ChatGPT and employability in higher education: exploring creativity, critical thinking, and problem-solving

Synthesis: Calle-Martinez and colleagues asked 30 students at Complutense University of Madrid to plan the same conference contingency with or without ChatGPT, then ran eight focus groups through grounded-theory coding. The two conditions produced different accounts of what makes a graduate employable: users described delegating routine production while keeping evaluation and refinement ("assisted autonomy"), and non-users described deliberation, ethical reasoning and teamwork as the source of originality ("slow creativity"). Team collaboration was the sharpest asymmetry — 8 of its 9 mentions came from non-users — and evaluation of AI suggestions and adjustment of output were mentioned only by users. The authors are explicit that these are perceptions from one institution, not measured skill gains.

Key Findings

1. Efficiency was the users' dominant gain. Time efficiency drew 15 mentions, 8 of them from the ChatGPT group, and information management 17, with participants describing a 20-minute turnaround on a task that would otherwise have run longer.
2. Users described a shift from production to evaluation. "Evaluation of AI suggestions" (11 mentions) and "adjustments to AI output" (13) appeared only in the ChatGPT groups, alongside metacognitive awareness (8 mentions against 1), which the authors read as self-regulation rather than offloading.
3. Non-users located originality in effort and deliberation. "Originality of ideas" drew 27 mentions, 16 from non-users, who described slower information gathering as the thing that generated better ideas, and alternative exploration was higher in that group (11 observations against 7).
4. Teamwork nearly disappeared when the tool appeared. Team collaboration drew 9 mentions overall, 8 of them from non-users; participants in both groups still ranked human interaction as an employability asset (27 mentions), so the value survived while the practice thinned.
5. The two modes were complementary, not ranked. The authors name "assisted autonomy" and "slow creativity" as parallel routes to perceived employability, and warn that treating speed as the outcome erodes depth; they propose a quantitative follow-up rather than claiming skill improvement.

A contingency plan, run twice

The design is small and interpretable. Thirty students from the same university completed a contingency-plan task for an international tourism conference whose venue fell through — an 800-word plan plus a relocation timeline — with one group working alongside ChatGPT and one without. Eight focus groups run on Microsoft Teams discussed four theoretical categories (complex problem-solving, critical thinking, Creativity and employability perception) through parallel question sets for users and non-users, and the transcripts were coded in ATLAS.ti through open, axial and selective coding to theoretical saturation. Quote frequencies are reported descriptively, and the authors are careful that network diagrams show how participants linked codes, not how often they appeared. The sample is 30 students at one Spanish university, self-selected into a qualitative protocol, so every finding here is a perception shaped in discussion rather than a measured competence.

What each mode made salient

The most useful result is not that one group did better but that each mode made a different capacity visible to students. Users talked about the mechanics of control: screening a draft for the one or two errors they expected, refining prompts until the output fit, and reworking AI text to match their intent. Their accounts of metacognitive awareness (8 mentions against 1) and of the "constant fine-tuning" the tool demanded suggest that assistance can install a monitoring loop — but participants also said the immediacy of answers discouraged questioning ("It gave ready-made responses, so we didn't question them much"), which is the same mechanism the cognitive offloading literature describes as a risk rather than a benefit. Non-users talked instead about justification: whether a limousine for VIP guests was coherent, how to divide the work, and why a slower route produced better ideas. Their creativity was described as invention under constraint, and the authors' term for it — slow creativity — is a claim about sequence, that deliberation precedes and generates the idea rather than verifying it afterward.

Assisted autonomy and slow creativity

The paper's conceptual contribution is the pairing. Assisted autonomy means delegating routine production while retaining judgment: the student keeps authorship of the criteria even when the tool writes the draft. Slow creativity means the friction is the method — ethical debate, coordination cost and self-assessment produce the originality that the assisted mode tends to relocate into selection and refinement. Read through the employability literature, the two pathways imply that a graduate's readiness depends on which of them they can enter deliberately, since workplaces increasingly expect the first and still reward the second. The authors connect this to the European Commission's skills agenda and to AI literacy training in prompt design, bias detection and ethics, and they frame the pedagogical task as designing activities that compare or revise AI outputs so that evaluation is rehearsed rather than assumed. The finding that teamwork mentions collapsed among users is the one most worth follow-up: if structuring a task with a tool removes the need to negotiate, the interpersonal capacity employers say they value stops being practiced in the very assignment meant to build it.

What this means for practice

• Instructors. Run the same task twice — once without the tool, once with a revision requirement — so students experience both pathways instead of settling into the one their habits favor; the paper's own design is a ready-made template.
• Program designers. Assess the re-authoring step, not the first draft: what students changed, why, and what they rejected is where the users' claimed self-regulation actually shows up.
• Administrators. Treat teamwork mentions as a warning signal: group assignments that route coordination through a chatbot no longer rehearse the negotiation employers name, so schedule at least one collaboration task where the tool is absent by design.

Limitations

• Thirty students at a single Spanish university, self-selected into focus groups, so the study supports no claim about skill or employability outcomes — the authors say as much.
• Perceptions were voiced in group discussion, where participants can converge on a shared account, and the quote frequencies are descriptive counts rather than measures.
• Assignment to condition is not described as randomized, and the two groups discussed different question sets, so the contrast is between accounts rather than between equivalent treatments.
• Grounded-theory coding with theoretical saturation claims is hard to audit from outside; no inter-coder reliability figure is reported.
• The design is cross-sectional and measures no later performance, employment status, or retention.

Citation

Calle-Martinez, C., Silvera-Roig, M., Recuero-Virto, N., & Paredes-Martinez, M. (2026). ChatGPT and employability in higher education: exploring creativity, critical thinking, and problem-solving. Creativity Studies, 19(2), 536–546.