto, Daniel Buschek, Albrecht Schmidt, Thomas Kosch, Chenxinran Shen, Robin Welsch",2026,AI makes you smarter but none the wiser: The disconnect between performance and metacognition,"Computers in Human Behavior 175, 108779 (Feb 2026)",Peer reviewed,United States (online panel),698,246 in study 1 and 452 in study 2,Experiment,Two experiments: LSAT logic problems with ChatGPT-4o; study 2 added a no-AI group,Metacognition,Work done with AI,"Logic-test scores with AI, and how well people estimated their own score","ChatGPT raised scores on 20 LSAT logic problems by about three points, but users overestimated their score by about four points, and those who knew more about AI judged themselves less accurately.","logic scores rose with ChatGPT, but people overrated how well they did",Mixed,One type of test; people were required to use the AI; study 1 compared with published scores rather than its own control group.,https://doi.org/10.1016/j.chb.2025.108779,full text,,2026-10-16
zhu-2026,"Qiuhan Zhu, Xiangnan Li, Yiang Dong, Pengcheng Chang, Mengmeng Fan",2026,Not all cognitive offloading is equal: distinguishing dependent and autonomous offloading to generative AI,"Frontiers in Psychology 17, 1878629",Peer reviewed,China*,589,589 students and early-career knowledge workers,Survey,Survey in three waves (time-lagged),Offloading,What people reported,"Two kinds of offloading to AI and self-rated independent capability, creativity, deep processing and judgement","Handing core thinking to AI went with worse self-rated thinking outcomes, while using AI as a scaffold went with better ones.","handing thinking to AI went with worse outcomes; using it as a scaffold, better",Mixed,Self-reported outcomes; survey design cannot show cause.,https://doi.org/10.3389/fpsyg.2026.1878629,full text,,2026-10-16
zhou-2026,"Yiyong Zhou, Qinghan Liu, Jihao Huang, Guiquan Li",2026,Creative scar without generative AI: Individual creativity fails to sustain while homogeneity keeps climbing,"Technology in Society 84, 103087 (2026); preprint arXiv 2401.06816 (2024)",Peer reviewed,China,61,"61 college students (31 with ChatGPT, 30 without)",Experiment,"Pre-registered 7-day lab experiment with ChatGPT vs without, plus a 30-day follow-up",Creativity,Tested later without AI,"Creativity and similarity of 3,302 ideas and 427 solutions, with ChatGPT and after it was taken away","ChatGPT boosted creative performance for five days, but scores fell back to baseline once it was taken away, on day 7 and day 30, while the ideas of former users kept becoming more alike.",the creativity boost vanished without ChatGPT while ideas stayed alike,Worse,Small sample at one university; figures are from the 2024 preprint of the published paper.,https://doi.org/10.1016/j.techsoc.2025.103087,full text,,2026-10-16
doshi-2024,"Anil R. Doshi, Oliver P. Hauser",2024,Generative AI enhances individual creativity but reduces the collective diversity of novel content,"Science Advances 10(28), eadn5290",Peer reviewed,United Kingdom (online panel),293,293 writers; 600 readers rated the stories,Experiment,"Pre-registered randomized experiment: write a short story with 0, 1 or up to 5 ideas from an LLM",Creativity,Work done with AI,"Novelty, usefulness and similarity of the stories, rated by readers who did not know the condition","Writers who could see up to five AI story ideas produced stories rated 8.1% more novel and 9.0% more useful, with the biggest gains for less creative writers, but the AI-assisted stories were more similar to one another.",AI story ideas made stories more creative and more alike,Mixed,"Eight-sentence stories written in one sitting; creativity was measured with AI help, not afterwards.",https://doi.org/10.1126/sciadv.adn5290,full text,,2026-10-16
leechung-2024,"Byung Cheol Lee, Jaeyeon Chung",2024,An empirical investigation of the impact of ChatGPT on creativity,"Nature Human Behaviour 8, 1906-1914",Peer reviewed,Not stated in the abstract,,Five experiments (the abstract gives no totals; the full paper is paywalled),Experiment,"Five randomized experiments: generate ideas with ChatGPT (GPT-3.5), with Google, or with no technology",Creativity,Work done with AI,"Creativity ratings of the ideas, by evaluators","Ideas produced with ChatGPT were rated more creative than ideas produced with Google or without technology, mostly for incremental rather than radical ideas; a later reanalysis of the same data found the ChatGPT ideas were less varied as a group.",ChatGPT ideas rated more creative; a reanalysis found them less varied,Mixed,"Ideas were produced with ChatGPT's help, so this measures the output, not the person's own creativity afterwards.",https://doi.org/10.1038/s41562-024-01953-1,abstract,,2026-10-16
urban-2024,"Marek Urban, Filip Děchtěrenko, Jiří Lukavský, Veronika Hrabalová, Filip Svacha, Cyril Brom, Kamila Urban",2024,ChatGPT improves creative problem-solving performance in university students: An experimental study,"Computers & Education 215, 105031",Peer reviewed,Czech Republic*,145,"145 university students (77 with ChatGPT, 68 without)",Experiment,Pre-registered experiment: a complex creative problem with or without ChatGPT,Creativity,Work done with AI,"Quality, elaboration and originality of solutions, mental effort and self-judgement accuracy","Students with ChatGPT produced better, more detailed and more original solutions to a creative problem with less mental effort, but their self-judgements no longer tracked how well they did.","better, more original solutions with ChatGPT and less mental effort",Better,"Creativity was measured with ChatGPT available, not afterwards; one task.",https://doi.org/10.1016/j.compedu.2024.105031,abstract,,2026-10-16
agarwal-2025,"Dhruv Agarwal, Mor Naaman, Aditya Vashistha",2025,AI Suggestions Homogenize Writing Toward Western Styles and Diminish Cultural Nuances,"CHI 2025 (ACM conference), doi 10.1145/3706598.3713564",Peer reviewed,India and United States (online panel),118,"118 adults (60 in India, 58 in the US)",Experiment,Randomized experiment: culturally grounded writing tasks with or without GPT-4o autocomplete suggestions,Writing,Work done with AI,Writing speed and the style and content of what people wrote,"AI writing suggestions saved Americans more time than Indians and pulled Indian participants' writing toward Western styles, making the two groups' writing more alike.",AI suggestions pulled Indian writers toward Western styles,Worse,Short English writing tasks; one autocomplete tool; two countries.,https://doi.org/10.1145/3706598.3713564,full text,,2026-10-16