“What Am I Really Doing with My Own Brain?”: College Students on Learning, Ownership, and AI

Abstract

Generative artificial intelligence gives college students new ways to support their learning, while also raising questions about what they still want to learn and do for themselves. This qualitative study draws on semi-structured interviews with 14 undergraduates at a Northeastern liberal arts college to examine how students distinguish between AI use that supports their own thinking and AI use that substitutes for it, what they believe is worth doing on their own, and whether and how purpose enters their reasoning about learning and AI use. Using the Framework Method with a cross-case matrix to compare patterns across participants and topics, we found that students repeatedly drew a line between using AI as a supportive tool and letting it take the place of their own thinking or authorship. Students most often talked about wanting to keep developing their own skills, thinking independently, and knowing what they could actually do without AI. They also wanted the work to still feel “like theirs,” citing ownership, pride, authenticity, and ethical concerns. Students also described pressures that made AI appealing, including heavy workloads, deadlines, grades, convenience, and concerns about falling behind their peers. Although few participants explicitly connected their AI choices to a stated purpose, many described concerns that may be relevant to purpose development, including developing valued skills and strengths, acting on personal values, and becoming the kind of person they hoped to be. We consider these concerns through a purpose-mindset framework while distinguishing our interpretation from participants’ own explanations of their AI choices. The findings suggest that understanding students’ AI use requires attention to what they believe they need to learn and do for themselves.

1. Introduction

Generative AI is now part of everyday undergraduate academic life. In the 2026 Student Generative Artificial Intelligence HEPI Survey, 95% of 1054 full-time UK undergraduates reported using AI in at least one way and 94% reported using generative AI to help with assessed work (Stephenson & Armstrong, 2026). Nearly half of respondents said AI had improved their student experience by saving time, improving understanding, and providing immediate support, while students who reported negative effects raised concerns about skill erosion, fairness, social isolation, and future employment. For educators, a central question is how students use AI while continuing to develop the skills and judgment that college is meant to build. Students make these decisions in the midst of ordinary academic pressures: earning grades, meeting deadlines, understanding professors’ expectations, and keeping up with their peers (Vallade et al., 2025). One HEPI respondent described using AI to “focus on critical analysis and deeper understanding,” while another said, “I’m not using my brain at all.” These responses capture two very different ways students can experience AI in their learning. Qualitative research can help explain how students understand AI’s role in their own learning and why they draw different boundaries around using it.

Recent research distinguishes AI use that supports students’ own thinking and AI use that substitutes for it. One question is whether students still see themselves as directing their own learning even when they use AI. For example, Essien et al. (2026) found that students who felt more in control of their AI use were more reflective about their learning, which was associated with stronger self-reported critical thinking. Zhu et al. (2026) make a similar distinction between “dependent cognitive offloading,” where AI takes over core thinking, and “autonomous cognitive offloading,” where AI serves as a scaffold while the student remains in control of thinking. Questions about who is doing the thinking also come up in research on psychological ownership. Students’ involvement in planning, reasoning, and revision has been linked to greater psychological ownership and confidence in their competence (X. Li et al., 2026). Other research suggests that students’ concerns about AI extend beyond their own learning and authorship. In interviews, undergraduates raised concerns about autonomy, creativity, trust, academic integrity, environmental sustainability, and the role of human teachers (Isaeva et al., 2026).

Ownership comes through especially clearly in Su et al.’s (2026) study of 239 university students majoring in education and preparing to become teachers in Eastern China. Students who were more dependent on GenAI were also more worried that AI-assisted work was not genuinely their own and that they were losing the ability to think and work independently. These were perceived losses, not demonstrated declines in ability. This closely parallels a question students in our study raised: if AI does more of the thinking, what remains mine, and what am I still learning to do?

Students’ approaches to AI may also depend on how they manage and value their learning. Prompts that encouraged students to reflect on and manage their own learning strengthened aspects of self-regulated learning in a generative AI environment (Xu et al., 2025). Motivation also appears to matter. Among economics students, expectations of success, perceived usefulness, interest, and perceived costs predicted GenAI use, while the personal importance students placed on doing well did not (Arthur et al., 2026). Chen et al. (2026) found that meaning in life partially mediated the relationship between self-control and academic engagement, with GenAI dependence also shaping this relationship. Together, these studies raise a further question: How might students’ decisions about AI relate to what they hope to gain from their education and why that learning matters to them? Purpose offers one way to consider this question. Related to meaning but more specifically future-oriented, purpose involves an intention to pursue something personally meaningful that also has consequences beyond the self (Damon et al., 2003). A purpose-mindset perspective draws attention to how people develop and use their strengths and skills, live according to their values, and consider the impact they hope to have on others (Liang & Klein, 2022). For college students, developing their skills and acting on their values can connect present learning to the people they hope to become and the contributions they hope to make.

Qualitative studies give us a closer look at how students themselves think about these questions. In interviews with undergraduate and postgraduate students, Hooley et al. (2026) found that students’ use of GenAI was connected to how they saw themselves as learners, particularly their confidence, independence, and sense of academic competence. Questions about ownership also emerged in Tan’s (2026) study of two advanced ESL students during their first year at U.S. universities. Over the year, both became more selective and critical in how they used AI while maintaining a strong sense of ownership over their work and learning.

Questions about future identity and purpose are explored more directly in Schans’s (2026) interviews with 13 undergraduates. Students discussed AI in relation to their professional identities, future work, abilities, and what they considered distinctly human or meaningful. Our study examines a more immediate decision: when AI can help with an assignment, what do students still want to do themselves, and why?

We analyzed interviews from Phase 1 of a larger, IRB-approved study of college students’ AI use. We asked three questions: How do students distinguish AI use that supports their own thinking from AI use that substitutes for it? What do they say is worth doing for themselves? And when, if at all, do purpose, future identity, moral values, and situational pressures enter their reasoning?

2. Materials and Methods

2.1. Participants and Procedure

The study was conducted at a Northeastern liberal arts college in collaboration with the institution’s Student Affairs office and a center supporting digital learning. During summer 2025, two trained undergraduate researchers conducted one-on-one, semi-structured interviews with 14 undergraduate students about their experiences with and attitudes toward generative AI. Interviews lasted approximately 35 to 45 min, were conducted via Zoom, and were audio-recorded, transcribed, and checked for accuracy. Participants were recruited through convenience and snowball sampling, provided informed consent electronically, and were assigned study identification numbers. Identifying information was stored separately from interview data. Participants represented a range of majors, with many studying psychology, neuroscience, or related fields. Eight participants reported he/him pronouns and six reported she/her pronouns. Among the 13 participants who reported their class year, one was a sophomore, five were juniors, and seven were seniors. Two participants were international students. The current analysis includes all 14 Phase 1 interviews.

The interview protocol asked about current and past AI use, professors’ and the institution’s approaches to AI, motivations for using or avoiding AI, and perceived effects on learning, writing, critical thinking, and relationships with faculty. Later questions addressed ethical decision-making, meaning and purpose, future goals, and students’ views of how AI might affect their futures.

At the time of data collection, the institution did not have a university-wide policy that either prohibited or permitted generative AI use in coursework. The institution’s academic integrity policy addressed unsanctioned or undocumented AI assistance, while expectations for appropriate AI use were otherwise set at the course level by individual instructors. Students therefore encountered different expectations for AI use across their courses.

2.2. Analytic Approach

Transcripts were examined using the Framework Method (Gale et al., 2013), which organizes qualitative data in a matrix so that patterns can be compared across participants and topics. Our analytic approach combined deductive and inductive coding. We began with categories drawn from the research questions and interview protocol, including students’ AI use, reasons for using or avoiding AI, stated purpose or future goals, and whether students connected purpose to their AI choices. As we read and compared the interviews, we refined the framework to capture reasons that emerged repeatedly in students’ accounts, including skill development, independent thinking, competence, ownership, pride, moral concerns, and task interest, as well as pressures such as time, workload, grades, convenience, and peer comparison.

For example, excerpts in which participants described wanting to complete parts of an assignment themselves to continue developing abilities they believed they needed were coded as skill development or competence. We then compared these excerpts across participants and examined whether participants also explicitly connected this reasoning to a stated purpose or desired contribution beyond the self. This comparison informed our distinction between skill development concerns and purpose-related reasoning; skill development alone was not coded as evidence of purpose.

The matrix recorded each participant’s stated purpose or what mattered to them, general pattern of AI use, reasons for doing parts of the work themselves, pressures toward greater AI reliance, and whether purpose, future identity, or responsibility to others entered their reasoning about AI use. The full transcripts were also reread alongside the matrix to check interpretations and quotes against the surrounding interview context. Although the matrix also included broad AI-use patterns developed for the larger project, this analysis focuses on students’ reasons for doing parts of the work themselves or turning them over to AI.

For each participant, we coded whether the interview showed one of four relationships between purpose and AI use: (1) an explicit connection, when the participant directly linked a stated purpose or desired contribution beyond the self to an AI-related choice or concern; (2) a contextually supported connection, when the surrounding interview supported such a relationship but the participant did not directly state it; (3) no established connection, when the interview did not provide sufficient evidence to connect the participant’s stated purpose with AI-related choices; or (4) a contradiction, when the participant explicitly rejected a connection between their stated purpose and their reasons for using or avoiding AI, rather than simply saying that AI did not affect their sense of purpose. These classifications captured the evidentiary relationship between participants’ stated purpose and their reasoning about AI use. Not all four classifications were represented in the final coding of this sample. Connections between AI use and future professional identity were recorded separately and were not treated as evidence of a purpose-AI connection unless participants themselves made that link.

Separately, we considered how students’ reasons for preserving parts of their own learning might relate to purpose development, including developing valued strengths and skills, acting on personal values, and becoming capable of contributing in ways that mattered to them (Liang & Klein, 2022). We used this purpose-mindset framework as an interpretive lens, not as evidence that participants themselves were acting from purpose.

Two researchers independently coded all 14 interviews using the matrix, and a third researcher independently coded a randomly selected four-interview subset (P2, P4, P6, and P8) to provide an additional check on the application of the coding framework. The two full-sample coders agreed on the purpose-to-AI classification for all 14 participants, and the third researcher agreed on all four cases in the subset. Other disagreements were checked against the transcripts and resolved through discussion. Because the analysis was qualitative and interpretive, we used the coding to compare cases systematically rather than treat the classifications as quantitative measures of interrater reliability.

The original Phase 1 sample was not recruited iteratively using saturation as a stopping criterion, and we therefore do not claim that thematic saturation was established. Although later interviews largely repeated or elaborated categories already identified rather than introducing new major themes, we do not make a claim of meaning saturation, particularly for purpose, which was one of several topics covered in these interviews.

2.3. Reflexivity

One of the undergraduate interviewers later contributed to the analysis and manuscript development. His interest in the topic grew partly out of his own experience as a student. When he saw a course as connected to how he hoped to contribute beyond the classroom, he found himself more invested in doing the readings, participating in discussions, and grappling with difficult material because the learning itself mattered to him. This made him curious about whether other students approached AI differently when their coursework felt connected to a larger purpose. His familiarity with the interviews helped him interpret participants’ responses, but his own experience may also have made him more likely to notice connections between purpose and AI use that participants themselves did not make. To address this possibility, we looked specifically for counterexamples and for students who did not connect purpose with AI use. We checked interpretations against the full transcripts rather than relying on isolated quotations, used independent coding, and did not infer a connection simply because a student discussed both purpose and an AI boundary in the same interview.

Peer interviewing may have helped students speak candidly about using AI under time pressure or in ways they knew instructors might question. It may also have influenced how they presented their choices. Direct questions about purpose could have prompted connections that were not otherwise salient to students, and undergraduate interviewers may have varied in how deeply they probed abstract concepts such as purpose.

The project grew out of an institutional student AI advisory initiative, in which undergraduate and graduate students work alongside faculty and staff to study questions about AI and learning. This approach was intended to keep students’ own perspectives at the center of questions about AI and learning, rather than beginning primarily with questions of academic misconduct or compliance with rules (Ying, 2026).

The analysis is qualitative and exploratory. It describes students’ reasoning and self-reported behavior; it does not establish causal relationships between students’ reasons, circumstances, and patterns of AI use.

3. Results

3.1. Supporting Thinking Versus Substituting for It

Students repeatedly distinguished between using AI to support parts of the learning process and turning over thinking they believed they should still be doing themselves. The boundary depended less on whether AI was used at all than on what students believed they were handing over. Participant 8 described the distinction directly:

“If it’s like I’ve already done work and then this is helping me organize that work, then I think that’s what I would view as okay. But if I’m having it produce work itself, that’s where I draw the line.”

Her advice to other students followed the same logic:

“Use AI to enhance, not to create…do your own thinking.”

Participant 1 described a similar process. When AI gave her an answer, she said that was “where my own thinking starts”: she would cross-check the response and build her own answer from it rather than submit the output as her work. For some students, AI assistance could even help them express thinking they already considered their own. Participant 7, an international student, described using AI when he knew what he wanted to say but had difficulty expressing it in English. Rather than experiencing this as giving up his voice, he said, “I personally feel like it enhances my voice.” He explained that he could keep revising the AI’s response until it better expressed what he was thinking.

3.2. Developing Skills, Competence, and Ownership

Across the interviews, students worried about whether relying on AI would interfere with developing the skills they believed college was supposed to build. They raised concerns about critical thinking, writing, problem-solving, independent judgment, and knowing what they could actually do without AI.

Participant 1 described the tradeoff plainly. AI saved time and made work easier, but she worried that relying on it made it harder to begin assignments on her own: “it’s making it quicker, but then I’m losing the ability to do it on my own.” Participant 2 explained why she wanted grades to reflect her own work:

“You’re in school for a reason, and the grade you get is there for a reason.”

Participant 5 raised the same issue at the level of education itself:

“It’s a waste of an education if I’m using AI to write every single paper and homework assignment that I have.”

She later asked:

“What am I really doing with my own brain?”

Participant 8 described wanting to remain a “free thinker,” and Participant 10 said he wanted to know which thoughts were actually his and to “take a little bit of pride” in work that was his own. He wanted to know what he actually understood, what he could do without AI, and whether the work reflected his own judgment.

Participant 3 felt wrong using AI when its output became part of work submitted for a grade because the work was no longer fully her own. Participant 12’s concerns were more explicitly moral: environmental harm, plagiarism, originality, and creative ownership informed her decision to avoid AI. Ownership, pride, authenticity, and moral values also shaped students’ boundaries alongside concerns about learning, though students did not treat these reasons as the same thing.

3.3. Purpose and Future Identity: Direct Connections and Their Limits

For some students, future plans gave them a concrete reason to do the learning themselves. We distinguished these career-related reasons from cases in which students directly connected an AI decision to their stated purpose. Participant 3 explained why she avoided AI for certain assignments:

“I’m taking classes that are very important for me and for my career so I want to spend the time to really learn and do the assignment by myself.”

Participant 5 also connected concerns about AI with her future in healthcare. She related concerns both about developing the skills she would need and about how AI might affect the healthcare work she hoped to pursue. Participant 13 similarly explained why his academic work still felt meaningful:

“Doing the work contributes to my education and thus contributes to the psychologist that I will be out in the world one day and make me a better one.”

This did not mean that he wanted to avoid AI. Looking toward his future work, he thought AI could be useful for finding research and working through ideas, but said, “I’m trying to do my best to know these things on my own and have that knowledge before I go and turn to it.” For him, AI could be useful while there were still things he wanted to know and be able to do for himself.

Participant 12 provided a clear direct connection between purpose and an AI-use decision. She described her purpose as wanting “to leave the world better than I found it.” Because she viewed AI’s environmental and human costs as inconsistent with that purpose, she said:

“I can’t use it if I want to maintain and preserve my sense of purpose.”

Participant 1 connected AI and purpose in a different way. She worried that relying on AI was weakening critical-thinking and problem-solving “skills that I value and skills that I hope to use in whatever my purpose is.” She also said that the ease of asking AI questions had made thinking for herself about her purpose more difficult.

Students did not always understand or talk about purpose in the same way. Participant 2 described a relational sense of purpose centered on helping others, family, friends, relationships, and service, but when asked about the connection to AI, said:

“AI does not affect that because my personal sense of purpose doesn’t come from my job or my career or anything like that. It all comes from my relationships.”

P2’s answer makes clear that she did not see AI as affecting where her purpose came from, but we cannot tell whether her purpose played a role in the boundaries she set around AI.

She nevertheless maintained strong boundaries because she wanted her grades and competence to be genuinely her own. Participant 8 articulated a clear vocation toward social work and a strong boundary against having AI produce her work, but did not explicitly connect that boundary to her stated purpose. Participant 6 wanted to become a therapist and valued improving his own writing, yet did not explicitly connect the two. Participant 10 described his purpose as “bringing joy to others” but said purpose and AI were “kind of separate”; his restraint centered more clearly on ownership, pride, and concern about losing fluency.

Across the interviews, students with a stated sense of purpose varied considerably in how they used AI. Having a purpose also did not necessarily mean drawing on it to explain their AI choices. Table 1 summarizes these patterns across all 14 participants.

Table 1. Cross-case patterns in AI use, reasons for retaining own thinking, pressures toward greater AI use, and relationship to purpose or future identity.

Person & Major	AI-Use Pattern	Reasons for Retaining Own Thinking	Pressures Toward Greater AI Use	Stated Purpose/Longer-Term Orientation	Relationship to Purpose or Future Identity
P1—Applied Psychology & Human Development	Weekly; varies by course	Skill development, independent thinking, authorship	Time saving, ease, and speed	Developing abilities that can contribute beyond herself; still clarifying the specific form of that purpose	Direct purpose connection: concerned AI could weaken skills she hoped to use in her purpose
P2—Theology	Frequent; all extensive readings in some capacity and about half of writing assignments	Competence, ownership, pride, academic honesty	Time-management difficulties, last-minute deadlines, and projects piling up	Helping others through relationships, family, friends, and service	No direct connection: did not see AI as affecting her sense of purpose
P3—Applied Psychology & Transformative Education	Selective; about 20% of assignments	Learning, ownership, career-relevant skills	Fear of falling behind peers who use AI, including concern that they could produce stronger work and grades	Helping children, including through speech pathology	Future-professional connection: career-relevant learning; no direct purpose-AI link
P4—Economics, Applied Psychology, Philosophy	Frequent for verbal/writing work, especially translation	Preserving his own meaning and ideas in translation	English-language demands and the practical advantage of translation	Providing valuable information to others and contributing to society	No direct connection: did not link purpose to AI decisions
P5—Applied Psychology	Frequent; estimated about 75% of assignments	Skill development, critical thinking, writing, and making education worthwhile	Peer disadvantage when others use AI; heavy workload and limited time, especially during finals	Helping people through healthcare	Future-professional connection: concerned about AI’s implications for healthcare work
P6—Psychology	Daily in some form	Writing, challenge, critical thinking, satisfaction in own work	Limited time for long readings; ease of using AI for assistance	Psychology/possible therapy; being a good Catholic and becoming the best person he could be	No direct connection: said AI had not affected his purpose
P7—Applied Psychology & Transformative Education	Selective; mainly summaries and efficiency support	Own thoughts, personal experience, and task relevance	Heavy reading and limited time; efficiency	Representing and caring for marginalized individuals or groups, particularly through research	No direct connection: said AI had not affected his purpose
P8—Applied Psychology & Human Development; Biology	Selective; study guides, summaries, explanations, and rewording	Free thinking, ownership, authenticity, and making use of her education	Stress, many assignments piling up, and insufficient time to complete all readings	Loving other people well; social work as one expression of that vocation	No direct connection: did not link vocation to AI boundaries
P9—Finance	Frequent/everyday	Ownership of work and ideas; attribution	Easy, immediate access to information; no specific deadline or workload pressure identified	Described purpose primarily in religious terms	No direct connection: said AI had not affected his purpose
P10—Theology; Environmental Studies	Regular; about 2–3 times per week and 10–15% of assignments	Ownership, cognitive skills, independent expression, and pride	Time crunches and deadlines; speed and ease	Bringing joy to others	No direct connection: described AI and purpose as “kind of separate”
P11—Neuroscience	Rare; mainly when instructors direct or permit its use	Meaningful learning, accuracy, and ethical/social concerns	Ease and quick access to simple information; no specific situational pressure identified	Helping people	No direct connection: said AI had not affected his purpose
P12—Applied Psychology & Human Development	Rare/avoids	Critical thinking, originality, and ethical and environmental concerns	No personal pressure toward use identified; recognized that AI can make tasks easier and more convenient	Leaving the world better than she found it	Direct purpose connection: viewed AI avoidance as consistent with her purpose
P13—Psychology	Low/selective; mainly a study aid plus instructor-required use	Learning, competence, professional preparation, and doing work himself	Time crunches, competing academic/extracurricular demands, deadlines, and the temptation of similar grades for much less time	Helping people through psychology; enabling others to leave interactions better than they entered them	Future-professional connection: linked independent learning to becoming a psychologist
P14—Neuroscience	Daily	Avoiding overdependence, checking accuracy, and retaining human judgment	Immediate clarification and quick help with practice problems; no specific deadline or workload pressure identified	Helping others, particularly children; interested in healthcare/psychiatry and human-centered work	Future-oriented discussion: considered AI’s implications for future work, without linking them to his AI decisions

Note. Academic disciplines reflect participants’ self-reported majors at the time of interview. Pressures are reported only when supported by the participant’s own account. Where no situational pressure was identified, the table distinguishes general convenience or ease from a personally described pressure. Future-professional connections are reported separately from direct purpose-AI connections.

3.4. Task Interest and Situational Pressures

Interest in the task may change how the same student uses AI. Participant 7 offered a useful contrast. He wanted academic work to contain his “own thoughts,” but when discussing motivation he said:

“If there is a field I’m not interested in, I would just let AI do it.”

He contrasted that with areas he cared about, where he sometimes used AI to find additional resources and go deeper. His account suggests that students may use AI differently depending on the task rather than having one fixed “AI-use style.”

Students’ reasons for doing the work themselves sometimes competed with pressures to work efficiently. Participant 2 captured the tension under deadline pressure:

“There’s no time to hold on to my pride right now. I need to just use AI to get it done.”

Participant 13 called AI under time pressure “that devil on your shoulder” and “the easy way out.” Participant 3 described an early motivation to use AI as fear of “falling behind” classmates who were using it. Participant 5 described AI summaries as attractive when she was “swamped during finals” and did not have time to complete a long reading. Across interviews, students’ desire to preserve their own learning and authorship often came into tension with the convenience of AI and the pressures of academic life.

4. Discussion

The students we interviewed were thinking about more than whether AI was allowed. They wanted to know what they could hand over without losing the chance to learn something important. Some worried about writing, problem-solving, or thinking independently. Others cared about whether their work still felt like their own. Purpose entered a few of these decisions but did not explain everyone’s boundaries. Their accounts suggest that what matters is not simply whether students use AI, but which parts of the thinking they want to keep doing themselves, and why.

Purpose was one part of the picture. We define it as a personally meaningful intention to contribute beyond the self (Damon et al., 2003), rather than treating any academic or career goal as purpose. A purpose-mindset perspective also draws attention to the skills people develop, the values they act on, and the contributions they hope to make (Liang & Klein, 2022). P12 explicitly connected her decision to avoid AI to her purpose, while P1 worried that relying on AI could weaken skills she hoped to use in pursuing a future purpose. For P3, P5, and P13, future professional goals shaped their thinking about AI. Other students, including P2, P6, P8, and P10, described purposes without connecting them to their AI choices. Having a sense of purpose, then, did not necessarily mean drawing on it when deciding how to use AI. An important question for future research is what makes that connection more or less likely.

Participant 2’s interview shows how these reasons can compete within the same student. She wanted her grade to reflect real effort: “you’re in school for a reason and the grade you get is there for a reason.” But, when facing a deadline, she said: “There’s no time to hold on to my pride right now. I need to just use AI to get it done.” Her account shows how a student can value ownership and still make a different choice under pressure. Because these interviews rely on retrospective self-reports, we cannot know how consistently students acted on their stated values across situations. Purpose research may need to pay more attention to these moments of tension, when students’ desire to develop their own abilities competes with immediate academic pressures.

The interviews also suggest that students’ reasons for keeping work their own should not all be grouped together. Worrying about losing critical-thinking or writing skills is different from wanting to feel pride in one’s work. Environmental concerns and plagiarism raise moral questions. Purpose and future identity can both connect learning to longer-term aims, but they are not the same thing. Future research should examine these reasons separately.

Cognitive offloading itself does not necessarily mean losing agency or learning less. It can reduce the demands of one part of a task while allowing students to focus on other kinds of thinking. What mattered to students in our interviews was whether they were still directing and judging the work, or handing over thinking they believed they needed to practice themselves. This distinction resembles recent work on dependent and autonomous offloading (Zhu et al., 2026), dependent outsourcing and reflective collaboration (X. Li et al., 2026), and epistemic dependence (Du & Yuan, 2026). Students described using AI to organize, explain, translate, summarize, generate practice material, and provide feedback without necessarily seeing these uses as giving up their own thinking. S. Wang and Zhang (2026) similarly found that cognitive offloading could coexist with cognitive vigilance and was associated with self-reported transformative learning. What students delegate may therefore matter as much as how much they delegate. A recent MIT report also calls for “augmentation not automation,” emphasizing AI use that preserves students’ agency, judgment, and engagement in learning (MIT Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training, 2026).

Whether AI supports or erodes learner agency depends on how students regulate its use and how it is incorporated into learning (G. Wang et al., 2026, reviewing 123 studies). Concerns about cognitive ownership also appear in Su et al. (2026), particularly among students reporting greater GenAI dependency. Along the same lines, Hamsiah et al. (2026) found concerns about growing dependence, reduced cognitive struggle, and shifting ideas about authorship among undergraduates using ChatGPT for academic writing. Our interviews help show why these concerns are meaningful to students: staying involved in the thinking allowed them to keep developing skills and know what they could do for themselves.

Belonging and relationships did not appear to be major reasons students gave for deciding whether AI should help with or take over their work. Still, several interviews pointed to effects on relationships worth studying separately: one student described less need for office hours because AI could provide immediate help, while others described anxiety about AI detection, faculty suspicion, or tension around suspected AI use. These mixed accounts are similar to Isaeva et al.’s (2026) finding that students continue to place human teachers within their learning networks for motivation and relational support.

These interviews also raise the question of whether some of the tension between students and faculty around AI is focused too narrowly on academic misconduct. Many students themselves were concerned about the potential loss of learning. They asked where AI was helping them and where it was doing something they needed to learn to do themselves. They worried about whether their work was still theirs, whether they could think and write on their own, and whether using AI too much would leave them less capable. These concerns were not so different from those of their instructors. Students’ ability to act on those concerns may also depend on the conditions in which they are making these choices. They described working for grades, facing deadlines and heavy workloads, and watching what their peers are able to produce with AI. If grades and evaluation in a class heavily emphasize the end product over how students think, grow, or learn along the way, it is understandable that students feel pressure to use AI even when they would rather do more of the learning themselves. This suggests that responding to AI cannot only be about changing student behavior. We may also need to look at the conditions we have created around their learning, including what we reward and how we evaluate them.

5. Implications for Educators

Because this study did not test an intervention, we cannot say that making purpose, values, or task meaning explicit will change how students use AI. But the interviews suggest that educators might begin with a different question: What would make students want, and be able, to do the learning themselves? Students in these interviews were already concerned about whether they were learning, whether the work was really theirs, and whether they were becoming less able to think and work on their own. Rather than focusing only on what kinds of AI use are prohibited, educators might make clearer what students are meant to learn, which parts of the work they need to do themselves, and why that effort matters. This is consistent with Stephenson and Armstrong’s (2026) recommendation that institutions teach students to use AI to enhance learning rather than outsource it.

This does not mean students should avoid AI altogether. A recent review of 67 studies found that ChatGPT was more likely to support critical and creative thinking when its use was structured around inquiry, evaluation, and metacognitive engagement, while less structured use was more often associated with cognitive offloading (C. Li et al., 2026). Assignments might therefore be designed around which parts of the thinking students need to do themselves rather than simply whether AI is allowed. In some assignments, students might first work through a problem or develop an argument themselves and then use AI for feedback, comparison, or revision. In others, AI might be part of the task from the beginning, with students learning how to question and evaluate its output, make thoughtful choices, and explain their final judgment.

The findings also raise questions about how students are evaluated. If the goal is learning, it could be helpful to place greater value on evidence of the learning process rather than only on the product students turn in. One possibility would be to include evidence of how thinking developed, such as early ideas, decisions made along the way, revisions, or an explanation of what changed and why. This is not primarily about creating more ways to catch students using AI. It is about making clear that what counts is the thinking that the assignment is supposed to develop. If students know that their judgment, growth, and ability to explain their thinking will be valued, doing that thinking may remain worthwhile even when AI could produce a better-looking final product faster.

Purpose may also be worth examining in relation to the pressures students face. For students who connect their education to a future contribution they care deeply about, doing the learning themselves may feel especially important. Yet succeeding academically may also feel more consequential when it is tied to that future. Our interviews were not designed to test this tension, but they raise a question for future research: What happens when students’ desire to become genuinely capable comes into conflict with the pressure to perform well enough to reach the future they care about?

Some pressures toward greater AI use cannot be addressed simply by asking students to make better choices. Students described deadlines, heavy workloads, grades, and the pressure to keep up with peers. If we want them to hold onto the parts of their learning they believe matter, we need to consider whether their courses make that realistic. This means examining workload, how assignments are evaluated, and whether expectations for AI use are clear and consistent.

6. Limitations

This study has several limitations. The interviews provide first-person accounts from 14 undergraduates, but all participants attended the same Northeastern liberal arts college and were recruited through convenience and snowball sampling. Their experiences may therefore reflect features of this particular institutional setting and should not be assumed to represent college students more broadly. The sample also included many students in psychology, applied psychology, neuroscience, and related fields, and most participants who reported class year were juniors or seniors. The prominence of these fields and class years may have shaped the kinds of concerns that emerged and may limit how readily these findings can be transferred to students in other disciplines or stages of college. Recruitment methods may also have favored students who were especially willing to discuss AI or whose experiences were relatively similar. Future research should examine whether similar patterns appear across different institutions, student populations, academic disciplines, and educational contexts. The interviews cannot tell us whether students’ concerns about losing skills correspond to actual changes in their writing, critical thinking, or other abilities. Future research could examine how students’ reasoning about AI relates to their observed AI use and learning over time.

The interview protocol was designed for a broader study of students’ AI experiences rather than specifically around the questions examined here. Interviews lasted approximately 35 to 45 min and covered several dimensions of students’ experiences with AI, limiting the depth with which any one topic could be explored. Although later interviews largely repeated or elaborated categories already identified, we do not make the stronger claim of meaning saturation, particularly for purpose. Participants also appeared to vary in how they understood the word “purpose,” with some discussing it primarily in terms of future careers. Interviews designed around a broader conception of purpose could examine more directly how students connect their everyday choices about learning to their strengths, values, identities, and desired contribution.

7. Conclusions

Generative AI is changing how students approach academic work, including which parts of the learning process they do themselves and which parts they ask AI to support. The students in this study were often aware of that distinction, but their reasons for wanting to do the learning themselves varied. They wanted to develop their own abilities, retain ownership of their work, and act according to their values. For some, these concerns were explicitly connected to a future purpose or professional identity. For others, they were important in their own right, without a stated connection to purpose.

Students’ reasons for retaining their own thinking did not always determine their choices. Deadlines, workload, grades, convenience, and peer comparison could make greater AI reliance appealing even when students valued learning and ownership. Understanding these decisions requires attention both to what students believe is worth learning for themselves and to the conditions under which they are trying to learn.