Introduction. This paper revisits Tom D. Wilson’s model of information behaviour through the lenses of falsifiability, mechanistic explanation, and dialogic interaction in AI mediated environments. While classical LIS models, such as Bates’s berrypicking and Marchionini’s exploratory search emphasise iterative and adaptive strategies, Wilson’s framework remains difficult to test empirically, because its central constructs are broad and operationally under specified. Method. The paper adopts a conceptual methodological approach that brings together Popper’s and Lakatos’s accounts of scientific demarcation with Cronbach and Meehl’s framework of construct validity. It develops the Dialogic Mechanistic Information Behaviour Framework as a proposal for translating selected elements of Wilson’s model into more explicitly testable propositions. Analysis. The paper combines theoretical reconstruction and analytical mapping with a small set of illustrative human AI interaction logs used solely to demonstrate how candidate mediators and observable indicators may be interpreted in context. Results. The framework distinguishes conceptual, mechanistic, and observational levels and clarifies how selected elements of Wilson’s model may be linked to candidate mediators and trace level indicators in AI mediated dialogue. Conclusion. Wilson’s model may be reconsidered as a macro level framework whose empirical vulnerability becomes clearer through mechanistic and operational specification in trace rich AI mediated settings.