Undergraduate research is credited with helping students develop cognitive abilities along with an understanding of the way that research happens, among other benefits, when done well. Some characteristics of high-quality undergraduate research include working on unsolved research problems (Bhattacharyya, Chan, and Waraczynski 2018) working closely with faculty  having some autonomy in the research decisions (Gilmore et al. 2016), and being integrated into the larger research community (Thiry et al. 2012). Just as faculty and corporate researchers have had to grapple with the affordances and risks of generative AI in their own research process, undergraduate researchers can face some of the same questions, but with special issues caused by their developmental role. In this post, I look at a few component tasks of more quantitative undergraduate research with an eye toward how AI support might affect the quality of student development. 

How Generative AI Is Changing Undergraduate Research 

Undergraduate research has specific learning goals. For example, the University of Missouri offers a list (University of Missouri Undergraduate Research n.d.) that can be summarized as: 

• Learning to develop research questions based in the discipline 
• Learning appropriate methods to investigate the question
• Assessing the results
• Communicating the results   
• Evaluating and citing appropriate literature 
• Develop skills in disciplinary methods 
• Conducting ethical research  
• Reflecting on the implications of the work 

When it comes to developing research questions, LLMs use can be both positive and negative for emerging scholars such as undergraduate researchers. There’s the question of whether the tools can do a good job, which scholars such as Yao et al. have examined, generally finding that although the generated hypotheses are not all winners, there is promise. However, studies like this tend to rely on expert evaluators to assess the quality of the hypotheses generated and that’s a real issue for novices for two reasons. The first is the lack of expertise and potentially the lack of opportunity to develop it. In my experience teaching research methods to undergraduates, problem definition is sometimes the most difficult challenge for them. Questions of theoretical interest aren’t too hard, but mastering filtering the questions through measurement issues and pragmatic issues like access to the thing to be measured, cost, time, etc. is a steep learning curve. AI-supported problem definition also might diminish the sense of curiosity that accompanies academic research. It is one of the more fun parts.  

The Risks and Benefits of AI Tools for Student Researchers 

When it comes to reviewing the literature, AI support might also be fraught. Research support tools like Semantic Scholar and Elicit have aspects that are helpful such as knowing a paper’s citation history at a glance. However, the papers these tools find do sometimes trend towards pre-prints and other unpublished reports. Experienced scholars, may be competent to judge the quality of work and the ability of a scholar to create it. For students, however, that ability is likely still developing. Relying on “smart” tools to identify papers could lead to ignorant choices of work should be included. It can also be difficult to know when the search should end. Caution is needed here.  

Generative AI systems can also summarize whole papers. As undergraduate research has become a valued activity in the academy, it has pushed both more broadly and earlier in the curriculum, which presents unintended consequences. One of these is that academic papers can be extremely difficult to read. The nature of undergraduate education means managing both major and core breadth courses as once, as well as expectations for out-of-class engagement and leadership. Undergraduate research is but a dish on a buffet of time-consuming experiences, yet it requires learners to acquire substantial skills. Summarizing existing research is a tempting application of GenAI, particularly for younger students. But, off-loading the reading for a literature review entirely could deny students the chance to learn the skill of interrogating the work of others. There is potential here – the summaries could play a role in helping students to develop that skill as they, for example, compare the summary to the full paper.  

Using Generative AI in Research Methods, Data Analysis, and Academic Writing 

There is also a learning curve in actually conducting research. Although generative AI likely can’t set your budget or know with nuance the ins and outs of a particular IRB, prompted well, it has promise as a thought partner. For example, a scholar could upload a research protocol and ask a tool “What aspects of doing this research would cost money?” Possibly, GenAI could also help by suggesting language for informed consent, instructions to participants, and more.  

When it comes to learning particular methods, this is field-dependent. Obviously, an AI is not going to teach a student how to load a centrifuge or feed a rat. For things like data analysis, professionals in research, business, journalism and more are widely using GenAI tools, and this seems like a plausible area to explore with students. However, tasks like this are usually advised to be “human-in-the-loop” so that the analytical results are evaluated to make sure they make sense in context.   

The thought partner role could also apply to drawing conclusions as well. A growing body of research has looked at the efficacy of using GenAI to data analysis, often comparing results from humans and tools. In the qualitative realm, some scholars have found a good amount of overlap in identifying themes from unstructured text (Wachinger et al. 2024; Zhou et al. 2025), in some cases nearly as well as manual coding. But there are risks as well. Just as GenAI can hallucinate citations, it can also make up convincing, but wrong identifications of patterns in data. Coding of open data is in some ways subjective, but use of LLMs could lead to reproducibility issues (Ball 2023).  There are issues with quantitative work as well. GenAI can produce correct code to run (Song, et al. 2025) statistical tests but may not choose appropriate tests for the problem. Although researchers have noted that GenAI tools can democratize some types of analysis (Kim and Jeong 2024) by removing the barrier of learning coding, the utility of the human-in-the-loop pervades analysis as well. For research trainees, relying on GenAI tools for the analysis portion of work may mean missing important skills needed to be that human.  

There are other posts in the series about the affordances of AI tools for writing. For students who are not used to writing in the academic format, a GenAI tool might add a measure of comfort for them. Again, however, if undergraduate research is done to train students to be researchers in the future, the ability to write up their own ideas is a meaningful part of the research process. You would not want to let AI tools take away from that learning. 

References 

Ball, Philip. 2023. “Is AI Leading to a Reproducibility Crisis in Science?” Nature 624 (7990): 22–25. https://doi.org/10.1038/d41586-023-03817-6. 

Bhattacharyya, Prajukti, Catherine W. M. Chan, and Meg Waraczynski. 2018. “How Novice Researchers See Themselves Grow.” International Journal for the Scholarship of Teaching and Learning 12 (2): 3. https://doi.org/10.20429/ijsotl.2018.120203. 

Council on Undergraduate Research. n.d. “Council on Undergraduate Research Issues Updated Definition of Undergraduate Research.” Accessed May 15, 2026. https://www.cur.org/council-on-undergraduate-research-issues-updated-definition-of-undergraduate-research/. 

Gilmore, Joanna, Michelle Vieyra, Briana Timmerman, David Feldon, and Michelle Maher. 2016. “The Relationship between Undergraduate Research Participation and Subsequent Research Performance of Early Career STEM Graduate Students.” Journal of Higher Education 86 (6): 834–63. https://doi.org/10.1080/00221546.2015.11777386. 

Kim, Seongmin, and Jiyong Jeong. 2024. “Examining Large Language Models’ Code Generation Abilities: Utilizing Deep Learning for Employee Attrition Prediction.” 아시아태평양융합연구교류논문지 10 (6): 157–69. English version accessed May 15, 2026. http://apjcriweb.org/content/vol10no6/13.pdf.

Song, Xinyi, Kexin Xie, Lina Lee, Ruizhe Chen, Jared M. Clark, Hao He, Haoran He, et al. 2025. “Performance Evaluation of Large Language Models in Statistical Programming.” arXiv. https://doi.org/10.48550/arXiv.2502.13117. 

Thiry, Heather, Timothy J. Weston, Sandra L. Laursen, and Anne-Barrie Hunter. 2012. “The Benefits of Multi-Year Research Experiences: Differences in Novice and Experienced Students’ Reported Gains from Undergraduate Research.” CBE—Life Sciences Education 11 (3): 260–72. 

University of Missouri Undergraduate Research. n.d. “Learning Objectives for Undergraduate Research at MU.” Accessed May 15, 2026. Learning Goals. https://undergradresearch.missouri.edu/about/learning-goals/. 

Wachinger, Jonas, Kate Bärnighausen, Louis N. Schäfer, Kerry Scott, and Shannon A. McMahon. 2024. “Prompts, Pearls, Imperfections: Comparing ChatGPT and a Human Researcher in Qualitative Data Analysis.” Qualitative Health Research 35 (9): 951–966. https://doi.org/10.1177/10497323241244669.

Yao, Lan, Heliang Yin, Chengyuan Yang, Shuyan Han, Jiamin Ma, J. Carolyn Graff, Cong-Yi Wang, et al. 2025. “Generating Research Hypotheses to Overcome Key Challenges in the Early Diagnosis of Colorectal Cancer: Future Application of AI.” Cancer Letters 620: 217632. https://doi.org/10.1016/j.canlet.2025.217632. 

Zhou, Y., Y. Yuan, K. Huang, and X. Hu. 2025. “Can ChatGPT Perform a Grounded Theory Approach to Do Risk Analysis? An Empirical Study.” Journal of Management Information Systems 41 (4): 982–1015. https://doi.org/10.1080/07421222.2024.2415772. 

About the Author  

Amanda Sturgill, associate professor of journalism, is the 2024-2026 CEL Scholar. Her work focuses on the intersection of artificial intelligence (AI) and engaged learning in higher education. Dr. Sturgill also previously contributed posts on global learning as a seminar leader for the 2015-2017 research seminar on Integrating Global Learning with the University Experience.  

How to Cite This Post 

Sturgill, Amanda. 2026. “The Affordances and Risks of Generative AI for Training Undergraduate Researchers.” Center for Engaged Learning (blog). Elon University. May 26 2026. https://www.centerforengagedlearning.org/the-affordances-and-risks-of-generative-ai-for-training-undergraduate-researchers.