Using generative AI for interview simulations to enhance student research skills in biology education

ABSTRACT

The Longevity Games Interview Simulator provides an innovative approach to preparing students for real-world research interactions by leveraging the capabilities of large language models (LLMs) like OpenAI’s GPT-4o and Claude-3.7. This paper outlines the development and demonstrates the benefits of the simulator, designed to mimic interviews with older adults to enhance students’ interviewing skills, empathy, and cultural competence. Key outcomes included preparing students for real-world interactions with interview subjects, improving their ability to identify and properly document protected health information (PHI), gaining experience in asking relevant follow-up questions, and directing conversations to achieve interview goals. The simulator used generative AI models to create realistic interview scenarios based on demographic data from Rochester, NY. Components of the simulator included a student interview-question selection and creation portion, an interview-guide worksheet, a post-simulation quiz on the materials, and a reflective exercise focusing on information gathering and ethical considerations regarding PHI. This tool was designed for the Science of Aging course’s CURE (Course-Based Undergraduate Research Experience) to provide students with practical, repeatable interview practice. A small pilot study with senior nursing students indicated that the simulator improved students’ confidence, preparedness, and understanding of ethical considerations. This paper also discusses how the simulator has potential for adaptation across educational contexts and encourages educators to develop their own custom interview simulations.

INTRODUCTION

Biology education often emphasizes coursework and labs, but real-world interactions build empathy and deepen student understanding of research subjects (1–4). Classroom role-play helps, but lacks the distinctiveness of real conversations (5). To address this observed gap in the Science of Aging course (BIOL 206), we leveraged the power of a Large Language Model (LLM), OpenAI’s ChatGPT, to simulate the interview process, thus preparing students for real interactions.

The Science of Aging course includes a CURE where students assess older adults’ health and design personalized plans to promote healthy aging. Called the Longevity Games, this CURE helps older adults “compete” for improved longevity—physical, cognitive, and social. Further details on the structure and goals of the Longevity Games are available in an attached Appendix S1 and through our website (https://bridges-across-generations.weebly.com/science-of-aging.html). Students needed a way to practice respectful, sensitive interviews in a controlled, repeatable format.

To fill this gap in the Science of Aging course (BIOL 206), we used OpenAI’s ChatGPT to simulate interviews and prepare students for real interactions. In a small pilot study with senior nursing students about to enter a geriatric clinical rotation, the simulator supported growth in confidence, ethical awareness, and comfort discussing age-related topics. This aligns with Constructivist and Experiential Learning Theories that emphasize active learning and reflection (6–8). It also demonstrates how AI-driven simulations can be used in undergraduate education to enhance communication skills before engaging with real-world subjects.

To aid in understanding the generative AI and chatbot terminology used throughout this paper, Table 1 provides definitions and their relevance to the content of this study.

TABLE 1.

AI and chatbot terms and definitions^a

Term	Definition	Connection to this paper/content
AI (Artificial Intelligence)	Artificial Intelligence (AI) is defined as a field of study focused on creating systems that exhibit intelligent behavior by analyzing their environment and taking actions with a degree of autonomy to achieve specific goals. It involves the development of algorithms and technologies that can perform tasks typically requiring human intelligence, such as reasoning, learning, and decision-making (9).	AI is the foundation of the tools used in this paper, such as ChatGPT and DALL-E, enabling the creation of realistic interview simulations.
Generative Artificial Intelligence	Generative AI refers to deep-learning models capable of creating high-quality text, images, and other media from training data. These models leverage neural networks to learn patterns and generate new, similar outputs based on prompts, with applications ranging from content creation to scientific discovery (10).	Generative Artificial Intelligence is key to the production of an interview simulation that can respond to a wide variety of questions and provide novel responses.
LLM (Large Language Model)	A type of artificial intelligence that uses machine learning to understand, generate, and manipulate human language. LLMs are trained on vast amounts of text data (11).	LLMs like OpenAI’s GPT-4 and Claude3 are used to simulate realistic interviews with older adults in the LGIS.
Chatbot	A computer program designed to simulate conversation with human users, especially over the internet (12).	The Custom GPT used in the LGIS is a specialized chatbot programmed to mimic conversations with older adults.
GPT (Generative Pre-trained Transformer)	A type of large language model developed by OpenAI that uses deep learning to produce human-like text. GPT models are pre-trained on diverse internet text data and can generate coherent and contextually relevant responses based on the input they receive (13).	The LGIS utilizes OpenAI’s GPT-4o to create dynamic and realistic interview scenarios.
Custom GPT	A tailored version of a standard GPT model with specific instructions to deliver targeted responses (14).	The Custom GPT in this paper is programmed with specific instructions to simulate interactions with older adults, incorporating demographic and health-related data.
DALL-E	An AI model developed by OpenAI that generates images from textual descriptions (15).	DALL-E 3 is used to create images of the simulated older adults based on randomized data, enhancing the realism of the interview simulation.
Natural Language Processing	Natural Language Processing is a branch of artificial intelligence that enables computers to understand, interpret, and generate human language, forming the foundation for large language models (LLMs) that can process and produce text in ways similar to human communication (16).	LLMs Like OpenAI’s ChatGPT4 and Claude3 use NLP to understand user input and respond in kind.

^a Definitions of essential AI and chatbot terms as they relate to the Longevity Games Interview Simulator, which uses tools like GPT-4o and DALL-E 3 to create realistic, ethical, and educational interview experiences.

Beyond the Longevity Games, practical simulations benefit students by preparing students for practical experiences (17–19). Unlike scripted interactions, LLMs create dynamic, realistic simulations from general instructions and conversational language, enhancing reusability and effectiveness.

This paper demonstrates the creation of the LGIS using Generative-AI, outlines the integration into a CURE, highlights its goals, presents a pilot study on its impact on older-adult engaged students, and its broader educational applications.

PROCEDURE

During the Longevity Games (CURE), students interview older adults to gather vital information on personal history, diet, social interaction, and activity levels, with the aim of understanding and improving their physical and cognitive health. To help prepare students for these real-world interactions, we developed the LGIS, an AI-powered tool designed to provide realistic, low-risk, repeatable interview practice with simulated older adults.

Students select and customize questions or create their own (Appendix S2). The simulator, built with a Custom GPT and DALL·E 3, generates realistic older adult profiles and portraits based on Rochester demographics (see Table 1; Appendix S3). To ensure HIPAA compliance, the simulator includes information on protected health information (PHI) relevant to research contexts (20). Detailed instructions for creating a Custom GPT are provided in Appendix S3.

While designed for the Science of Aging (BIOL 206), we conducted a pilot evaluation with senior nursing students who were preparing to enter clinical rotations focused on geriatric care. As BIOL 206 was not offered in 2024–2025, we piloted the simulator with senior nursing students entering geriatric rotations, for whom the tool is immediately relevant. These students tested the simulator’s effectiveness and usability in an academic setting. Students completed pre- and post-surveys on comfort, confidence, and ethical/legal awareness, including HIPAA (Appendices S4–S5).

Students conducted respectful PHI-aware conversations using open-ended and follow-up questions. Afterward, they recorded their observations and reflections in the worksheet (Appendix S6), completed the embedded quiz, and documented their rationale for each quiz response. They then submitted the completed materials along with the post-survey for evaluation.

While the LGIS is a specific use case, the concept can be broadly applied. To set up an interview simulation for different contexts, follow the steps outlined in Appendix S3. These steps can be entered into OpenAI’s Custom GPT generator (see Appendix S3) or can be used to generate a detailed prompt to instruct an LLM to run a similar interview simulation (see Appendix S7).

These steps help educators create interview simulations tailored to their needs. The LGIS serves as a practical example of how such a tool can be implemented to enhance student learning and preparation for real-world interactions.

CONCLUSION

To evaluate the LGIS prior to classroom deployment, we conducted a small pilot study with six senior-level nursing students preparing to enter geriatric rotations. Students completed pre- and post-simulation surveys assessing their comfort and confidence in communication and ethics. Surveys included Likert-scale and open-ended questions (see Appendices S4 and S5).

Students reported improved comfort and confidence post-simulation. Interview comfort showed the greatest improvement (median score: 3.0–4.0, P = 0.030; Fig. 1A and B). Communication confidence and awareness of ethical issues also showed modest gains (insignificant P > 0.05). Students with prior experience in geriatric care reported smaller shifts, suggesting the simulator may be most impactful for learners with limited prior exposure. Students valued the low-pressure environment. Some found responses overly polished, but still useful.

Fig 1.

Improvements in student comfort, confidence, and awareness following the use of the Interview Simulator. (A) Average Likert-scale scores (1 = strongly disagree, 5 = strongly agree) from pre- and post-simulation surveys completed by senior nursing students (n = 6) assessing comfort interviewing older adults, communication confidence, ethical understanding, and HIPAA awareness. Only the increase in interviewing comfort reached statistical significance (P = 0.030, Wilcoxon signed-rank test), while the increases in communication confidence (P = 0.083) and ethical understanding (P = 0.157) did not reach a significance threshold of ≤0.05. (B) Distribution of Likert responses for the “Comfort interviewing older adults” item before and after simulator use, showing a shift from disagreement and neutrality toward greater agreement. Stacked bars represent the percentage of students selecting each response option at each timepoint.

Five out of six students felt the simulator helped improve their interviewing skills, and four recommended it be used earlier in the nursing curriculum for maximum benefit. Participants noted that the guided format and opportunity to reflect on ethical challenges made the tool particularly helpful as preparation for real patient interactions.

The LGIS provides practical training for sensitive interviews with older adults. Students, especially those with limited experience, gained confidence and ethical awareness. The simulator also introduced students to key ethical considerations, including HIPAA compliance, through interactive reflection and guided questioning.

While early results are promising, feedback suggests that the LGIS may be most effective in early or non-clinical learners. Some improvements in realism were also suggested. Clear steps and flexible customization make the simulator adaptable across health sciences and biology education.

Planned use in BIOL 206 is expected to expand its application and allow additional evaluation with students who have limited prior exposure to older adult care. Instructors are encouraged to explore the included setup instructions and consider integrating the simulator into their own teaching to provide students with critical interview experience in a low-risk environment.

For practical advice on employing an interview simulator, educators can refer to Appendix S8, which outlines various use cases across different educational disciplines. This demonstrates the simulator’s versatility and potential impact in the classroom.

Future deployments will use iterative design based on student and older adult feedback. In addition, exploring advanced AI technologies to enhance realism, such as simulated audio and video, voice recognition, and body language recognition, will be key.

See Appendix S9 for limitations and lessons learned. Educators can explore the simulator (https://chatgpt.com/g/g-FGcO2aOdX-longevity-games-interview-simulator) and adapt it for diverse courses.

ETHICS APPROVAL

All activities involving student participants were reviewed and determined Exempt, Category 1 (educational practices), by the St. John Fisher University Institutional Review Board (IRB #4457-011625-01); written informed consent was obtained from every participant prior to data collection, and all student data were secured in a password-protected computer system.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/jmbe.00122-25.

Appendix S1. jmbe.00122-25-s0001.docx.

Overview of the course-based undergraduate research experience and its six-phase model.

Appendix S2. jmbe.00122-25-s0002.docx.

Full list of semi-structured interview questions used to guide student interviews.

Appendix S3. jmbe.00122-25-s0003.docx.

Step-by-step guide to creating the interview simulator using OpenAI's Custom GPT tool (21, 22).

Appendix S4. jmbe.00122-25-s0004.docx.

Survey administered before simulation to assess baseline comfort and knowledge.

Appendix S5. jmbe.00122-25-s0005.docx.

Survey used after simulation to assess learning outcomes and simulator usability.

Appendix S6. jmbe.00122-25-s0006.docx.

Worksheet used during the simulation for guided questioning, reflection, and HIPAA quiz.

Appendix S7. jmbe.00122-25-s0007.docx.

A detailed prompt for using GPT-4o in a free ChatGPT (or other LLM) session to simulate interviews.

Appendix S8. jmbe.00122-25-s0008.docx.

Examples of how the simulator can be used in various educational fields beyond biology.

Appendix S9. jmbe.00122-25-s0009.docx.

Summary of feedback, lessons, and identified limitations from the pilot testing phase.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.