Skip to content

How to Write Better AI Prompts

The same task prompted badly then well, with the real ChatGPT outputs side by side, plus the few plain habits that reliably get better results.

Eddie Ochieng

Eddie Ochieng

July 3, 2026· Updated Aug 26, 2026

5 min read
A prompt input field for Claude.ai
Photo by Planet Volumes on Unsplash

People who get disappointing results from AI usually blame the tool. Nine times out of ten it is the prompt. To show the difference rather than assert it, I gave ChatGPT the same job twice, once lazily and once properly, and kept what it returned each time. This was the free tier, August 2026.

How I tested this

Everything below is a real ChatGPT output, unedited. The point is not that one prompt is clever, it is that a few plain habits, being specific, giving a role, and refining, visibly change what you get back.

The Same Task, Badly Then Well

First, the lazy version, the kind of prompt most people actually type.

The weak prompt
Write a blog intro about email marketing.
ChatGPT · actual output

Email marketing is one of those things that seems almost too simple to still work. You write an email, send it to a list, and wait for people to open it. Yet while social platforms change their algorithms and advertising costs keep moving, email remains one of the few channels where you can build a direct relationship with an audience.

It is competent and completely generic. It could sit atop anyone’s post and speaks to no one in particular. Now the same task with the length, the audience, the tone, the format and a finish all named.

The strong prompt
Write a 150-word intro for a blog post about email marketing, aimed at
small business owners who are new to it. Friendly and practical, no jargon,
and end with a question that makes them want to read on.
ChatGPT · actual output

If you run a small business, email marketing can sound like one more thing you need to figure out. But it doesn’t have to mean complicated software, constant newsletters, or filling your customers’ inboxes with promotions. At its simplest, email marketing is a practical way to stay in touch with people who already know your business and give them a reason to come back. So, what should your first email look like, and how can you make sure people actually want to open it?

Same tool, same topic, minutes apart. The second one talks to a specific reader, respects the length, drops the jargon and lands on a question that pulls you into the piece. Nothing about that required a clever trick, only saying what I actually wanted.

Give It a Role, and It Explains Itself

Telling the model who to be shapes how it answers. I asked it to act as an editor and to show its working.

The prompt
Act as an experienced copy editor. Tighten this paragraph for clarity and
tell me what you changed and why:

"In today's world it is basically very important for businesses to always
try to make sure that they are consistently staying in touch with their
customers on a regular and ongoing basis."
ChatGPT · actual output

Businesses need to stay in regular contact with their customers.

Twenty-seven words down to nine, with the meaning intact, and then a list of exactly what it cut and why, the empty "in today’s world," the weak "it is basically very important," the repetition of "regular" and "ongoing." The role turned a rewrite into a short lesson in self-editing.

Keep Talking to It

The first answer is rarely the best one. Instead of rewriting my prompt, I just told it what to change about the strong intro above.

The follow-up
Make it shorter and less formal, and cut the last sentence.
ChatGPT · actual output

If you run a small business, email marketing can feel like one more thing to figure out. But it doesn’t have to be complicated. At its simplest, it’s just a way to stay in touch with people who already know your business.

It did the three things I asked and nothing else, shorter, looser, closing sentence gone. Treating the model as a conversation beats trying to write one perfect mega-prompt on the first try.

The Habits That Did the Work

  • Be specific. Name the length, the format, the audience and how to finish. Vague in, vague out.
  • Give context. Tell it the background it cannot know, who it is for and what matters.
  • Assign a role. "Act as a copy editor" or "a sceptical reviewer" changes the whole register.
  • Show an example. If you want a particular style, paste one and say "like this."
  • Refine in conversation. React to the reply and adjust, rather than restarting.

A prompt worth keeping

A reusable shape for a strong prompt: a role, then a specific task with length and audience, then any context, then the format and tone, and an example if you have one. You rarely need all five, but reaching for two or three is what separates a useful answer from a generic one.

These habits pay off everywhere. See them applied in our guides to using ChatGPT to write SQL queries and using ChatGPT for programming help.

FAQ

Why do I get bad results from AI?+

Usually the prompt, not the tool. In my test the same task produced a generic intro from a lazy prompt and a genuinely useful one when I named the audience, length, tone and a finish.

What is the single most useful habit?+

Being specific. Tell the model exactly what you want, including format and audience. It cannot read your mind, and the difference in the output is immediate.

Do I need long, complicated prompts?+

No. A clear two-line brief usually beats a 300-word mega-prompt. When the first answer is close but not right, refine it in conversation instead.

Does assigning a role really help?+

It did here. Asking it to act as a copy editor produced a tighter rewrite plus an explanation of every change, which a plain "fix this" would not have.

Eddie Ochieng

Eddie Ochieng

Eddie Ezekiel Ochieng is a software developer and the editor of The Test Card. He has been writing code since 2016 and has spent the last six years building production web applications for clients, work that runs from a non-profit’s platform to a community dictionary and a personal-safety service. He builds mostly in TypeScript, React and Next.js, with Node, Python and PostgreSQL behind them.

He started The Test Card out of mild irritation. Most AI tool coverage is written by people who never ship anything and never have to live with a bad tool choice. He does. The question that interests him is the practical one, which of these tools survive contact with real work, and which are just a subscription you forget to cancel.

He is not an AI researcher and does not pretend to be. What he brings instead is a builder’s scepticism, a habit of actually reading the documentation and the pricing page, and a bias toward simplicity over novelty. Good software, as he puts it, should feel as good as it works.

▶ eddie-ezekiel.com

Leave a Comment

Share your thoughts about this article