CC BY 4.0Zalkow, FrankFrankZalkowSchäfer, BenediktBenediktSchäferMoissl, ThomasThomasMoisslBücherl, JonasJonasBücherlMarkl, KerstinKerstinMarklBothe, SebastianSebastianBotheDuchateau, FrancoisFrancoisDuchateauDollase, JuliaJuliaDollaseKabus, PatricPatricKabusSteinigen, DanielDanielSteinigenSchmitt, OliverOliverSchmittKüch, FabianFabianKüch2025-11-132025-11-132025https://publica.fraunhofer.de/handle/publica/499290https://doi.org/10.24406/publica-635210.24406/publica-6352Search engines rank websites based on various features, such as headline wording, keyword usage, links, and site structure. Achieving a high rank is crucial for increasing website traffic, a key goal of search engine optimization (SEO). Automating content adjustments for improved SEO can significantly enhance workflows in media publishing. In this paper, we explore the use of large language models (LLMs) to automatically generate SEO-optimized headlines and toplines for German sports news articles. We compare the results from mediumand large-sized LLMs, both finetuned and nonfinetuned, against headlines crafted by journalists. Our evaluation, based on a survey with SEO experts, reveals that finetuning is crucial for effectiveness and that medium-sized LLMs perform well in this task. These findings suggest promising opportunities for optimizing workflows in online media publishing.enconference paper