AI can compress a week of synthesis into an afternoon—and confidently erase the most important contradiction in the room. The answer is not avoidance. It is a protocol that makes judgment visible.
1. Prepare
Remove identifying information, define the research question, and establish what the model may and may not infer. Keep raw evidence separate from generated interpretation.
2. Extract
Ask for atomic observations with source references, not themes. Each observation should be traceable to a participant, moment, or artifact.
3. Cluster
Generate several plausible groupings rather than one definitive taxonomy. Compare what each grouping reveals and conceals.
4. Challenge
Prompt explicitly for counterevidence, minority experiences, unresolved tensions, and observations that do not fit the dominant story.
5. Judge
A researcher reviews every consequential claim against source material. Confidence belongs to the human accountable for the decision—not the model producing the sentence.
6. Record
Publish the evidence chain, prompts, model and date, human revisions, limitations, and the decisions the synthesis is intended to inform.
AI may propose the pattern. A researcher must remain accountable for believing it.