The same prompt can produce different results each time. This is by design and reflects how models sample from probability distributions.
Similarly, the quality of the response depends on the quality and relevance of the context provided in the prompt.
This is what makes AI so difficult for me. There's so many best practices and dogma being thrown around, but how can you meaningfully prove what's better vs different. How can you rely on something that has inherently unreliable mechanisms? There's a lack of consistency, which is scary.