Getting Better Results From Any AI Chatbot
You don't need special "prompt engineering" tricks — you need to give the chatbot the same context you'd give a person who wasn't there for the backstory.
My mother-in-law once showed me a chatbot conversation where she'd typed "write me a letter" and gotten back something generic enough to be useless, and she'd concluded the tool just wasn't very good. The tool was fine — it had no idea who the letter was to, what it was about, or what tone she wanted, because none of that was in the message.
The single biggest improvement for anyone, technical or not, is including context a person would need too. "Write me a letter" gives an assistant nothing to work with; "write a short, warm letter to my neighbor thanking her for watering my plants while I was on vacation" gives it a recipient, a tone, and an occasion — the same information you'd naturally include if you asked a friend to help you write it.
Second, it's fine to correct and continue in the same conversation rather than starting over. If the first draft is too formal, saying "make this more casual, like I'm talking to a friend" works better than abandoning the chat and trying a completely different opening prompt — the assistant keeps the context from the first attempt and just adjusts.
Third, ask it to show its work when the answer matters — for anything you're going to rely on, like a medical or legal question, or a factual claim you plan to repeat to someone else, ask where the information came from, and treat "I'm not certain, you should verify this" as the honest and correct answer rather than a disappointing one.
None of this is a special skill reserved for programmers — it's the same communication instinct you already use talking to a new coworker who doesn't have the backstory yet: more context in, less guessing required, better result out.