Preserving Participant Meaning in AI-Mediated Multilingual Qualitative Research

<ns5:p>Artificial intelligence is rapidly entering multilingual qualitative research through automated transcription, translation, coding, summarisation and quotation editing. Existing scholarship has examined many of these practices separately. Yet qualitative claims are usually produced through a sequence of transformations, and an apparently minor alteration at one stage can shape every subsequent stage. I therefore ask: How does AI mediation across the multilingual qualitative evidence pipeline alter meaning, epistemic authority and responsibility, and what methodological safeguards should follow? The research conducted a critical integrative study of cross-language qualitative methods, AI-assisted qualitative analysis, machine translation bias, reflexivity, and epistemic justice. The synthesis identifies an algorithmic interpretive layer located between participant expression and researcher interpretation. This layer does not merely transmit language. It selects, normalises and reorganises it. Five linked mechanisms explain how meaning can be progressively narrowed: semantic compression, linguistic normalisation, category anchoring, evidentiary laundering and accountability diffusion. Their interaction produces cascading meaning loss, in which transformations that appear acceptable in isolation become consequential when inherited by subsequent analytical decisions. To address this problem, the study proposes an AI-Mediated Interpretive Chain of Custody. The framework requires versioned source preservation, transformation logs, model and prompt provenance, risk-based human adjudication, interpretive checkpoints, and claim-to-source traceability. It treats translation as situated interpretation rather than a search for one mechanically correct equivalent. The article advances qualitative methods by shifting attention from the accuracy of individual tools to the integrity of the complete evidence pathway. It offers researchers, ethics committees, editors and reviewers a practical basis for judging when AI-supported multilingual findings remain meaningfully connected to participant expression.</ns5:p>