Building and Maintaining Trust in AI Chatbots: Introducing Generative Information Retrieval Accountability Assessment Framework (GIRAAF)

Social integration of AI requires public trust and appropriate governance. This paper introduces the Generative Information Retrieval Accountability Assessment Framework (GIRAAF), designed to appraise accountability for Large Language Models and their applications, including AI chatbots. Drawing on Dahl’s (1998) notion of enlightened citizens, the framework comprises six interlinked dimensions: Information Readiness, Factual Correctness, Normative Alignment, Sourcing Practice, Transparency and Explainability, and Responsiveness and Adaptivity. Together, these dimensions link observable text outputs to intelligible standards, enabling systematic evaluation and accountable oversight. For scholars, developers and policymakers, GIRAAF provides a diagnostic heuristic and shared vocabulary for critical analysis of AI chatbot performance.