AI-Supported Inquiry-Based Learning in Photosynthesis and Respiration: Implications for Sustainable Science Teacher Education

Synthesis. In an eight-week AI-supported guided inquiry intervention that integrated problem-based and design-based learning, Aydın (2026) found that 48 Turkish pre-service science teachers made significantly larger gains than a conventional-instruction control in conceptual understanding of photosynthesis (F(1,41)=22.79, p<0.001, ηp²=0.357) and cellular respiration (F(1,44)=25.70, p<0.001, ηp²=0.369), but showed no significant change in AI literacy or self-perceived computational thinking skills. Qualitative reflections reported benefits for conceptual learning, inquiry, collaboration, critical evaluation of AI-generated outputs, and digital content creation, suggesting AI-supported inquiry can strengthen disciplinary learning even when broader competencies require longer or more explicit instruction.

Key Findings

1. A quasi-experimental pretest–posttest control group design with an embedded qualitative component was used with 48 second-year pre-service science teachers in Türkiye (25 control, 23 experimental), taught the same photosynthesis and respiration curriculum by the same instructor.
2. The eight-week experimental program combined guided inquiry, problem-based learning, and design-based learning, with AI tools (ChatGPT, Consensus, NotebookLM, Elicit, Canva, Runway, Animaker) used as scaffolding for question formulation, experiment design, literature synthesis, and digital product creation; the control received conventional instruction with verification-focused labs.
3. Mixed-design ANOVA yielded significant time×group interactions for conceptual understanding of photosynthesis, F(1,41)=22.79, p<0.001, ηp²=0.357, and cellular respiration, F(1,44)=25.70, p<0.001, ηp²=0.369; both remained significant after Holm correction, with adjusted post-test differences of 8.12 and 8.71 points, respectively.
4. No significant group-by-time differences emerged for AI literacy (F(1,42)=0.740, p=0.395) or self-perceived computational thinking skills (F(1,35)=2.384, p=0.132); the AIL scale's low post-test reliability (α=0.557) constrained measurement sensitivity.
5. Qualitative self-evaluation forms reported perceived gains in correcting misconceptions, digital content creation, critical evaluation of AI outputs, collaboration, and engagement, providing contextual but not confirmatory evidence.

Design and Context

The study addressed a gap in teacher education research: most work on AI focuses on AI literacy, attitudes, and self-assessed competence, whereas discipline-specific interventions that engage pre-service teachers in generating, evaluating, and communicating scientific evidence are rarer. Photosynthesis and cellular respiration were chosen as contexts because both require coordinating matter-and-energy transformations across molecular, cellular, organismal, and ecosystem levels, and because pre-service teachers persistently struggle to distinguish respiration from breathing and to move beyond input–output descriptions of photosynthesis.

Using guided inquiry, the researcher designed an integrated instructional environment in which AI supported disciplinary reasoning rather than serving as a standalone tool. Participants were drawn from two intact classes of a General Biology II course via purposive sampling; the higher-achieving section was assigned to the control condition to avoid giving the experimental group an initial advantage, though the authors note that class membership, prior achievement, and intervention effects cannot be fully separated. Outcome instruments included the Artificial Intelligence Literacy Scale, the Computational Thinking Skills Scale (self-report, hence "self-perceived computational thinking skills," SPCTS), and researcher-developed photosynthesis and respiration concept tests scored with a four-level rubric (91% inter-rater agreement).

Intervention

Over eight weeks the experimental group worked in groups of five on authentic biological problems (e.g., optimizing seed germination, yogurt fermentation, biofuel production). Students formulated and refined investigable questions, identified variables, designed their own experiments, interpreted evidence, revised explanations, and created scientific posters, videos, and animations. Design guidelines and instructor guidance served as scaffolds that supported learner independence rather than prescribing procedures, and explicit instruction in prompt writing helped students use ChatGPT for research-problem formulation. AI tools supported Visualization, literature synthesis, and Multimodal AI product development, with all AI-generated content verified against original academic sources and refined through rubric-based formative feedback and revision cycles. The control condition relied on textbook lessons, question-and-answer sessions, and predetermined verification labs.

Findings

Quantitative results showed significant, large gains in conceptual understanding. Photosynthesis scores rose from 22.58 to 38.53 in the experimental group versus 23.54 to 30.63 in the control, and respiration scores from 22.86 to 33.68 versus 24.54 to 25.08, with both interactions surviving Holm-adjusted correction and multiple sensitivity analyses. In contrast, no significant effects were found for AI literacy or SPCTS; a supplementary ANCOVA even favored the control on self-perceived computational thinking after baseline adjustment, though the primary interaction was nonsignificant. The authors caution that the AI-literacy null result is inconclusive given the instrument's unstable post-test reliability and reduced complete-case samples.

Qualitatively, participants reported that the activities helped them understand and reinforce concepts, correct misconceptions (e.g., that light is required in all photosynthesis stages), relate ideas to daily life, and develop digital content. They described engagement in Problem Solving, planning and decomposition, pattern recognition, debugging, and collaboration. These reflections align with science-specific computational thinking practices, yet the authors stress they represent perceived engagement rather than evidence of improvement in the broader measured constructs.

What this means for practice

• Instructors. Use AI-supported guided inquiry when the goal is disciplinary conceptual learning: the experimental group's photosynthesis scores rose from 22.58 to 38.53 against 23.54 to 30.63 in the control group (F(1,41) = 22.79, p < 0.001, ηp² = 0.357), with a comparable effect for cellular respiration (F(1,44) = 25.70, p < 0.001, ηp² = 0.369).
• Faculty developers. Make AI Literacy and Computational Thinking explicit instructional goals rather than implicit by-products: eight weeks in which AI was embedded as scaffolding produced no significant group-by-time gains in either (F(1,42) = 0.740, p = 0.395; F(1,35) = 2.384, p = 0.132).
• Faculty developers. Embed these competencies gradually across science methods, laboratory, STEM, and instructional technology courses instead of treating them as isolated technical skills, in line with the design-based and inquiry pedagogies the study used.
• Instructors. Require verification of AI output against original academic sources and teach prompt writing as part of the task: participants reported gains in correcting misconceptions, inquiry, collaboration, and critical evaluation of AI-generated content, though those are perceived rather than measured improvements.

Limitations

• The study used two intact classes of 48 second-year pre-service science teachers (25 control, 23 experimental) with purposive rather than random assignment; the higher-achieving section was assigned to the control group deliberately, so class membership, prior achievement, and intervention effects cannot be fully separated.
• The eight-week window is too short for standardized self-report measures to register change in broader competencies, which the authors give as a reason the AI literacy and self-perceived computational thinking null results should not be read as evidence of no effect.
• The AI literacy scale's post-test reliability was poor (α = 0.557) and complete-case analysis shrank the analytic samples (degrees of freedom of 41, 44, 42, and 35 rather than the full roster), reducing measurement sensitivity and statistical power.
• The qualitative component relied on reflective self-evaluation forms and therefore captures perceived engagement rather than measured skill, and the sustainability competencies the framing invokes were not directly measured at all.

Citation

Aydın, G. (2026). AI-Supported Inquiry-Based Learning in Photosynthesis and Respiration: Implications for Sustainable Science Teacher Education. Sustainability 18(18), 8643.