5 ChatGPT prompting mistakes most people make (and how to fix them)

I have written thousands of AI prompts over the past two years. In the early days, most of my results were mediocre. Generic. The kind of thing that makes you think "is this really what all the fuss is about?"

The problem was not the AI. It was me. I was making the same mistakes that almost everyone makes when they first start using ChatGPT, Claude, or Gemini.

Once I fixed these five things, the quality of my results went through the roof. I went from getting bland, cookie-cutter responses to getting outputs I could use with minimal editing. Let me show you what I mean.

Mistake 1: being too vague with your prompt

This is the number one mistake I see. People type something like:

And then they are surprised when the result is generic and useless. Of course it is. You gave the AI almost nothing to work with.

The AI does not know who the email is to, what the meeting is about, what tone you want, how long it should be, or what the key message is. So it guesses. And guesses are rarely what you need.

See the difference? You have told the AI exactly who, what, why, how long, and what tone. The result will be dramatically better.

The rule: if you would not give a human colleague that instruction and expect a good result, do not give it to AI either.

Mistake 2: not giving the AI a role

This was a game-changer for me. When I started telling the AI who to be, the quality jumped immediately.

When I was building our AI tutor for students, I experimented with this extensively. The difference between asking ChatGPT to "explain binary" versus telling it to "act as a patient, encouraging GCSE computer science teacher explaining binary to a 14-year-old who is struggling with maths" was night and day.

The first response was dry and technical. The second used simple language, relatable analogies, and broke things down step by step. Same AI, same topic, completely different result.

How to do it: start your prompt with "You are a..." or "Act as a..." followed by a description of the expert you need.

Some examples that work well:

• "You are an experienced marketing copywriter who specialises in small business websites."
• "Act as a friendly HR manager helping me draft a difficult conversation."
• "You are a financial advisor explaining pension options to someone with no finance background."

The more specific the role, the better the output. Telling it you are a "writer" is fine. Telling it you are "a direct-response copywriter who writes in short, punchy sentences for UK small business owners" is much better.

Mistake 3: treating it as a one-shot conversation

Most people type one prompt, read the response, and then either use it or give up. That is like asking a colleague to draft something and accepting the first draft without any feedback.

AI conversations are meant to be iterative. The first response is a starting point, not the final answer.

Here is what I actually do in practice:

1. Give my initial prompt with as much detail as possible.
2. Read the response and identify what is good and what is not quite right.
3. Ask it to revise specific parts. "Make the opening more direct." "Remove the third paragraph." "Change the tone to be less formal."
4. Repeat until I am happy.

I recently used this approach to write a set of exam revision notes. The first draft was about 70% there. After three rounds of refinement, it was exactly what I needed. Total time: about five minutes. Writing it from scratch would have taken over an hour.

The rule: think of AI as a conversation, not a vending machine. You put in a request, review what comes out, and refine.

Mistake 4: not showing it what you want

If I asked you to write something "in my style" but gave you no examples of my writing, you would struggle. AI is exactly the same.

One of the most powerful techniques I use is giving the AI an example of what good looks like. This is sometimes called few-shot prompting, but you do not need to remember that term. Just think of it as "show, do not just tell."

The AI will match your tone, length, structure, and style. It is remarkably good at this.

I use this constantly for our business. When I need new content that matches our existing brand voice, I paste in two or three examples and tell the AI to follow the same style. The results are consistent every time.

The rule: whenever possible, show the AI what you want rather than trying to describe it. One good example is worth a hundred words of description.

Mistake 5: not setting constraints

Here is something I learned the hard way. Without constraints, AI tends to waffle. It will give you long, verbose responses packed with filler phrases like "in today's fast-paced world" and "it's important to note that."

Sound familiar? That is because the AI is trying to be helpful by giving you more. But more is not always better.

Setting clear constraints fixes this immediately. Here are the ones I use most often:

• Length: "Keep this under 200 words" or "Write exactly 5 bullet points."
• Format: "Use bullet points, not paragraphs" or "Format this as a numbered list."
• Tone: "Write in a casual, conversational tone" or "Keep this formal and professional."
• Audience: "Write this for a business owner with no technical knowledge."
• What to avoid: "Do not use jargon" or "Do not include an introduction, just get straight to the point."

That last one is my favourite. Telling the AI what not to do is just as powerful as telling it what to do. When I tell it "do not start with a generic introduction" or "do not use the phrase 'in today's world'", the output is immediately more natural and useful.

What does a really good prompt look like?

Let me put all five fixes together in one example. Say you need to write a LinkedIn post about a new service your business is offering.

That prompt includes a role, specific context, constraints on length and tone, things to avoid, and an example. The result will be incomparably better.

Quick reference: the 5 rules for better prompts

Want to go beyond the basics? There is a technique most people do not know about

The five rules above will dramatically improve your results. But there is a technique I use almost every day that takes things to another level entirely.

The first time I tried it, I was genuinely shocked at the difference. It completely changed how I approach every AI conversation. I am not going to give the full method away here because it is one of the most valuable things we teach. But I will say this: once you learn it, you will never go back to writing prompts the old way.

What should I try next?

The best way to improve is to practise. Take something you need to write this week - an email, a report, a social media post - and try applying all five rules to your prompt. Compare the result with what you would normally get.

Or if you want us to train your team directly, get in touch about a workshop. We run half-day sessions where your team practises these techniques with their own real work tasks. It is the fastest way to level up.