The co-analyst in the loop: Claude’s affordances for applied linguistics research methods

Applied linguistics is being transformed less by “artificial intelligence” in the abstract than by the concrete affordances of specific large language models (LLMs) acting on its research-methods lifecycle. This paper is a narrative, tutorial-style methodological review advancing a model-specific, affordance-based account of how Anthropic’s Claude can support applied-linguistics research across design, data collection, transcription, coding, analysis, and reporting. It is not a systematic synthesis in the sense of Chong and Plonsky (2024); we specify our source-selection procedure and claim no exhaustive coverage. Against a tendency to treat LLMs as an undifferentiated category typified by ChatGPT and evaluated mainly as productivity aids, we reconceptualize Claude as a co-analyst in the loop: a configurable participant whose decisions can be made auditable. We identify, and explain how we derived, four affordances useful for method: extended-context document grounding, behavioural steerability, rationale-exposing output, and agentic tool use. Through worked examples—including one grounded in real interview data with a computed reliability figure—we demonstrate their application to instrument design, elicitation, transcription, coding, rating, and synthesis, using Claude as an example of a broader class rather than a uniquely capable system. We are explicit about limits: a stated rationale is not a faithful trace of internal computation, and LLMs remain only partially interpretable. We specify a human–machine analytic loop with reliability procedures addressing session-to-session variability and cross-session carryover. Claude’s contribution, we argue, is to turn tacit judgements into inspectable procedures, repositioning the human expert toward auditing and adding auditability to—not replacing—existing criteria of agreement, reliability, and construct validity.