How to Write Better ChatGPT Prompts Without Complicated Formulas

A prompt is simply the instruction, question or information you give ChatGPT to tell it what you want. That's it. There is a small industry built around making prompting seem technical, but the part that actually works is ordinary: communicate the task clearly, the way you would to a capable but uninformed colleague.

This article is about the handful of things that reliably improve a prompt — and why no single formula is "correct."

What a prompt is

A prompt can be one sentence or several paragraphs. It can be a question, a request, a role to play, a set of facts to work from, or any combination. The model responds to whatever you write, so the quality of your request sets the ceiling for the quality of the answer.

Why context matters

ChatGPT doesn't know your situation unless you tell it. The same question asked with and without context produces very different answers.

Before: "Write a follow-up email."

Improved: "I sent a proposal to a small business owner two weeks ago and haven't heard back. Write a short, warm follow-up email that nudges gently without sounding pushy."

The second version gives the model a person, a situation, a goal and a tone. It has far less to invent, so the answer lands much closer to what you need.

Clear instructions

State what you want the model to do, in plain verbs. "Analyse," "summarise," "list," "rewrite," "compare," "draft." Vague verbs ("look at," "think about") tend to produce vague answers.

Before: "Look at this paragraph and tell me what you think."

Improved: "Read this paragraph and point out any sentences that are vague or unsupported, then suggest a specific fix for each."

The second version tells the model exactly what kind of feedback you want.

Desired output

Tell the model what shape you want the answer in. A few sentences? A table? A numbered list? A bullet list with bold headings? Specifying the format saves a follow-up.

Before: "Give me tips for sleeping better."

Improved: "Give me five practical tips for sleeping better, as a numbered list, each with a one-line explanation."

Constraints

Constraints are the edges of the task — what to avoid or include, length, audience, tone. They stop the answer from drifting.

Before: "Explain retirement saving."

Improved: "Explain retirement saving for a 30-year-old in the UK who has never invested, in under 200 words, avoiding jargon and not giving specific financial advice."

The last constraint matters: it tells the model the limit of what it should do, which is often as important as what it should do.

Examples

Showing the model an example of what you want is one of the most powerful things you can do, especially for formatting or style.

Improved, with an example: "Rewrite these product descriptions in a calm, factual tone. Example of the style I want: 'A lightweight rain jacket with a quiet, matte finish — made for everyday commuting rather than the mountains.' Now rewrite these three: …"

The example does more work than a paragraph of description.

Follow-up prompting

Your first prompt rarely produces the perfect answer, and it doesn't need to. Treat the exchange as a conversation:

• "Too long — cut it by half."
• "Good, but the second point is off; replace it with something about backups."
• "Rewrite the whole thing for a non-technical reader."

Most genuinely useful answers arrive after one or two refinements, not on the first try.

Why there is no universally perfect prompt formula

You'll see frameworks that promise a fixed structure — role, task, context, format, examples, constraints, in a particular order. These can be helpful training wheels, and they're not wrong. But the order and the ingredients that matter depend entirely on the task.

A brainstorming prompt needs room and options; a drafting prompt needs constraints and tone; a summary needs source material and a length. Trying to force every request through the same template often produces longer, more artificial prompts without better answers.

The reliable skill underneath all of it is the one you already have: say what you mean, give the relevant detail, and tell the other party what good looks like. If you'd like ready-made examples to adapt, The Everyday ChatGPT Prompt Library collects them by everyday task.