Hallucinations in AI-generated data, where incorrect information appears factual, present a significant challenge, particularly in both structured and unstructured data. To address this, the article discusses the use of Guardrails AI provenance validators, which help reduce inaccuracies by confirming AI outputs against source documents using embeddings and large language models (LLMs). These validators, including provenance v0 and v1, can automatically detect and remove unsupported claims in AI outputs by leveraging tools like Cohere and OpenAI. Provenance v0 confirms responses sentence-by-sentence using embeddings, while provenance v1 performs self-checks by prompting another LLM. The article provides a step-by-step approach to implementing these validators, emphasizing their potential to enhance the reliability and accuracy of AI outputs, thus building greater trust in AI-powered solutions.