A Two-Phase AI-Assisted Research Protocol for Bias-Controlled Evaluation and Ideation

This paper proposes a practical, two-phase interaction protocol for AI-assisted research designed to mitigate author-induced bias and contextual conditioning in conversational AI workflows. Rather than modifying AI models or relying on prompt phrasing alone, the protocol restructures the interaction itself by explicitly separating evaluation and collaboration into two sequential phases. In Phase 1, an anonymized and memory-isolated AI interaction is used to obtain critical assessment of a manuscript without authorship disclosure. In Phase 2, authorship is explicitly revealed and the AI transitions into a collaborative brainstorming role. A qualitative case study demonstrates how this deliberate phase separation alters interaction dynamics, producing clearer evaluative feedback and more productive ideation without changing the underlying AI system. The protocol functions as a bias-aware scaffold for early-stage research, complementing traditional peer review rather than replacing it. This work highlights that the effectiveness of AI-assisted research depends not only on model capability, but also on the design of human–AI interaction structures. The proposed protocol is intended as a reproducible, tool-agnostic guideline for researchers integrating conversational AI into evaluative and exploratory workflows. License: CC0 (Public Domain)