You typed something into ChatGPT once. Maybe it was "write me a business plan" or "help me with an email." You got back three paragraphs of generic mush that sounded like it came from a corporate training video, and you closed the tab thinking, "Yeah, this isn't for me."

Here's what nobody told you: that wasn't your fault, and it wasn't the AI being useless. It was a bad prompt. Not a stupid prompt — just an incomplete one. AI tools like ChatGPT and Claude aren't mind readers. They respond to what you actually give them, and most people give them almost nothing to work with.

The good news is that writing a good prompt isn't a technical skill. You don't need to learn code words or "prompt engineering" certificates. You need to explain things the same way you'd explain them to a new employee on their first day — someone smart, but who knows nothing about your business, your situation, or what's in your head. Once you start thinking of it that way, everything gets easier.

Why "write me an email" gets you garbage

Imagine you hired a temp worker and the only instruction you gave them was "write me an email." No context, no topic, no idea who it's going to, no idea what tone you want. They'd have to guess at everything, and their guess is going to be average at best. That's exactly what happens when you give a vague prompt to AI. It fills in the blanks with the most generic, safest version of what you might mean — because it has to guess, and generic is the safest guess.

The fix isn't complicated. You just have to fill in the blanks yourself before you hit enter. Who is this for? What's the goal? What tone do you want? What do you already know that the AI doesn't? The more of those blanks you fill, the less guessing the AI has to do, and the better the output gets.

The four things every good prompt needs

You don't need a formula memorized word for word, but every strong prompt tends to cover four things: the situation, the goal, the audience, and the constraints. Situation is the background — what's going on. Goal is what you want to walk away with. Audience is who's going to read or use this. Constraints are the limits — how long, what tone, what to avoid.

Compare these two prompts and notice the difference in what you're handing the AI to work with.

Try this prompt: Write a follow-up email to a customer who hasn't responded to my quote in two weeks. I run a small landscaping business. Keep it short, friendly, not pushy, and give them an easy way to say no if they went with someone else.

That prompt has a situation (quote sent, no response), a goal (get a reply), an audience (a customer, likely busy, maybe forgot), and constraints (short, friendly, not pushy). Compare that to "write a follow-up email" and you can already picture how much better the first one is going to turn out — not because the AI got smarter, but because you gave it something real to build from.

Stop asking for the finished product on the first try

One of the biggest mistakes people make is treating AI like a vending machine — put in one prompt, expect the perfect final answer to fall out. That's not how any of this works, and it's not how you'd work with a real person either. If you asked an employee to draft a proposal, you wouldn't expect it to be client-ready on the first draft. You'd read it, tell them what to fix, and let them revise it.

Treat AI conversations the same way. Your first prompt gets you a rough draft. Your second message is where you actually shape it into something usable. "Make this shorter." "This sounds too formal, make it sound like me." "Cut the third paragraph, it's repeating the first one." Each of those follow-ups costs you ten seconds and makes the output dramatically better. People who give up after one prompt are the same people who think AI doesn't work for them. People who go three or four rounds deep are the ones building entire businesses on this stuff.

Give it your own words as a sample

If you want something to sound like you instead of sounding like every other AI-generated email on the planet, paste in a real example of something you've already written. A text you sent a client. An old estimate. A voicemail transcript, even. This does more for tone-matching than any instruction you could type.

Try this prompt: Here's an email I wrote to a client last month: [paste the email]. Match this tone and style, and write a new email to a different client explaining that their project will be delayed two weeks due to a material shortage.

Now the AI isn't guessing what "sound like me" means. It has your actual voice sitting right there to copy. This trick alone fixes the number one complaint people have about AI writing — that it sounds robotic and fake.

Being specific about the audience changes everything

A lot of people write prompts as if the output is going into a void. But the person reading your email, proposal, or text message is a real human with a specific level of patience and a specific relationship to you. Tell the AI who that person is and the output shifts immediately.

If you're writing to a homeowner who's never worked with a contractor before, say that — the AI will explain things in plainer language and skip the industry shorthand. If you're writing to another business owner who's busy and just wants the bottom line, say that too, and you'll get three sentences instead of three paragraphs. This is one line of extra context that costs you nothing and changes the entire output.

Constraints are your best friend, not a limitation

People often leave out length, tone, and format because they think giving the AI freedom will produce something more creative. In practice, the opposite happens — freedom produces the safest, most average answer possible. Constraints force better thinking, the same way a tight deadline forces better focus.

Try this prompt: I need a text message, not an email, to send a client whose payment is 10 days late. Keep it under three sentences. Friendly but direct. No corporate language, no "please don't hesitate to reach out."

Notice how much that prompt rules out. It tells the AI exactly what not to sound like, which is often more useful than telling it what to sound like. If you've ever gotten an AI response that felt stiff and overly formal, it's usually because nothing told it not to be.

Use AI to fix your own prompt

Here's a move almost nobody uses, and it's one of the most useful ones available. If you're not sure how to ask for something, just ask the AI to help you ask better.

Try this prompt: I want to write a prompt that gets you to help me plan next week's schedule around three job sites, but I'm not sure what details you need from me. Ask me questions until you have enough to give me a solid plan.

This flips the entire interaction. Instead of you guessing what information matters, the AI asks you directly, and you end up with a much stronger result because the right questions got asked in the first place. This is especially useful for anything complicated — business plans, project timelines, budget breakdowns — where you know what you want but don't know how to phrase the ask.

The habit that actually matters

None of this sticks if you only try it once. The people who get real value out of AI aren't the ones who found one magic prompt — they're the ones who use it daily for small, real tasks until explaining themselves clearly becomes automatic. It's the same skill as giving clear instructions to a new hire. The first few times feel clunky. After a couple weeks, it's just how you talk.

Start noticing when an AI response falls flat, and instead of blaming the tool, look at what you left out. Did you say who it's for? Did you say what to avoid? Did you give it an example of your own voice? Nine times out of ten, that's the missing piece, not some secret trick you haven't learned yet.

Do this today

Pick one real task you're avoiding right now — an email you haven't sent, a text you've been putting off, a proposal sitting half-finished. Open ChatGPT or Claude and write a prompt that includes the situation, the goal, who it's for, and one constraint on tone or length. Then read what comes back and give it one round of feedback before you use it. That's the whole skill. You already have it — you just haven't practiced it yet.