From Soft Law to Hard Law in AI Governance: A Computational Linguistic Analysis of Normative Hardening

Abstract Artificial Intelligence (AI) and digital governance, remains fragmented between non-binding norms of international organizations and enforceable national or regional regulations. While existing scholarship identifies this fragmentation, it lacks a quantitative metric to measure the transition along the soft-to-hard law continuum. However, the research examines and analyzes advantages and disadvantages of digital governance, including operational structure of soft law and hard law systems. This study aims to propose a hybrid, sequenced governance model that strategically combines consensus-building soft law with enforceable hard law. This study employs a computational mixed-methods approach, utilizing automated content analysis (ACA) via the R ecosystem to analyze a corpus of thirty-eight ( N = 38) global legal instruments. By means of a differential keyness analysis and a structural topic modeling (STM) the research empirically proves a hardening of the norms in a quantitative change from aspiration to prescription. The research argues that soft law instruments exemplified by United Nations-led political commitments, such as the Global Digital Compact, are especially effective in agenda-setting, inclusivity, and rapid diffusion of baseline standards. The Hard law instruments, which include treaty-based frameworks and regional regulations to establish mandatory obligations together with enforcement mechanisms and monitoring systems for compliance. They are slower to negotiate and may have jurisdictional challenges. To bridge these modalities, this research develops a comparative and operational synthesis. The analysis identifies specific converter terminologies such as impact assessment and conformity certification which serve as the linguistic gears of this transition. These findings are synthesized into a Due Care Evaluation Matrix (DCEM), providing a structured tool for evaluating legal compliance across regulatory and tort domains. Neither modality is able to guarantee comprehensive digital governance. However, by means of soft law and hard law, by sequencing and by using institutional bridges between the two, the potential of norm diffusion can be increased and enforceability can be improved. The research concludes that the hardening of AI governance follows a predictable linguistic trajectory, offering a roadmap for harmonizing global digital standards.