AutoGEO

A research framework for automatic generative engine optimization, accepted at ICLR 2026. It learns what generative engines prefer and rewrites web content so it gains higher visibility in answers produced by large language models. The pipeline has three parts: rule extraction, which mines content preferences from the engines, an API variant that uses a prompt-based model driven by the extracted rules, and a Mini variant, a cost-effective model trained with those rules. Results are measured with a GEO score for visibility, covering position, token count and citation frequency, and a GEU score for utility, covering citation quality, keypoint coverage and response quality. Rule extraction must be rerun when switching to a different engine, dataset or domain. Supported datasets, engines and metrics are listed. Written in Python under the MIT license, open source and free to use.