Consistency Is the Key: Detecting Hallucinations in LLM Generated Text By Checking Inconsistencies About Key Facts

Large language models frequently produce factually inaccurate outputs, a phenomenon known as hallucination. We present CONFACTCHECK, a detection method that works on the simple intuition that responses to factual probes within the generated text should be consistent — both within a single model and across different LLMs. Our approach requires no external knowledge base and fewer API calls than existing methods. Empirical evaluation across multiple datasets shows that CONFACTCHECK achieves superior accuracy compared to prior work.