Microsoft Copilot Researcher critiques its own drafts for higher accuracy

Microsoft Copilot Researcher now uses two AI models: one for planning, retrieval, drafting; a second critiques for facts, claims, structure, evidence. DRACO benchmark shows 7-point overall gain, +2.58 factual accuracy, +3.33 analysis depth. Auto mode for Frontier users reduces manual checks in professional research.