Does AI help students design climate solutions?
In a peer-reviewed UAE study, 69 students reported greater creativity and agency when using AI, but the research measured perceptions in one course—not learning gains, stronger climate solutions or real-world environmental impact.
The short version
- What happened
- Researchers surveyed 69 of 82 eligible undergraduates in one United Arab Emirates University course, an 84.1% response rate, and conducted 25 qualitative interviews across policy, faculty and student groups.
- Who it affects
- Students may gain a structured way to explore complex climate problems, but confidence should not be mistaken for competence.
- What remains unproven
- The 69 survey respondents came from one UAE university course, with disproportionate representation from the College of Education.
Keep in mind
The 69 survey respondents came from one UAE university course, with disproportionate representation from the College of Education.
The design was cross-sectional and had no comparison group, pre/post assessment or random allocation.
Learning, creativity, critical thinking and agency were self-reported rather than measured through objective performance tasks.
Only seven respondents reported experience in a specific university climate project, so evidence about real solution design is thin.
The 25 interview participants were purposively selected and include only five student innovators; their views are not population estimates.
Source analysis · 1 linked source. Primary analysed source date: 5 October 2026.
Read the full report · 7 min The evidence, methods, implications and limitations
The answer is that students felt helped; the study did not show that they learned more
A peer-reviewed study from United Arab Emirates University found substantial enthusiasm for using AI in student-led climate work. Among 69 respondents, 60, or 86.9%, agreed that AI made project design more creative and independent. Fifty-three, or 76.8%, felt more empowered to propose climate solutions, while 51, or 73.9%, said AI helped them move from data collection toward higher-level solution design. These are meaningful signals about readiness and perceived usefulness in one educational setting.
They are not evidence that AI improved critical-thinking scores, climate knowledge or the quality of a solution. The survey was cross-sectional and asked students to rate their own experience. There was no pre-test, post-test, random allocation, comparison course or independent scoring of project work. The authors explicitly describe the outcomes as perceived cognitive scaffolding rather than measured skill acquisition. The defensible finding is therefore about attitudes and a proposed teaching framework, not proven educational effectiveness.[1]
The quantitative denominator was 69 students from one course
The eligible population was 82 undergraduates enrolled across three sections of a foundational AI and Education course at United Arab Emirates University. All had completed a prerequisite module on AI ethics and data privacy. Every eligible student was invited and 69 returned a completed questionnaire, an 84.1% response rate; 13 did not respond. The paper reports no statistically significant gender or academic-major difference between respondents and non-respondents, although the sample remained concentrated in a single course and institution.
All 69 respondents reported using ChatGPT, and 75.4% said they used AI daily or weekly. Fifty-one had used AI in a group or project setting. Only ten reported participation in a student-led project or hackathon, and only seven—10.14% of respondents—reported work on a specific sustainability project. That last denominator matters: most survey answers describe anticipated or general academic use rather than sustained experience developing, testing and implementing a climate intervention.[1]
Twenty-five interviews added regional and teaching perspectives
The qualitative strand used purposive sampling and included 25 non-overlapping participants. Eleven were policy or strategic experts: five Arab Youth Center leaders drawn from Bahrain, Saudi Arabia, Kuwait, Jordan and the UAE, plus six regional sustainability engineers. Nine were UAEU faculty members, including four AI or STEAM specialists and five educators working in educational technology or geography. Five were student innovators involved in climate projects.
These interviews were intended to test whether the proposed AI-Empowered STEAM Climate Action Model was feasible and connected to regional policy needs. Participants described AI as a way to reduce the burden of handling complex datasets and to support prototyping, while also warning that speed can be mistaken for accuracy. The five student innovators reported checking AI outputs against local sources, but five purposively selected accounts cannot establish how consistently ordinary students validate evidence or whether the curriculum caused that behaviour.[1]
The strongest result is perceived creativity, alongside a persistent trust gap
On a five-point scale, perceived creativity and independence had the highest mean score at 4.12, with a standard deviation of 0.79. Perceived empowerment to propose climate solutions averaged 3.84, and strategic solution design averaged 3.78. Forty-seven students, or 68.1%, agreed that AI supported critical thinking and problem solving. Because these items ask people to judge their own cognition, agreement may reflect confidence, convenience or enthusiasm as well as genuine skill.
Concern remained high despite 71% of respondents reporting formal AI training. Forty-two students, or 60.9%, raised accuracy and reliability; 41, or 59.4%, raised plagiarism or academic honesty; 36, or 52.2%, raised privacy and data protection; and 30, or 43.5%, raised over-dependence. The pattern supports teaching verification and responsible use, but it does not test whether the proposed framework actually reduces those risks. That requires observed behaviour and independently graded work.[1]
The review was substantial, but the intervention evidence remains preliminary
The study also incorporated a systematic literature review. The authors report identifying 1,000 records, retaining 815 after duplicate removal, assessing 245 full texts and including 80 studies after exclusions. They used Elicit for initial identification, metadata extraction and preliminary screening, then say two human reviewers independently checked every extracted data column against the full text. Gemini and Grammarly were disclosed for language and formatting support rather than analysis or conclusions.
That review helps place the teaching model within prior work, but it does not turn the 69-student survey into a trial. The included education literature was often single-institution and non-randomised, and the new study shares those constraints. The project was funded by United Arab Emirates University under grant G00005117; the authors declared no commercial or financial conflict. Institutional ethics approval and written consent are reported. Those disclosures support transparency while leaving generalisability and causal impact unresolved.[1]
What this could mean for students—and what would change the assessment
For students, a structured workflow could make difficult climate datasets less intimidating and create room for design, discussion and local problem framing. The useful teaching principle is not simply to permit a chatbot. It is to assign AI bounded roles, require validation against scientific and regional evidence, disclose its use and assess the human reasoning behind a proposal. Without those safeguards, fluent output could conceal weak climate science or make students more confident without making their work more accurate.
Confidence would rise with a longitudinal, multi-university study comparing the proposed curriculum with a credible alternative. Researchers should measure climate knowledge, critical reasoning, source verification and project quality before and after teaching, using blinded rubrics and recording which tools and prompts were used. Follow-up should examine whether student projects are implemented, whether claimed environmental effects are measured and whether results hold across disciplines, languages and access levels. Until then, this paper is best read as a well-documented regional feasibility signal rather than proof of learning gains or climate impact.[1]