You open ChatGPT, type "write a follow-up email to a customer," and get back something that reads like a greeting card wrote it. Too polite, no mention of the actual job, three exclamation marks you'd never use. So you rewrite the whole thing yourself and quietly file AI under "overhyped." The model usually isn't the problem. You handed it a blank slate and expected it to already know your trade, your customer, and how you talk.
Why Vague Prompts Come Back Useless
A model answers the question you typed, not the one in your head. A one-line request forces it to guess everything you left out: what you sell, who you're writing to, your price, your voice. It fills those gaps with the statistical average of the entire internet, and average reads as filler. Every business owner who says "the AI just gives me generic junk" is describing an underspecified prompt.
Here's the test. If you spend more time rewriting the output than you'd have spent writing it from scratch, the prompt is too thin, not the tool broken. A good prompt does one job above all others: it removes the model's need to guess. The more of your real situation you put in, the less it invents, and the closer the first draft lands to something you can actually send.
The Four Parts Every Working Prompt Names
Almost every prompt that returns usable work names four things, roughly in this order. You don't need clever wording. You need these four present.
Say them in plain sentences, not keywords. "You are" for the role, "Here's the situation" for context, "Write me" for the task, "Keep it to" for format. That's the whole trick most of the prompt-course industry sells back to you.
- Role: who the model is for this task. "You're the owner of an Edmonton landscaping company replying to a homeowner." This sets vocabulary and default assumptions.
- Context: the specifics only you know. The customer's name, the job, the price you quoted, the fact that it's late June and you're booked three weeks out.
- Task: the exact deliverable. Not "help with" — "write a two-paragraph email." Name the thing you'll paste somewhere.
- Format: the shape and limits. Length, tone, what to leave out. "Under 120 words, warm but direct, no exclamation marks, no pricing."
A Worked Example: The 8 PM Quote Email
Say it's late, you're at the kitchen table, and you owe a quote follow-up to a customer named Dave who wanted seasonal snow removal for a corner lot in Sherwood Park. The weak prompt is "write a follow-up email about a snow removal quote." You already know what comes back: hollow.
Now the same request with the four parts filled in. "You're the owner of a small snow-removal company in Sherwood Park. Dave asked for a season-long residential contract on a corner lot with a long driveway. I quoted $780 for the season, up front, covering every snowfall over 5 cm through March. He hasn't replied in four days. Write a follow-up email under 130 words. Warm, no pressure, remind him spots fill before the first heavy snow in November. Don't re-explain the price; just nudge him to lock it in."
The second version writes itself in one pass because there's nothing left to invent. That gap — forty minutes of rewriting versus three minutes of light editing — is the entire return on learning to prompt. And it compounds every time you reuse the shape for the next Dave.
Feed It Your Own Words
The fastest way to kill the generic tone is to show the model how you actually write. Paste two or three of your own past emails, an old quote, a policy you already use, and tell it to match that voice. This is the single highest-leverage move most owners skip, and it works because you're giving the model a target instead of asking it to imagine one.
One caution worth the space: don't paste anything you wouldn't want stored on someone else's server. Customer addresses, card numbers, anything covered under Alberta's privacy rules — strip it or swap in placeholders before it goes in the box. The model doesn't need Dave's real house number to write the email.
Fix It, Don't Start Over
When the first draft is close but wrong, resist deleting it and rewriting the prompt from zero. Correct it the way you'd correct a new hire: "Second paragraph's too stiff, loosen it," or "cut the last line, add a sentence about the November cutoff." Short, specific edits get you there faster than a fresh essay of instructions.
One habit to drop: shouting. Piling on "CRITICAL," "NEVER," and "YOU MUST" made older tools behave, but on current models it tends to backfire and produce stiffer, worse output. Calm and specific beats loud and vague. Tell it plainly what to change and it will.
When a Prompt Isn't the Answer Anymore
Here's the honest limit. Prompting is the right tool for one-off writing and thinking work. It's the wrong tool the moment you're retyping the same prompt every day. If you rebuild that snow-quote email from memory a dozen times a week, you don't have a prompting problem, you have a system waiting to be built — one that pulls the price, the job details, and your voice automatically and hands you a finished draft.
That's the line where a text box stops paying off and real software starts. We built Zebra Landscaping a quoting tool that took their estimates from about four hours to under twenty minutes, and we built a system for an Alberta contractor that cut bid assembly from two or three hours down to minutes. Same idea, both times: take the prompt someone was retyping and turn it into something that just runs. Our bid-proposal platform, Bid Pro's, is launching soon on exactly that principle.
So learn the four parts. They'll make this week's emails and quotes noticeably better, tonight. Then pay attention to which prompt you keep typing over and over, because that one is telling you what to automate next.