If ChatGPT keeps handing you answers that sound like a brochure, the model is not the problem. It is doing exactly what you asked — you just asked for very little.
A good prompt is not clever wording. It is complete information. Here is how to write one every time.
The 5 parts every good prompt has
- Role — who the AI should act as.
- Goal — what the output is for, in one sentence.
- Context — who your customer is, what you sell, what you already tried.
- Format — bullets, table, 200 words, three options, plain English.
- Constraints — tone, reading level, what to avoid, what to never claim.
Why vague prompts produce vague answers
When you type "write me a marketing email," the model has no idea who the reader is, what you sell, what tone fits, or what a win looks like. So it averages everything it has ever seen about marketing emails. That average is exactly what generic output feels like.
Give it three specific details and the average collapses into something that could only be about your business.
Before and after
Before: "Write a marketing email for my product."
After: "Act as an email copywriter. Write a 150-word email to past customers of a home cleaning service in Dallas who have not booked in six months. Goal: get them to rebook. Mention that spring is the busiest season so slots go fast. Warm and direct, no exclamation points, no discount offers."
The second prompt names the reader, the situation, the goal, the length and the boundaries. That is the whole skill.
Five mistakes that quietly ruin prompts
1. Asking for everything at once. A prompt that asks for strategy, copy, and a posting schedule gets shallow versions of all three. Split it up.
2. Skipping the audience. "Small business owners" is a category. "Solo house painters who bid jobs from their truck" is a person.
3. No format. If you do not say how you want the answer, you get paragraphs. Ask for a table, five bullets, three options, or 100 words.
4. Accepting the first draft. The first answer is a starting point. Reply with what to change, not a brand new prompt.
5. No constraints. Constraints are where voice lives: no jargon, no hype, no emojis, 8th-grade reading level, never promise results.
Give it your raw material
The single biggest upgrade most people can make is pasting in real information: your actual service list, your actual customer email, your actual notes. The model cannot invent facts about your business, and when it tries, you get the made-up-sounding output everyone complains about.
Feed it truth, and ask it to shape that truth.
Ask for options, then pick
Instead of "write a headline," ask for five headlines in different angles: practical, curious, blunt, story-based, and objection-first. You will learn more about what you actually want by reacting to options than by describing it in advance.
Refine in small moves
Useful follow-ups you can reuse on any output:
- "Same structure, but make every example specific to [industry]."
- "Cut this by 30% without losing meaning."
- "Rewrite it the way you would explain it to a customer standing in front of you."
- "What is the weakest part of this? Fix only that."
A template to keep
Act as [role].
Goal: [what this output is for].
Audience: [who reads it, what they already know].
Context: [facts about your business, what you tried].
Format: [length, structure].
Constraints: [tone, what to avoid, what never to claim].
Fill it in once for a common task and save it. You now have a reusable prompt for every future version of that task — which is how people who "seem good at AI" got that way.