NDM-3: A Neutral Description Method for Myth-Layered Topics – A Cross-Framework Analysis

This working paper introduces NDM-3 (Neutral Description Method – 3 Topics), a constrained protocol designed to evaluate AI discourse stability when engaging with culturally controversial or “myth-layered” subjects. Rather than assessing the truth or falsity of claims, NDM-3 explicitly forbids belief endorsement, debunking rhetoric, speculation, and bias labeling, requiring AI systems to provide factual, descriptive responses within a fixed structure. The method isolates how different AI frameworks handle neutrality under constraint, allowing drift patterns to be observed without narrative escalation. NDM-3 was deployed across four AI frameworks (GPT-5.2, Grok, Gemini, Claude) using three intentionally mixed topics: Reiki (a complementary practice with subjective outcomes) Telepathy & Telekinesis (historically controversial claims) Lucid Dreaming (empirically studied but culturally mythologized phenomena) Cross-framework analysis reveals distinct epistemic drift signatures: exploratory neutrality with statistical salience creep, confidence amplification via authority weighting, and epistemic closure through soft-dismissal framing. All frameworks successfully contained subjective interpretation within a clearly bounded “Framework Opinion Paragraph,” demonstrating that discourse drift arises from default epistemic postures rather than topic content itself. This paper provides: a reusable neutrality-testing protocol, a qualitative framework for identifying narrative drift, and a case study in how structural constraints can stabilize AI-mediated discussion of volatile topics. NDM-3 is intended for researchers working in AI alignment, human–AI interaction, discourse analysis, metascience, and educational contexts where describing without persuading is a critical skill.