The AI Buddy Students Actually Asked For

By Jean Rhodes

What do college students want from an AI companion? When researchers at the University of Zurich surveyed nearly 1,000 students about a hypothetical “AI Buddy,” the most popular answers had nothing to do with friendship. Most (78%) just wanted an editor and proofreader, about 69% wanted help managing references, and 59% wanted help organizing their time. Far fewer wanted help with social networking (34%) or a discussion partner (31%), despite the social role the researchers had envisioned (Martínez Moreno et al., 2026). This preference for practical assistance raises useful questions about where AI belongs in student support.

Conducted in spring 2024, the Zurich survey asked 926 students about a digital companion that would accompany them across courses and semesters, helping with academic tasks and needs such as stress and motivation. Students were generally receptive. Mean willingness to adopt was 4.79 on a 7-point scale, with about 44% choosing 6 or 7. These were not AI novices. About 97% had already used AI tools and 82% had used them for their studies. A Buddy would extend a familiar practice (Martínez Moreno et al., 2026). Or course, AI tools and students’ experiences with them are changing rapidly, so a survey conducted in 2024 may not reflect current attitudes.

What students wanted from it was concrete. In addition to proofreading, references, and time management, 66% wanted help finding relevant courses and educational content. When asked to describe the benefits in their own words, the most common term was “time,” followed by “information” and “help.” The appeal was largely that of an academic assistant. Their priorities suggest that practical assistance deserves at least as much attention as a tool’s conversational personality (Martínez Moreno et al., 2026).

Students were also selective about who should have access to their information. About 80% were willing to share their academic schedules, 71% their learning preferences, 69% their interests and career goals, and 68% their study materials. Yet only 39% supported access by university officials for administrative purposes, and 23% supported access for assessment. The authors recommend excluding the Buddy from grading and limiting information provided to faculty and administrators to aggregated, anonymized data (Martínez Moreno et al., 2026).

This distinction matters for programs considering whether an AI assistant should also identify students who need help. A tool students use to work through confusion privately serves a different purpose from a system that reports that confusion to an advisor or instructor. Consent to use an assistant should not be treated as consent to every subsequent use of the information it collects.

Students with more at stake in disclosure may be particularly attentive to these distinctions. Pierrès et al. (2025) interviewed 20 university students with ADHD about disclosing their diagnosis to AI-based educational technologies. Participants generally viewed AI tutors more favorably than early warning systems, which they considered more likely to stigmatize them. Their willingness to disclose depended partly on who could access the information. They wanted selective control, including the ability to decide which lecturers would learn about their diagnosis. The possibility of more personalized support did not eliminate concerns about how that information might be used.

Similar concerns emerged when 132 Australian university students viewed video scenarios involving AI systems that detected attention or emotion during learning activities. Students responded more negatively when sensing was disclosed, whether it involved tracking gaze or interpreting facial expressions. They were particularly concerned about being identified as struggling and having help summoned without their consent. When asked to compare ethical priorities, they placed autonomy and privacy first (Han et al., 2026). The findings suggest that a technically capable system can still provide support in ways students find intrusive.

The Zurich researchers also examined whether willingness to adopt a Buddy was associated with students’ reasons for being on campus. Of 10 motivations, 8 had no statistically significant association with adoption interest. These included discussing in seminars, studying with classmates, socializing, and participating in student events. The significant associations were with motivation to attend lectures and consult or borrow library books, and both were negative and very small, at r = -0.084 and r = -0.105, respectively (Martínez Moreno et al., 2026).

The pattern suggests that students see AI as a way to obtain information more efficiently while continuing to value campus as a place to spend time with other people. Looking up an answer can also be an occasion to speak with a librarian, instructor, or classmate, and those incidental encounters deserve attention when evaluating what a new tool replaces.

Research with distance learners points to a similar combination of interest and caution. Across 3 studies conducted over 11 months at the Open University, students valued immediate academic feedback, 24/7 availability, and assistance tailored to their needs. Concerns about academic integrity, privacy, and ethics persisted across the studies. Rienties et al. (2025) identified groups ranging from highly critical to keenly supportive, with differences in prior AI experience, age, and educational background. The findings argue against treating students as a single market with a uniform appetite for AI. An institutionally developed assistant may be useful, but its relevance and acceptability need to be established with the students expected to use it.

Institutions also have a role in helping students make informed decisions. In Zurich, about 89% of students turned to the Internet for AI-related help, compared with 48% who consulted peers, 14% who consulted faculty, and 5% who used central university services (Martínez Moreno et al., 2026). These figures suggest that much of the guidance students receive comes from outside formal support systems. Programs should be asking whether students know how to check an answer, protect sensitive information, and recognize when a question requires human judgment.

The choice between AI and human assistance also depends on how difficult it feels to ask for help. In interviews with 8 computer science graduate students, Pal et al. (2026) found a distinction between trust and reliance. Students trusted human experts more but often relied on them less. Fear of appearing incompetent or being judged discouraged them from asking instructors and teaching assistants. AI was accessible and anonymous, even when students doubted its accuracy. As a participant explained, “I know AI hallucinates and makes mistakes, but I can check it privately without anyone judging me, so I use it way more than I probably should.”

The distinction is important for mentoring. Confidence in someone’s expertise does not necessarily make that person easy to approach. A student who uses a chatbot may be responding to the social costs of asking a person, rather than expressing a preference for artificial companionship. Programs should therefore consider what makes human help difficult to seek, including limited availability, uncertainty about what questions are appropriate, and the possibility of feeling embarrassed. Providing an AI alternative leaves those barriers in place unless programs also address them.

Interviews with 18 students in programming courses identified a related pattern. Students valued generative AI’s availability and convenience, but described human assistance as more emotionally supportive and, in their view, more conducive to deeper learning. Some also described investing less effort in AI interactions or ending them without fully understanding an answer. Their accounts suggest that the ease of obtaining an answer and the value of working through a problem are not interchangeable (Chao et al., 2026).

Meyer et al. (2026) examined the situational nature of these preferences with 282 U.S. college students who evaluated 6 feedback scenarios. High prior effort and a salient need for support were associated with greater preference for human feedback. High anxiety about evaluation was associated with greater preference for AI. In a separate study of 114 undergraduates completing an academic writing task, a strong overall preference for human tutors persisted after the feedback experience. However, students using a conversational AI interface were more receptive to AI than those using a structured writing-analytics interface (Le et al., 2025). Thus, both the circumstances of help-seeking and the design of the interaction matter.

Students’ judgments also depend on whom they believe they are hearing from. Lee and Esposito (2025) asked 127 college students to evaluate responses to student problems without identifying their sources. Perceptions of whether a response came from a human or AI shaped evaluations more strongly than its actual source. Students who regarded their human mentorship as effective were less inclined to seek further support from responses they believed were AI-generated. In a related qualitative analysis, responses perceived as human were described in terms of warmth, authenticity, and contextual detail. Those perceived as AI-generated were more often described as generic or impersonal, even when students had misidentified the source (Lee et al., 2025).

This concern also emerged in our study of 65 undergraduates at a minority-serving institution, in press in Translational Issues in Psychological Science (Werntz et al., in press). Students rated descriptions of peer mentoring without AI more favorably than mentoring in which AI suggested answers for a human mentor to personalize. Average usefulness ratings were 7.34 and 6.14, respectively, on a 9-point scale. Students were more comfortable with AI assistance for career questions than for mental health concerns, and about 65% wanted to know whether their mentor was using it. The findings extend the question raised in Zurich to situations in which a person remains involved. Students may welcome practical assistance while wanting to understand what the human helper contributes.

These findings help explain why a well-written response may not be sufficient in a mentoring relationship. Students are evaluating both the advice and what it suggests about the person providing it. A mentor who checks a suggested resource, considers whether it fits, and follows up is contributing something different from a mentor who forwards a polished answer. For programs using AI assistance, training should make those responsibilities explicit. Disclosure should also allow students to understand how the mentor has used the tool.

The distinction becomes especially important when students are seeking emotional support. In a survey of 428 U.S. college students conducted in 2024, 49% reported having used a chatbot, but only 5% had sought mental health support from one during the previous year. Attitudes toward chatbot-delivered support were neutral and substantially less favorable than attitudes toward traditional mental health services. Among students who had used neither, the most common reason for not seeking chatbot support was no perceived need, followed by doubts that it would help. Cost, time, and stigma were reported less often as barriers to chatbot support than to traditional services (Rackoff et al., 2025). Greater convenience did not necessarily translate into confidence that the support would be useful.

In an open-ended survey of 14 college-aged clients with recent counseling experience, Bates et al. (2026) found conditional acceptance of AI for nonclinical tasks, concerns about data security, and a strong preference for human judgment in crises. Participants valued the human relationship at the center of counseling and were more receptive to AI when it supported that work. These preferences echo the importance students placed on role boundaries in the mentoring studies.

When Kuhail et al. (2024) asked 224 college students to evaluate transcripts of active-listening interactions without identifying whether the helper was a human or AI, students identified the source correctly only 47.5% of the time. They also rated the AI transcripts more favorably. Read alongside the Lee studies, this suggests that AI can produce language students experience as supportive, while knowing that it comes from AI can change its meaning.

Students also report finding value in ongoing AI companionship. In a survey of 1,006 student users of Replika, Maples et al. (2024) found high levels of loneliness alongside substantial perceived social support. Students described using the app as a friend, a therapist, and a way to reflect on their own thinking.

Among 176 university and graduate students in South Korea, loneliness and social anxiety declined during a 4-week period of chatbot use (Kim et al., 2025). In a qualitative study of 17 neurodiverse graduate students, many found a brief interaction with an AI Virtual Mentor affirming and emotionally responsive, while others found it mechanical or impersonal. Participants described a possible supplement to human mentorship, with considerable variation in what they found helpful (Syharat & Zaghi, 2026).

There are also reasons to examine whether AI support displaces opportunities for human connection. In a survey of 387 university students, Crawford et al. (2024) found that greater perceived social support from AI was associated with lower perceived support from other people. AI use itself did not have a significant direct association with self-reported academic performance. Programs should examine how students’ use of AI relates to their opportunities for human connection.

A randomized study offers a more direct comparison. Li et al. (2026) assigned university students in their 1st semester to 2 weeks of daily texting with a supportive chatbot, texting with a randomly assigned peer, or brief daily journaling. At the end of the study, the peer group reported less loneliness than either of the other groups. The chatbot group did not differ significantly from the journaling group. The findings point to the value of measuring changes in loneliness alongside satisfaction with an interaction.

Taken together, the studies describe students who are interested in AI but selective about its role. Many value practical assistance, immediate feedback, and a private place to ask questions. Their willingness to use these tools varies with the task, the interface, the information being collected, and their access to people they trust. Some also find companionship helpful.

For universities and mentoring programs, the practical implications are fairly straightforward. Begin with the tasks students want help with, explain how AI is used, and give students meaningful control over their information. Keep private assistance separate from assessment unless students have clearly understood and agreed to the arrangement. Train mentors to check suggested information and use their knowledge of the student to decide what fits. At the same time, make human support easier to access and less intimidating to seek.

An assistant that helps a student organize work or a mentor find the right resource could free time for conversation, follow-up, and connection across campus. Programs should build those opportunities into how they use the technology. The Zurich students’ interest in a proofreader is a useful starting point. The more demanding task is to make sure that a more efficient university is also a place where students can find people who know them and are willing to help.

References

Bates, D., Antor, C., Kammeyer, P., Lorey, W. W., & Hakenewerth, T. J. (2026). “I don’t trust AI”: A generic qualitative analysis of college-aged mental health clients’ perceptions of artificial intelligence used in mental health counseling. Behavioral Sciences, 16(5), 754. https://doi.org/10.3390/bs16050754

Chao, A., Chen, M., Yang, Y.-K., Griswold, W. G., Porter, L., & Soosai Raj, A. G. (2026). Student perspectives on the role of teaching assistants in the age of GenAI. Proceedings of the 2026 ACM Conference on International Computing Education Research, Vol. 1, 303–316. https://doi.org/10.1145/3765964.3811648

Crawford, J., Allen, K.-A., Pani, B., & Cowling, M. (2024). When artificial intelligence substitutes humans in higher education: The cost of loneliness, student success, and retention. Studies in Higher Education, 49(5), 883–897. https://doi.org/10.1080/03075079.2024.2326956

Han, B., Ma, Y., Coghlan, S., McKay, D., Buchanan, G., & Smith, W. (2026). AI sensing and intervention in higher education: Student perceptions of learning impacts, affective responses, and ethical priorities. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, Article 699, 1–21. https://doi.org/10.1145/3772318.3790360

Kim, M., Lee, S., Kim, S., Heo, J.-I., Lee, S., Shin, Y.-B., Cho, C.-H., & Jung, D. (2025). Therapeutic potential of social chatbots in alleviating loneliness and social anxiety: Quasi-experimental mixed methods study. Journal of Medical Internet Research, 27, e65589. https://doi.org/10.2196/65589

Kuhail, M. A., Alturki, N., Thomas, J., & Alkhalifa, A. K. (2024). Human vs. AI counseling: College students’ perspectives. Computers in Human Behavior Reports, 16, 100534. https://doi.org/10.1016/j.chbr.2024.100534

Le, H., Shen, Y., Li, Z., Xia, M., Tang, L., Li, X., Jia, J., Wang, Q., Gašević, D., & Fan, Y. (2025). Breaking human dominance: Investigating learners’ preferences for learning feedback from generative AI and human tutors. British Journal of Educational Technology, 56(5), 1758–1783. https://doi.org/10.1111/bjet.13614

Lee, J., & Esposito, A. G. (2025). ChatGPT or human mentors? Student perceptions of technology acceptance and use and the future of mentorship in higher education. Education Sciences, 15(6), 746. https://doi.org/10.3390/educsci15060746

Lee, J., Wang, S., & Esposito, A. G. (2025). Content analysis of college student perceptions: Mental health-related support from generative AI versus faculty mentors. American Journal of Qualitative Research, 9(4), 70–100. https://doi.org/10.29333/ajqr/16815

Li, R.-N., Folk, D., Singh, A., Ungar, L., & Dunn, E. (2026). Is a random human peer better than a highly supportive chatbot in reducing loneliness over time? Journal of Experimental Social Psychology, 125, 104911. https://doi.org/10.1016/j.jesp.2026.104911

Maples, B., Cerit, M., Vishwanath, A., & Pea, R. (2024). Loneliness and suicide mitigation for students using GPT3-enabled chatbots. npj Mental Health Research, 3, 4. https://doi.org/10.1038/s44184-023-00047-6

Martínez Moreno, J., Christen, M., & Bernstein, A. (2026). Towards an AI Buddy for every university student? Exploring students’ experiences, attitudes and motivations towards AI and AI-based study companions [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2603.20909

Meyer, J., Keller, M. V., & Daumiller, M. (2026). When do learners prefer AI or human feedback? Situation-specific psychological predictors of feedback source preference. Proceedings of the Thirteenth ACM Conference on Learning @ Scale, 254–263. https://doi.org/10.1145/3774398.3811625

Pal, O., Duddu, V., Goyal, A., Goel, D., & Saha, K. (2026). Do we know what they know we know? Calibrating student trust in AI and human responses through mutual theory of mind. Proceedings of the Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems, Article 239, 1–8. https://doi.org/10.1145/3772363.3799386

Pierrès, O., Darvishy, A., & Christen, M. (2025). Perceived risks and benefits of disclosing ADHD to AI-based educational technologies: Semi-structured interviews. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, 8(3), 2030–2042. https://doi.org/10.1609/aies.v8i3.36693

Rackoff, G. N., Zhang, Z. Z., & Newman, M. G. (2025). Chatbot-delivered mental health support: Attitudes and utilization in a sample of U.S. college students. Digital Health, 11. https://doi.org/10.1177/20552076241313401

Rienties, B., Tessarolo, F., Coughlan, E., Coughlan, T., & Domingue, J. (2025). Students’ perceptions of AI digital assistants (AIDAs): Should institutions invest in their own AIDAs? Applied Sciences, 15(8), 4279. https://doi.org/10.3390/app15084279

Syharat, C. M., & Zaghi, A. (2026). “A supplement, not a substitute” – relational experiences of neurodiverse graduate students with an AI Virtual Mentor. Frontiers in Education, 11, 1769327. https://doi.org/10.3389/feduc.2026.1769327

Werntz, A., Hersch, E., Lamont, M., & Rhodes, J. (in press). Understanding undergraduates’ attitudes toward AI-assisted peer mentoring: Balancing authenticity with efficiency. Translational Issues in Psychological Science.