For 15 Moroccan postgraduates, AI support came with a confidence cost
A small interview study found that students valued grammar help, explanation and faster research tasks while some felt less able to submit work without AI validation. It records lived experience; it does not prove that AI reduced achievement or critical thinking.
The short version
- What happened
- Researchers interviewed 15 AI-using postgraduate students from two Moroccan public universities; participation required English and prior AI experience.
- Who it affects
- Postgraduate students need language and research support without losing ownership of their judgement; transparent task boundaries can also help supervisors assess genuine learning rather than guessing how AI was used.
- What remains unproven
- One primary source; 15 purposively selected, English-speaking AI users at two public universities; short self-report interviews; no comparator, longitudinal follow-up or measured achievement, confidence or critical-thinking outcome.
Keep in mind
One primary source; 15 purposively selected, English-speaking AI users at two public universities; short self-report interviews; no comparator, longitudinal follow-up or measured achievement, confidence or critical-thinking outcome.
Source analysis · 1 linked source. Primary analysed source date: 17 October 2026.
Read the full report · 4 min The evidence, methods, implications and limitations
Convenience sometimes turned into a need for validation
The important finding is a tension, not a verdict. Fifteen postgraduate students at two Moroccan public universities described AI as useful for explaining difficult concepts, finding starting points, planning lessons and polishing English grammar. Some said these tools reduced pressure and freed time for analysis. Yet the same convenience could make work feel less secure without a machine's approval: one participant said they were no longer fully confident submitting work before consulting AI, while others worried that quick solutions weakened independence.
The researchers grouped the interviews into three themes: AI as an academic co-pilot, a paradox between greater capability and lower self-reliance, and uncertainty about originality and dependency. Students did not speak with one voice. Some believed critical use improved learning; others felt their learning had declined. That variation matters because the paper's evidence is what participants perceived about their own habits—not an objective test of whether AI improved or damaged their thinking.[1]
The study was narrow by design
The researchers purposively selected students who already used AI, were enrolled in public postgraduate programmes and could be interviewed in English. Academic backgrounds included English studies, economics or finance and philosophy. Sessions took place through WhatsApp or Zoom, lasted 15 to 20 minutes and were manually analysed for recurring themes. The sample offers detail from an under-reported national context, but it excludes students who avoid AI, cannot interview in English, attend private institutions or study in other educational settings.
The title refers to achievement, confidence and critical thinking, but the study measured none of those with grades, before-and-after tasks, validated scales or a non-AI comparison group. It also did not observe actual prompts or submitted work. A statement such as being '60% 70% dependent' is one participant's self-description, not a prevalence estimate. The authors themselves describe the work as an initial exploration and say its small, single-country sample cannot support general psychological or behavioural conclusions.[1]
The distinction that matters
The same tool can shift roles
Participants described benefits and risks according to how much of the task they delegated.
Support
Explain, proofread or suggest
Student still evaluates the work
Validate
Check every draft through AI
Confidence may become external
Substitute
Delegate the core reasoning
Original learning becomes harder to see
Universities can respond without pretending the causal evidence is settled
For supervisors and lecturers, the interviews identify concrete points for discussion: which language edits are acceptable, when an AI-generated explanation must be checked, which parts of research require an independent first attempt and how students can disclose assistance without stigma. A simple assessment design could separate drafting, verification and reflection, asking students to preserve their reasoning and explain why an AI suggestion was accepted or rejected. Institutions could pair this with language support so students are not forced to choose between clear English and intellectual ownership.
Those interventions remain proposals, not tested results from this paper. The authors report no external funding or competing interests, and the interview data are not public because of privacy concerns. Stronger evidence would follow students over time, compare supported and unsupported tasks, measure both performance and confidence, and include Arabic- and French-speaking programmes as well as institutions beyond two public universities. For now, the responsible conclusion is modest: these students experienced AI as both scaffold and dependency risk, and university guidance should make that trade-off discussable and observable.[1]