How do you get better answers from ChatGPT, Claude or Gemini? Prompt engineering without the hype

I use these tools daily for writing and research and I still get generic mush half the time. Everyone sells a "prompt engineering course". What actually makes a difference in practice?

The things that changed my results, roughly in order of impact:

1. Give it the actual material. Paste the document, the code, the data. Most bad answers come from the model guessing what you are looking at.
2. Say who it is for and what "good" looks like. "Rewrite this for a customer who is angry and technical" beats "make this better".
3. Show one example of the output you want. One good example beats three paragraphs of description.
4. Ask for a plan before the work on anything longer than a paragraph. Fix the plan, then say "go".
5. Tell it to ask you questions first when the task is underspecified. The newer models are good at this and it saves a round trip.
6. Constrain the format: length, headings or not, bullet or prose, what to leave out.
7. Iterate in the same conversation instead of restarting. "Same, but shorter and drop the second point" is a perfectly good prompt.
8. Verify anything that looks like a fact, especially numbers, dates and citations. Ask for sources and then actually open them.

The "act as a world-class expert" and "think step by step" incantations matter far less than they did two years ago; the reasoning models do that on their own. Context and examples are the whole game.

For technical problems: paste the exact error text, the versions of everything involved, and what you already tried. Then ask for the three most likely causes ranked, and how to test each one, instead of "fix this". You learn something and you can tell when the answer is wrong.

For creative work, ask for ten rough variations in one go, pick the two you like, and ask it to combine them. Asking for "the best" version gets you the average one.