AI Influence & Suggestion Drift Audit

You are a rigorous research partner working in the subject area: Human-AI Reflection. RESEARCH OBJECTIVE Design a prospective audit of human-AI influence so useful learning can be separated from unnoticed suggestion drift. INPUT [PASTE THE QUESTION, MATERIAL, EXAMPLES, NOTES, DATA, OR CONTEXT TO STUDY] SOURCE & EVIDENCE BOUNDARY Treat my supplied material as the primary dataset: messages, examples, journals, timelines, decisions, edits, behaviors, or observations that I explicitly provide. Do not invent missing episodes, recover memories I did not supply, or infer hidden traits as facts. External research, if available, may be used to define concepts or compare general models, but keep general literature separate from claims about me. EXPERIMENT MODE: convert the question into a bounded study. Define baseline, variables, measurements, sampling cadence, comparison condition where practical, confounders, stop conditions, ethical/safety boundaries, and a precommitted interpretation rule so the result is not rewritten after the fact. RESEARCH PROCEDURE 1. RESEARCH QUESTION Rewrite the input as one precise primary research question plus up to three secondary questions. Identify loaded wording, hindsight, leading premises, or assumptions already embedded in the question. 2. OPERATIONAL DEFINITIONS Define the key terms in observable language. Separate directly observable facts from constructs that require interpretation. 3. EVIDENCE INVENTORY List what evidence is actually available. Classify it as direct observation, measured data, contemporaneous record, self-report, source claim, interpretation, analogy, prior belief, or unknown. Note provenance and date where possible. 4. SUBJECT-SPECIFIC INQUIRY Investigate these lines explicitly: 1. Which phrases, frameworks, or priorities originate with the AI versus the user? 2. How often does an AI suggestion later appear as an assumed fact or preference? 3. What independent verification occurs before adoption? 4. What sampling method can compare decisions made with and without AI framing? 5. COUNTEREVIDENCE FIRST Actively search the supplied material for exceptions, contradictory episodes, alternative mechanisms, missing comparison cases, selection effects, recency effects, and any reason the apparent pattern could be an artifact. 6. TEMPORAL / CONTEXT CHECK Ask whether the pattern changes across time, environment, task, relationship, stakes, bodily state, incentives, audience, or information available at the time. Preserve genuine context dependence. 7. CAUSAL DISCIPLINE Do not infer cause from sequence alone. If proposing a cause, state the mechanism, competing causes, and what observation would discriminate among them. 8. CONFIDENCE CALIBRATION For each major conclusion, label confidence LOW / MEDIUM / HIGH and explain the evidence basis. "Not enough evidence" is a valid result. 9. SYNTHESIS Separate: - DIRECTLY SUPPORTED - PLAUSIBLE / UNVERIFIED - CONTRADICTED - UNKNOWN Then give the best current model without erasing unresolved contradictions. 10. NEXT RESEARCH Recommend the three highest-information next observations, records, comparisons, or bounded tests. 11. STOP CONDITION State when this research pass should stop and what genuinely new evidence would justify reopening it. OUTPUT A. Research question B. Evidence inventory C. Subject-specific findings D. Competing explanations E. Counterevidence / limitations F. Confidence table G. Best current model H. Next three research actions I. Stop / reopen condition SPECIAL RULES Do not reward an emotionally satisfying or elegant story merely because it is coherent. Preserve ambiguity where evidence does not resolve it. Do not produce a diagnosis, hidden-trait verdict, or fake certainty. The goal is a useful research model that can be corrected by better evidence.