Answers are grounded in this paper and cite their sources.
Generative artificial intelligence tools such as ChatGPT are increasingly being explored as assistants for chemistry instructors, including the preparation of teaching materials. Here, we examined whether ChatGPT, using Images 2.0, can generate scientifically correct images for first-year undergraduate chemistry students, focusing on visual and symbolic representations central to the discipline. A total of 444 images covering six chemistry areas were generated. Short, minimally descriptive prompts were used first; when important errors appeared, more detailed corrective prompts were introduced. We also generated instructional infographics combining several representations to explain chemical concepts. Scientific accuracy depended strongly on image complexity, with simpler orbitals, titration curves, and molecular structures generally performing better than more demanding examples. When the six categories were equally weighted, 43.5% of basic-prompt images, 45.6% of images from corrective prompts, and only 8.3% of instructional infographics were directly usable. Embedded text was generally more accurate than the images. Expert checking therefore remains essential before classroom use.
No takes yet. Share an insight, caveat, or question.
Fernández et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: