Writing Science for Human Readers and AI Systems

Researchers spend less time reading each paper as publication volumes explode, while AI systems increasingly process scientific literature but struggle with unstructured content. We propose restructuring scientific papers to serve both human readers and AI systems through front-loaded narratives and structured appendices. Front-loading places key findings and significance in opening paragraphs, enabling time-pressed researchers to assess relevance immediately. Structured appendices transform supplementary materials from passive data dumps into machine-readable knowledge bases containing executable code, documented protocols, and FAIR data repositories. This dual-audience framework addresses current communication failures while creating infrastructure for responsible AI integration. Publishers can differentiate themselves by hosting interactive knowledge packages rather than static PDFs, enabling AI-assisted querying, translation, and analysis. The approach restructures peer review by allowing efficient evaluation of front-loaded narratives and precise auditing of computational components through executable appendices. Widespread adoption would generate a curated corpus of structured, verified research data superior to current AI training materials, thereby creating a virtuous cycle where better scientific infrastructure produces more reliable AI assistants.