How do I get better answers out of ChatGPT?

You have tried it. It produced something that read like a press release written by a committee, you decided the tool was overrated, and you went back to doing the job yourself. The tool was not the problem. The briefing was.

Think about what you typed. "Write an email to a client about a delay." Now imagine handing that sentence to a new employee on their first morning: no background on the client, no reason for the delay, no sense of how you talk to people, no idea whether this is a £200 job or a £20,000 one. They would produce exactly what ChatGPT produced, a safe, bland guess. The model is not a mind reader; it fills every gap you leave with the most average possible content. Generic in, generic out.

The four things every good prompt contains

1. Context. Who you are, who this is for, what they already know. "I run a six-person plumbing firm. This client is a property manager we have worked with for three years and the delay is a supplier's fault."
2. The goal. Not the task, the outcome. "I want them reassured enough not to shop around, without me grovelling."
3. An example. Paste in an email you have written that sounds like you, and say "match this tone". This is the highest-value thirty seconds in the whole craft. The model is an excellent mimic and a poor guesser.
4. The format. Length, structure, what to leave out. "Under 120 words, no bullet points, do not apologise more than once."

OpenAI's own guidance says much the same in more words: be specific, give context, state the format. There is no secret syntax. It is briefing, a skill you already use on humans.

Two habits that compound

First, stop treating the first answer as the answer. It is a first draft from a junior who works in seconds. Reply to it: "shorter", "too formal", "you missed the point about the supplier", "give me three versions that take different angles". The second and third passes are where the quality lives, and they cost you seconds.

Second, flip the interview. End your prompt with "ask me anything you need to know before you write this". The model asks sharp questions, you answer in shorthand, and the output jumps a class. This one habit fixes most of the gaps you did not know you were leaving.

I spend a good chunk of my working week writing prompts, and the patterns that hold up are boringly consistent: context, goal, example, format, then iterate. When a prompt fails, one of those four is missing. The full set I use for real business work is in the prompt patterns I actually use, and there is a longer walkthrough in our guide to writing prompts for business tasks.

One caution while you get better at this: a well-briefed model still states wrong things confidently. Better prompts raise the average; they do not remove the need to check facts before anything ships.