How Do Undergraduate Students Use ChatGPT for Career Development? An Explanatory Mixed-Methods Study of Employability-Related Activities

Synthesis: Loidl, Cao, and Huang report an explanatory sequential mixed-methods study of how undergraduates in one College of Liberal Arts and Sciences at a Midwestern R1 university used ChatGPT for career development, surveying 102 students between August 2024 and January 2025 and following up with 25 in focus groups. The design pairs the Technology Acceptance Model with a three-item measure of perceived ethical considerations, so Trust and responsible use sit beside usefulness and ease of use. About 81% had used it for employability-related activities in the previous 12 months, most often for resumes (52.94%), workplace communication (38.24%), and cover letters (35.29%). Only attitude toward the tool (β* = .48) and behavioral intention to use it (β* = .36) predicted perceived usefulness in the final regression. Interviews show students treating it as a prompting-heavy revision and rehearsal partner, wary of accuracy and overreliance.

Key Findings

1. Most students had used ChatGPT for career tasks. About 81 percent (83 of 102) reported using it at least once in the previous 12 months, and 21.57 percent (22) more than ten times.
2. Resumes led, followed by workplace communication and cover letters. Resume building was most common (52.94%), then workplace communication (38.24%), cover letters (35.29%), and suggested interview responses (28.43%).
3. Frequent users rated the tool more useful. A one-way ANOVA found a significant difference across usage groups (F[4, 95] = 2.59, p = .042), with users of ten or more times outscoring never-users (t(95) = -2.96, p = .031).
4. Attitude and behavioral intention predicted perceived usefulness; ease of use, credibility, and ethics did not. The final two-predictor model explained R² = .61, retaining attitude (β* = .48, p < .001) and intention (β* = .36, p < .001).
5. Credibility beliefs were lukewarm. Perceived credibility averaged 3.04 out of 5.00 (SD = 0.98), the lowest of the seven construct means, against 3.95 for ethical considerations.
6. Students tailored and rehearsed rather than outsourced. Interviewees pasted job descriptions beside their drafts to surface sought-after skills, drafted emails, and brainstormed interview questions, revising rather than copying.

A Two-Phase Design in One College

Phase 1 collected survey responses from August 2024 to January 2025 through flyers, course emails, and campus listservs; 119 students opened the survey and 102 complete responses from undergraduates were retained (87.93%). In February and March 2025, 25 students joined Zoom focus groups of up to six participants, and cancellations turned three sessions into individual interviews of 20 to 70 minutes. Two doctoral researchers coded the transcripts with thematic analysis, resolving differences by consensus.

What Predicted Perceived Usefulness

Mean perceived usefulness of ChatGPT for career tasks was 3.63 out of 5.00 (SD = 1.00), ease of use 5.33 out of 7.00 (SD = 1.34), and behavioral intention 4.80 out of 7.00 (SD = 1.70). A regression with backward elimination began with six predictors and explained R² = .63; the final two-predictor model (F[2, 97] = 74.77, p < .001) explained R² = .61, retaining only attitude toward ChatGPT (β* = .48) and behavioral intention to use it (β* = .36). Ease of use, credibility, perceived social influence, and perceived ethical considerations dropped out. Students who had used the tool ten or more times rated it more useful than never-users (t(95) = -2.96, p = .031). All of it rests on self-reported responses.

Tailoring, Rehearsal, and Tone

All 25 interview participants had used ChatGPT for employability work. For resumes and cover letters, students described two strategies: having the tool catch errors and suggest revisions, or pasting a job description beside their draft so it would highlight the skills an employer asked for — "specific skill sets or specific tools that the company actually asks for," one said. A smaller group supplied rough notes and asked for polished paragraphs. For workplace communication, students drafted and revised emails, used the tool as a grammar checker or tone adjuster, or preferred their own voice where authenticity mattered. Interview preparation meant brainstorming from job descriptions and partial answers. The authors read this pattern of student-AI interaction as translation and calibration work, not outsourcing.

Trust, Authenticity, and the Ethics Extension

The study added three items on perceived ethical considerations — discussing AI use with peers or mentors, verifying information before applying it, and awareness of potential biases. The construct correlated positively with perceived usefulness (r = .36) but was the weakest correlate and dropped out of the final regression, so ethical awareness alone did not predict seeing the tool as useful. Students questioned generic responses, inaccurate suggestions, and whether AI-drafted materials were credible, and described revising output, writing more detailed prompts, or comparing tools rather than copying. The authors read that reflective engagement as a developing form of critical AI literacy that keeps Learner Agency with the student, and warn that overreliance may narrow the self-reflection career preparation should build.

What this means for practice

• Instructors. Build critical-evaluation tasks into career-readiness work: have students critique AI-generated resumes, cover letters, and interview answers, then tailor them to a real job description, since reported value came from tailoring, not generating.
• Instructors. Use the tool's low-stakes rehearsal affordance deliberately — timed mock interviews, drafted emails students rewrite in their own voice — since participants valued practicing professional communication before higher-stakes use.
• Administrators. Fund the career-services interventions the paper recommends: prompt-design workshops, resume revision labs, and advising protocols that teach students to evaluate and personalize AI output for the tasks they already bring (resumes, 52.94%).
• Administrators. Clarify institutional expectations for AI-assisted application materials — transparency, authenticity, acceptable use — and fold AI literacy into career-readiness programming, guiding use rather than discouraging it.

Limitations

• The study ran at a single Midwestern R1 university in one College of Liberal Arts and Sciences: 102 valid surveys from 119 responses (87.93%) and 25 interviewees, with no exact response rate determinable from flyer and listserv recruitment.
• All measures are self-reported and retrospective over a 12-month window, which the authors say may introduce recall bias and overrepresent tech-engaged students; the data are not publicly available.
• The backward elimination regression is exploratory rather than confirmatory, and the final model is reported inconsistently: the text names attitude (β* = .48) and behavioral intention (β* = .36), while Table 4's Model 5 labels the surviving rows perceived ease of use (.48) and attitude (.36).
• The paper never reports which ChatGPT version students used, and its data were collected between August 2024 and March 2025; the tool has changed since, and the cross-sectional, ChatGPT-only design cannot show how perceptions evolve.

Citation

Loidl, V., Cao, Y., & Huang, W. D. (2026). How Do Undergraduate Students Use ChatGPT for Career Development? An Explanatory Mixed-Methods Study of Employability-Related Activities. International Journal of AI in Pedagogy, Innovation, and Learning Futures.