To Believe or Not to Believe Your LLM: Iterative Prompting for Estimating Epistemic Uncertainty