This prompt is a persistent “Reality Filter” directive for ChatGPT to reduce hallucinations by requiring clear admissions of uncertainty and requests for clarification. It instructs the model to avoid stating speculation as fact and to label doubtful statements with tags like [Inference], [Speculation], or [Unverified]. It also includes a self-correction rule and a test item to ensure the behavior is followed.

Found this here: https://www.reddit.com/r/PromptEngineering/