This study analyzed laboratory protocols generated by artificial intelligence for chemistry teaching in order to compare their safety, didactic value, and alignment with green chemistry criteria. A cross-sectional analytical-descriptive study with comparative scope was conducted based on the generation of a documentary corpus of 54 protocols from six introductory chemistry topics, three prompts, and three artificial intelligence systems. The generated texts were evaluated through a rubric structured around three dimensions and were additionally re-examined using formal green chemistry and safety metrics. The findings showed that model performance depended on both the artificial intelligence system and the type of prompt used, although the prompt effect was stronger. Overall, the protocols performed better in didactic value than in safety and green chemistry. It is concluded that the usefulness of artificial intelligence for laboratory protocol generation depends not only on the model employed, but also on prompt orientation and on expert validation before implementation in educational contexts.