How to write better ChatGPT prompts
Published by ModelCoach.ai — an AI prompt coaching and skills platform.
To write better ChatGPT prompts, give it the context it cannot infer, name who the output is for, and describe the format you want back. Then use the conversation rather than restarting: correcting one specific thing in a follow-up usually beats rewriting the whole prompt. ChatGPT will happily produce a confident answer from a vague request, so specificity is what separates a usable draft from a generic one.
What ChatGPT responds to
ChatGPT is agreeable by default. It will attempt almost any request, which means a vague prompt still returns a polished-looking answer — just not a useful one. That is why the failure mode here is rarely refusal and usually blandness.
It also carries the conversation forward, so the earlier turns are part of the prompt whether you intended them to be.
Practical tips
- Lead with the instruction, not the background. Put the task in the first sentence.
- Name the reader. "For a customer who has already complained twice" changes the whole draft.
- Say what to leave out. Suppressing the preamble and the closing summary is often the biggest readability win.
- Give one example of the style you want when tone matters more than content.
- Ask for a specific length in words or sections rather than saying "short".
- Correct in a follow-up. "Same email, but drop the apology and keep it under 100 words" is faster and more precise than starting again.
Using the conversation well
Because context carries between turns, a long thread can start dragging old assumptions into new answers. When the topic genuinely changes, start a new conversation rather than fighting the old one.
When you get a good result, note what you asked for. The reusable part is usually a single clause — an audience, a format, a constraint — not the whole prompt.
Common mistakes
- Asking a question you have not decided the answer criteria for, then rejecting drafts on instinct.
- Pasting a long document with no instruction about what to do with it.
- Rerolling the same prompt hoping for variance instead of changing one element.
- Assuming a confident answer is a verified one. Check anything factual before it leaves your hands.
- Sharing confidential or customer information with any external tool without checking your organisation's policy.
The same principles travel
Almost none of this is specific to one assistant. Task, context, audience, format, and constraints matter the same way in Claude, Gemini, and Microsoft Copilot. The interface changes; the reason a prompt underperforms does not.
Frequently asked questions
- What makes a ChatGPT prompt effective?
- A clear instruction, the context the model cannot infer, a named audience, and a described output format. Those four cover most of the gap between a generic answer and a usable one.
- Should I start a new chat or keep going?
- Keep going while you are refining the same piece of work — the accumulated context helps. Start fresh when the topic changes, because old context will keep influencing the answers.
- Do the same tips work for Claude, Gemini and Copilot?
- Largely yes. The wording style differs slightly between assistants, but the underlying reasons a prompt fails — missing context, missing audience, unstated format — are the same everywhere.
Related resources
ChatGPT, Claude, Gemini and Microsoft Copilot are trademarks of their respective owners. ModelCoach.ai is an independent product and is not affiliated with, endorsed by, or sponsored by OpenAI, Anthropic, Google, or Microsoft. Product names are used for identification and compatibility only.