The same model can give a noticeably better answer depending on how it is asked. Prompt engineering includes giving the model examples of the format you want, stating what it should do when information is missing, and being explicit about the audience and length of the response, rather than assuming the model will guess correctly.
The common misconception is that prompt engineering is a one-time task. As business needs change and models are upgraded, a prompt that worked well can start producing worse answers, which is why prompts used in production should be tracked and checked with evals the same way code changes are tested, not just written once and left alone.