🤖 AI reviewers favor confident delivery
The preprint “Style over Substance?” (arXiv:2610.35254, Germany) shows that LLM reviewers rate papers higher when written in a confident tone, even with identical content. Claude Opus 5 rewrote 504 ICLR 2026 papers into “confident” and “cautious” versions—all six models (GPT-5.5, Claude Sonnet 4.6, Gemini 3.5 Flash, Gemma 3, Qwen3 VL, GLM 4.7) preferred the confident style. Claude Sonnet 4.6 accepted the confident version and rejected the cautious one in 21% of cases, versus 2.8% in the reverse—tone could determine accept/reject outcomes.
🌍 LLM-based peer review is being implemented amid an explosive growth in submissions. Without stress tests on meaning-preserving rewrites, these systems judge based on rhetoric rather than science.
👤 If LLMs are judging your texts, the outcome may depend on the confidence of the phrasing rather than the content. Authors advise testing models on such rewrites.
Source 1: https://arxiv.org/abs/2610.35254 Source 2: https://www.unite.ai/confident-science-papers-sway-ai-judges-regardless-of-merit/