Prompt Archaeology: A Five-Stage Protocol for Documenting and Analysing Human-AI Research Dialogues (DRAFT)

Prompt Archaeology is a five-stage protocol for retrospectively reconstructing, classifying, and descriptively analysing the recoverable prompt exchanges that shaped a human-AI research project, including human-to-LLM, LLM-to-sub-agent, and cross-model prompts. As large language models move from writing aids to active research instruments, the chain of questions, corrections, and micro-confirmations that produced an analysis is routinely lost, while methodological discussion still treats prompts as prospective design objects (prompt engineering) rather than as retrospective research artefacts. Prompt Archaeology is the retrospective counterpart: after a project is finished or paused, it documents which prompts actually shaped the results and what that reveals about the epistemic structure of the inquiry. The protocol produces five cumulative plain-text artefacts (raw log, topic filter, typed classification, per-analysis aggregation, descriptive statistics) and an eight-type prompt taxonomy, together with a minimum reporting-and-redaction bundle and three automation routes (post-session script, in-session hook, system-level tracker). It is illustrated with a single worked pilot application to a 208-prompt regulatory investigation (127 human, 81 LLM prompts). (i) Contribution: a five-stage documentation protocol with defined, cumulative artefacts; an eight-class prompt taxonomy; and three automation routes with effort estimates; (ii) Evidential status: the pilot is a single, author-produced and author-coded illustration, not an external validation; the taxonomy is derived from and demonstrated on the same corpus; (iii) Recommended but not yet computed: inter-rater reliability (Cohen's kappa), repeated-run stability, and a prompt-freeze date are part of the protocol but are deferred to an independently coded, externally sourced case study; (iv) Positioning: complements protocol analysis, conversation analysis, dialogue-act tagging, prompt engineering, and prompt-management tools, reframed for retrospective research provenance. This version is an advanced research draft (DRAFT). It presents Prompt Archaeology as a methods proposal supported by a single illustrative pilot. Major revisions are not excluded; in particular, external multi-coder validation is planned for a future version. Changes in Version 1.0 (June 2026) Initial release. Five-stage protocol with defined cumulative artefacts and an eight-type prompt taxonomy. Minimum Reporting and Redaction Set, plus three automation routes with effort estimates. Single illustrative pilot (208 prompts) with a full, mutually exclusive type distribution; presented as illustration, not validation. Prominent limitations: circular self-coded demonstration, recommended-but-uncomputed reliability metrics, single case study. DE/EN: English and German PDFs published synchronously; combined PDF generated from EN plus DE.