Answers are grounded in this paper and cite their sources.
We analyze 1.17 million papers across five engineering disciplines (civil/construction/environmental, mechanical, electrical, chemical, and materials) together with computer science as a benchmark, and report that large language model (LLM) marker frequency increased 1.72–2.05 × (Cohen's d = 2.3–2.9) post the release of ChatGPT in 2022. Our estimation suggests that, by early 2025, approximately one in four engineering abstracts exhibits LLM-marker density at or above the pre-LLM 95th percentile. Our findings specifically note that LLM marker density correlates positively with citation counts (Spearman's ρ = 0.18, p < 0.001). In addition, we report that nonnative English speakers exhibit steeper adoption with higher marker rates than native speakers (1.97 × vs. 1.46 ×), and first-author affiliations were dominated by China (20%) and the United States (11%). Early-career researchers show 32% higher adoption of LLM markers in collaborative work than senior scholars; however, this pattern reverses in single-author papers, where established mid-career researchers lead adoption. We also document coevolutionary adaptation of the most commonly identified LLM markers and observe their rise and fall. Collectively, our results indicate that the rise in LLM-characteristic language is consistent with reduced linguistic barriers to publication in English-dominant venues, while the more substantial uptake among early-career authors indicates a generational gradient likely to widen as cohorts progress through the academic pipeline. If the observed trajectories continue, majority prevalence appears imminent (within 2 to 3 years) and is likely to fundamentally redefine baseline standards for how engineering research is written and presented.
No takes yet. Share an insight, caveat, or question.
McCreery et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: