Abstract

Expert witnesses in the U.S. have begun using AI-generated outputs in their analyses. This paper critically examines such uses in “linguistic” analyses and expert reports submitted to court. We discuss how such practices raise significant ethical, constitutional, and evidentiary concerns and are largely inconsistent with U.S. legal requirements governing expert testimony including admissibility doctrines (e.g., _Daubert_, _Frye_) and statutory frameworks. Using a case study example of a “ChatGPT analysis” in an opposing expert linguist’s report, we identify tensions between judicial expectations of methodological transparency, reliability, and replicability, and the nature and function of many current AI models. Moreover, we discuss how the use of AI often fails to assist triers-of-fact in the context of specific statutory frameworks, such as the Lanham Act. We further highlight a growing tendency for courts and practitioners to conflate AI-generated output with, e.g., corpus linguistics, which risks mischaracterizing AI output as methodologically equivalent and, in doing so, undermines both established linguistic methods and uses of AI in other judicial discursive practices (e.g., legal interpretation, drafting support). As courts increasingly confront AI use, this paper aims to support judicial gatekeeping by offering a linguistically-grounded critique of AI’s role in forensic evidence and expert reasoning, and also examine how AI use in expert evidence reconfigures the semiotics of expertise itself. Specifically, we argue that these practices shift epistemic authority from human experts to AI, reshaping processes of evidentiary meaning-making and altering how expertise is performed, evaluated, and legitimated in the courtroom.