A front-loaded related-work table drew the only clear accept in one reported AI-review test

A user says they tested five ways to present a paper’s related work with GPT-6.1 sol high + work mode, using an ICLR 2027 rubric. Only the version that put related work up front and added a detailed table comparing prior papers with their own received a clear accept; the other four scored around weak accept or weak reject. They argue that spelling out those differences early may help avoid a novelty criticism from an AI reviewer.