Grzegorz Chodak

As large language models (LLMs) gain millions of individual and business users, their rapid adoption has raised critical questions about economic sustainability and commercialization strategies. This paper examines monetization strategies for LLMs, providing a structured overview of existing and emerging revenue models, including subscription and pay-per-use access plans. Particular emphasis is placed on the potential of advertising-based models that leverage the unique strengths of LLMs. The paper also presents eight potential scenarios for the commercial future of LLMs, ranging from advertising-based, hybrid, and public sector funding models to the dominance of open source solutions and blockchain-based microtransactions. Drawing parallels to the evolution of search engine monetization, the authors argue that hybrid approaches (e.g., combining advertising revenue with other sources) may represent the most viable future for funding LLMs. This work contributes to a deeper understanding of the commercial forces shaping the next phase of generative artificial intelligence deployment.

Image crises pose significant challenges for organizations and public figures, often requiring rapid identification and classification to mitigate reputational damage. This study introduces a novel typology of brand crises and demonstrates its application using large language models (LLMs) to enhance crisis detection and classification. We review the current state of knowledge of brand crises and LLMs, underlining their relevance in real‐world text analytics tasks. Based on an analysis of 300 actual crisis cases, we propose an original typology that captures various types and causes of crises. Our methodology combines expert data annotation with automatic crisis type annotation using a generative LLM. This approach enables both classification and early detection of crises in media texts. The results demonstrate that the GPT‐4‐turbo achieved strong performance in distinguishing ideological from nonideological crises (accuracy: 0.903; F1: 0.874), while GPT‐5 with a 2‐shot prompt and GPT‐4o‐mini excelled in identifying affected actors (accuracy and F1: 0.984). Performance was comparatively lower for detailed cause classification, highlighting the greater complexity of fine‐grained categorizations. This study highlights the potential and limitations of LLMs in developing automated crisis management systems to enhance organizational resilience.