Run the same query across Claude, Perplexity, Gemini, and a research-accurate GPT-4o - then compare accuracy, confidence, and hallucination risk side by side.