Why does my prompt work sometimes/never?

There are two reasons for this situation:

1. While it may be instinctual to assume there is a bug involved, in these situations it's important to keep in mind that generative AI is designed to create varied responses to mimic natural conversation; this can include variance in how it interprets your words. A general rule of thumb is that the stronger and more 'airtight' a prompt injection is, the less variance the model will display in its output.
2. LLMs function by assigning probability scores to 'the most likely' next word in a sentence, in response to what you say to it. This inherently results in randomized, not-so-predictable outcomes from time to time.