A business owner I know typed one line into ChatGPT last month: "Write me a marketing email for my spring sale." Thirty seconds later she had an email. It was fine. It was also completely generic — no mention of her products, her customers, or her tone of voice. She copied it, felt a little deflated, and closed the tab. "AI just isn't that helpful for real work," she told me.

I hear a version of this story constantly. The AI didn't fail her. The prompt did. She gave the tool almost nothing to work with, so it gave her back almost nothing in return. The good news is that fixing this has nothing to do with learning code or memorizing technical commands. It's a communication skill, and once you see the pattern, you can't unsee it.

What "Prompt Engineering" Actually Means for You

"Prompt engineering" sounds like something only programmers do. Strip away the jargon and it just means: giving clearer instructions. That's it. Every AI chat tool — ChatGPT, Claude, Gemini, whatever you use — is trying to guess what you actually need from very little information. The vaguer your request, the more guessing it has to do, and the more generic the result.

Think of it less like typing a search query and more like briefing a new hire. A search engine can work with three words because it's just matching keywords. An AI assistant is trying to produce original writing, and original writing needs direction. The instructions you'd give a new employee on their first day — who this is for, what tone to use, what "done" looks like — are exactly the instructions an AI needs too.

Start With Who, What, and Why

Before you even state your request, give the AI three things: a role, an audience, and a purpose. This single habit fixes more bad AI answers than anything else.

  • Role: Tell it what kind of expert to act as. "Act as an experienced email marketer" produces a different email than no instruction at all.
  • Audience: Tell it who will read the result. A message to loyal repeat customers should not sound like a message to first-time visitors.
  • Purpose: Tell it what you want to happen after someone reads it. Click a link? Book a call? Reply with a question?

You don't need fancy language for this. A single sentence at the top of your prompt does the job: "You're helping me write a marketing email for my boutique's spring sale, aimed at repeat customers, and I want them to click through to the sale page."

Say Exactly What You Want It to Look Like

Once the AI knows who it's writing for, tell it what the finished piece should look like. Length, tone, structure, and format all matter, and none of them are obvious unless you say them.

Instead of leaving it to guess, spell out:

  • Length: "Keep it under 150 words" or "three short paragraphs."
  • Tone: "Warm and casual, like a text from a friend" versus "professional and formal."
  • Structure: "Start with a subject line, then a two-sentence hook, then the offer."
  • Constraints: "Don't use the word 'exciting.' Don't include emojis."

This is the part people skip most often, and it's the part that saves the most rework. An AI assistant will happily write you five paragraphs when you needed three sentences, simply because you never told it otherwise.

Before and After: One Request, Two Results

Here's the same task, asked two different ways.

Before: "Write me a marketing email for my spring sale."

After: "You're an email marketer for a small boutique clothing store. Write a marketing email announcing our spring sale (20% off, ends April 30) to loyal repeat customers. Keep it under 120 words, warm and friendly in tone, with a clear subject line and one call-to-action button linking to our sale page. Don't use the words 'exciting' or 'don't miss out.'"

The first prompt could describe any business, anywhere, selling anything. The second prompt could only produce an email for this business. That's the entire difference between a generic answer and a useful one — and it took about fifteen extra seconds to type.

Your First Answer Is a Draft, Not the Final Answer

Even with a great prompt, the first response usually isn't perfect. That's normal, and it's not a sign you did something wrong. Treat it the way you'd treat a first draft from a junior employee: give feedback and ask for a revision.

Instead of starting over with a new prompt, just respond to what it gave you:

  • "Make this shorter."
  • "This sounds too formal — make it more like how I actually talk."
  • "Add a line mentioning our free returns policy."
  • "Give me two more versions of the subject line."

The conversation format is the real advantage over a search engine. You're allowed to go back and forth. Most people stop after one try because they assume the first answer is as good as it gets. It almost never is — the second or third round, after a bit of steering, is usually where the useful version shows up.

This same instinct — being specific, giving context, and expecting a back-and-forth — is just as valuable when talking to a human developer about a project. Clear communication gets better results no matter who, or what, is on the other end.

Make It a Habit, Not a One-Time Fix

None of this requires technical knowledge. It requires slowing down for fifteen seconds before you hit enter, and treating your AI assistant like a capable new team member rather than a vending machine. Give it a role, an audience, a purpose, and a picture of the finished product. Then talk back to what it gives you.

Do this a few times and it stops feeling like a technique. It just becomes how you ask.

If your business is leaning on AI more and more, and you want to make sure your tools, processes, and team are actually set up to get the most out of it, that's exactly the kind of conversation I have with clients. Let's talk through your situation.