As generative AI tools such as ChatGPT, Claude, and Gemini become increasingly integrated into academic writing, ethical concerns around transparency and accountability are growing. While major journals now require authors to disclose AI use, few demand records of how these tools were used or what prompts were given. This Perspective highlights six overlooked risks of undocumented AI use, including hallucinated citations, unclear responsibility, and reproducibility barriers. We propose a simple, practical solution: saving AI prompts and outputs as timestamped PDFs. This practice promotes traceability, aligns with FAIR principles, and can help rebuild trust with editors and reviewers. We also share open-access templates for documenting AI use in line with journal policies. Preserving prompts is not just a technical measure—it affirms the foundational principle that humans direct, and remain responsible for, the work they publish. Cultivating this culture of prompt transparency is essential for the future credibility of science.
Kimio Shiraishi (Mon,) studied this question.