Organizations like national statistics offices increasingly share data used by AI tools in online searches, but AI-generated answers can misattribute or overinterpret evidence. To ensure responsible AI use, publishers should provide clear definitions, limitations, and example questions with annotated guidance alongside their data. This approach helps AI developers verify that answers are faithfully supported by evidence, transparently acknowledge uncertainties, and maintain consistency between human-readable and machine-readable information. Collaboration between data producers and AI developers, including shared test questions, enhances the reliability and accountability of AI-generated knowledge.