A product launch email. Nine words into the prompt box: "Write a marketing email for my product launch." What came back was a perfectly structured email, subject line, opening hook, body copy, call to action. Beautifully formatted. Completely useless. No product name. No audience. Benefit language so vague it could have applied to a kitchen appliance or a SaaS tool or a children's book.
I closed the tab. Told myself the tool was overhyped. Then I rewrote the prompt, properly this time, and got one of the best first-draft emails I've ever seen. Same tool. Same day. The only thing that changed was me.
That gap, between nine words and a real brief, is the whole lesson.
The Prompt I Actually Sent (And Why It Failed)
"Write a marketing email for my product launch" is not a prompt. It's a search query with delusions of grandeur.
Compare it to what I'd give a real copywriter: the product name, what it does, who buys it, what they're worried about before they buy, the tone of the brand, what I want them to do at the end of the email.
I gave ChatGPT none of that. I gave it nine words and expected it to read my mind.
The model doesn't have access to my product, my audience, or my brand. So it did the only thing it could: it generated a plausible-looking email that fit the shape of what I asked for.
This is the core mistake most people make with ChatGPT in the first week. They treat it like Google, a place where you type a few words and the right thing appears.
Why ChatGPT Mirrors Your Vagueness Back at You
ChatGPT is a language model. Its job is to generate plausible, coherent, helpful-sounding text based on the input you give it.
When you're vague, it doesn't stop and say "I need more information." It fills the gap with probability, what would a marketing email typically look like? It generates that. Fluently. Confidently. Without any indication that it's working with basically nothing.
The practical implication: you can't tell a bad prompt from a good one by looking at the output's formatting. A beautifully structured, well-written response can be completely off-target. Fluency is not accuracy. Confidence is not correctness.
Until you understand this, you'll keep blaming the tool.
Prompts vs. Briefs: The Most Important Distinction
Designers don't work from search queries. Developers don't work from search queries. Good writers don't work from search queries.
They work from briefs.
A brief has: a goal, an audience, constraints, tone, format, and context. When I rewrote my email prompt as a brief, the output transformed. Here's roughly what I included:
- Product: a PDF guide for non-technical professionals who want to understand AI
- Audience: people in their 40s and 50s who feel left behind by the AI conversation
- Tone: direct, warm, not technical, not hyped
- Goal: get them to click through to the sales page
- Format: short, three paragraphs max, casual subject line, single CTA
Same tool, same underlying model, same day. The email I got back was specific, in-voice, and actually useful as a starting draft.
Think of every prompt you write as a brief. Here's the template I use, copy it, fill in the brackets, and send it:
Product: [what you're launching or working on]
Audience: [who they are, and what they're worried about before they buy]
Tone: [what to avoid, formal, salesy, jargon, etc.]
Goal: [what you want them to do at the end]
Format: [length, structure, any constraints]
Thirty seconds to fill that in. It will outperform five minutes of back-and-forth on a bad prompt every time.
The Hallucination Problem: When Fluency Becomes Dangerous
There is one area where the prompt quality problem stops being about output quality and starts being about accuracy.
ChatGPT will sometimes make things up. Confidently. With footnotes, if you ask for them.
I asked it once to summarize some research on a particular topic. Several of the citations it included were fabricated. Not paraphrased or misattributed. Invented. Real journal name, plausible-sounding paper title, author names that don't exist.
This is called hallucination, and it's an inherent feature of how language models work. They predict what a correct-sounding response looks like. Sometimes what a correct-sounding response looks like is a made-up citation.
My rule: if I'm using ChatGPT to draft, explore, brainstorm, or rewrite, I don't verify heavily. If I'm using it to cite, fact-check, or research, I verify everything.
What ChatGPT Is Good For (And What to Use Instead)
ChatGPT is genuinely excellent at a specific category of tasks: drafting, restructuring, explaining, brainstorming, and adapting tone. Give it a rough draft and ask it to tighten the structure, good. Give it a complex concept and ask it to explain it in plain English, very good.
What it's not the right tool for: live research, anything where accuracy is load-bearing without verification, and tasks that require deep nuanced judgment without significant guidance from you.
For research where you need real sources, I use Perplexity. It's built to retrieve and cite. For drafting where I need careful reasoning, I tend to use Claude.
Using the right tool isn't betrayal. It's competence.
What I Do Differently Now
Six specific habits that changed my results:
I write the context before I write the request. First line of any prompt: who I am, what I'm working on, and what role I want ChatGPT to play.
I specify the audience explicitly. Not "write this for readers." Write it for someone who has never used AI tools but is curious and slightly skeptical.
I include format constraints. Length, structure, tone, what to avoid. Constraints aren't limitations on creativity, they're the shape the work has to fit.
I treat the first output as a draft, not a result. Almost nothing I get back in round one is publish-ready. That's fine.
I don't ask it to research; I ask it to reason. "What are the facts about X?" is a weak prompt. "Here are the facts about X, help me think through the implications" is better.
I verify anything specific. Statistics, names, dates, citations. Every time.
The Gap Between Bad Results and Good Ones Is Prompt Quality
Most people who say ChatGPT is overhyped or useless are describing their own prompt quality, not the tool's capability.
The difference between beginners and experienced users isn't intelligence, it's knowing the tool's actual behavior.
These six habits cover the prompt quality problem. They don't cover what comes after: how to route tasks between tools, knowing when to reach for Perplexity instead of ChatGPT, and when Claude is the better call, and how to build these habits into a brief template you actually run every time rather than carrying them in your head.
The guide below closes that gap. It takes the brief format from this article and gives you the prompts for every task that eats your time, organized by job, ready to run. The door is open.
Go deeper
ChatGPT for Business Owners
The brief format this article covers, plus the prompts for every task that eats your time: proposals, discovery calls, SOPs, hiring docs. Organized by job, ready to run.
Get the guide on Amazon, $9.99If you'd rather hire the employee than read the guide, we built that. Meet the team →