Study: gap-search method cuts homogenization in GEO content

A new arXiv preprint introduces Gap Detection, a framework that searches a shared text embedding space for relevant ideas a language model’s existing answers haven’t already covered, aiming to counter LLMs’ tendency to give near-identical responses to open-ended questions. Across five curated topics and 90 queries from the INFINITY-CHAT benchmark, the authors report mean overlap reductions of 12.8% and 10.8%; they also instantiated the method for generative engine optimization across four unnamed product markets and for synthetic training-data generation, where it cut overlap from 0.324 to 0.298 against undirected methods.