Persuasion Index: A Theory-Guided Framework for Persuasion Analysis

A team of researchers has introduced the Persuasion Index, a new taxonomy of 15 dimensions designed to analyze persuasive language in text, according to a paper submitted on 12 Jun 2026 [1]. The framework maps rhetorical strategies to established theories in psychology and communication [2]. The Persuasion Index (PI) is structured around two core components. The first is a theoretical taxonomy of 15 dimensions, which the authors ground in the Aristotelian triad of Logos, Ethos, and Pathos [3]. The second is a transparent implementation that uses 55 granular sub-features, built from lexicons and rule-based detectors [1]. Unlike neural network models that treat persuasion as a black box, PI extracts features directly from text, ensuring every score can be traced back to a specific lexical or structural pattern [4]. The framework is modular by design, meaning individual detectors can be replaced without disrupting the overall theoretical structure [5]. The researchers evaluated the framework on four public datasets that varied in domain, style, and outcome measures [2]. They found that linear models using PI features carried a meaningful predictive signal while remaining computationally lightweight [1]. A dimension-level analysis revealed recurring associations between specific PI dimensions and persuasion outcomes across the different datasets, while also highlighting variation linked to topic and stance [3]. The work addresses a persistent tension in persuasion research between predictive performance and interpretability [5]. The study of persuasion has long been a focus in social science, a field that encompasses disciplines such as psychology, communication studies, and sociology [8]. Foundational models like the heuristic-systematic model of information processing, developed in the 1980s, describe how individuals process persuasive messages either carefully or by using mental shortcuts [6]. The PI framework provides a computational method to operationalize such theories for the analysis of both human and AI-generated text [1]. The authors have released PI as an open-source package and a web interface to support auditable analysis of communication [2].