Haritz Puerto

"Our experiments reveal that most current C-SEO methods are not only largely ineffective but also frequently have a negative impact on document ranking, which is opposite to what is expected," Puerto and his co-authors wrote in C-SEO Bench: Does Conversational SEO Work? The benchmark, built with Martin Gubri, Tommaso Green, Seong Joon Oh and Sangdoo Yun, covered question answering and product recommendation across six domains, from retail, video games and books to web, news and debate questions, and measured what happens as more competitors adopt the same tactics.

Moving a document higher in the list the model read did far more than rewriting it: "traditional SEO strategies, those aiming to improve the ranking of the source in the LLM context, are significantly more effective." As the number of adopters grew, the gains shrank, which the authors described as the "congested and zero-sum nature of the problem." A breakdown of the benchmark's results charts the rank changes by method and domain.

Puerto earned a BSc in computer science and engineering at the University of Málaga in 2017 and an MS in computer science at KAIST in 2021, where he was a research assistant in the IR&NLP Lab. Before his doctorate he worked at the Coleridge Initiative, which organized the Kaggle competition Show US the Data.

He completed his PhD summa cum laude at the UKP Lab at TU Darmstadt under Iryna Gurevych, defending in September 2026, and interned during it at Parameter Lab, one of the affiliations on C-SEO Bench. In March 2026 he joined Sahar Abdelnabi's COMPASS group at ELLIS, with a co-affiliation at the Max Planck Institute for Intelligent Systems. His personal site lists later papers on membership inference attacks against LLMs, privacy leaks in the reasoning of large reasoning models, and whether models that know how evaluations are designed score safer.