Chenyan Xiong

Xiong is an associate professor in the Language Technologies Institute at Carnegie Mellon, with a research agenda around large language models and information retrieval. His faculty page and LTI profile place him in the same CMU neighborhood as other retrieval and language researchers who treat search as a systems problem.

Xiong's group released AutoGEO, accepted at ICLR 2026, as a framework that extracts generative-engine preference rules and rewrites web content for more visibility while tracking generative-engine utility. The project includes AutoGEO_API, a plug-and-play prompt-based approach for frontier models, and AutoGEO_Mini, a smaller model fine-tuned with reinforcement learning for lower-cost rewriting.

Code and papers are public through the group's GitHub and project site. AutoGEO tries to learn engine preferences and apply them systematically.

Earlier GEO practitioner advice often listed content edits that seemed to help: citations, statistics, clearer answers. AutoGEO treats those as things a system might discover and apply, then evaluates whether rewrites help without destroying utility for users or engines.

Pranjal Aggarwal's GEO-bench work defined how to measure visibility; AutoGEO pushes toward scalable optimization under learned preferences.