So basically, here’s the secret nobody tells you about AI: most bad AI answers come from bad prompts.
You type something vague, the AI guesses what you meant, and the answer comes back… fine. Generic. The kind of thing that reads okay for ten seconds and then helps you zero. Sound familiar?
I spend all day with these tools — ChatGPT, Claude, Gemini, the whole lineup — and here’s what I’ve learned. The difference between a “meh” answer and a “holy cow, that’s exactly what I needed” answer usually isn’t a better AI. It’s a better prompt. And better prompts aren’t some mystery. They follow patterns.
I’ve got five of them, and now they’re on a sticky note next to my monitor. Roxie says calling them “patterns” is generous. She calls them “the stuff people skip because they’re in a hurry.” She’s not wrong.
“90%” is shorthand, not a lab-tested number. What I mean is: almost every bad AI answer I’ve seen was a prompt problem, not an AI problem. Your mileage will vary, but these five patterns will genuinely move the needle on the output you get.
1. Give the AI a Job Title
This is the single fastest upgrade you can make to any prompt.
Before: “Write a product description for my widget.”
The AI has no idea what a product description even means in your world. Amazon listing? Landing page? One-pager for a sales meeting? It’ll pick something, and it’ll probably pick wrong.
After: “Act as a direct-response copywriter for small businesses. Write a 100-word product description for my widget, aimed at busy homeowners. Focus on the one benefit that saves them the most time.”
See the difference? You gave the AI a job title, a word count, an audience, and a priority. That’s four guardrails in one sentence, and the answer comes back shaped like actual copy instead of a generic blob.
Actually, this one matters more than people think. An AI with no assigned role defaults to “helpful internet encyclopedia.” A role gives it a personality, a standard, and a point of view — the three things that separate useful writing from filler.
Try it with anything: “Act as a friendly math tutor,” “Act as a grumpy-but-fair code reviewer,” “Act as a real estate copywriter who never uses the word ‘cozy.'” The more specific the role, the better it works. Have fun with it.
2. Show One Example Instead of Writing Ten Lines of Instructions
Here’s where it gets interesting. If you want the AI to write in a certain style, you could spend three paragraphs describing that style…
…or you could just show it one example.
Before: “Write a social media caption in a fun, punchy, slightly edgy tone with emojis but not too many, aimed at millennials, with a hook in the first line and a call to action at the end.”
That’s a lot of adjectives doing very little work. “Punchy” means something different to everyone — including the AI.
After: “Here’s the vibe I’m going for: [paste one caption you like]. Write three captions in this exact style for my product.”
One example is worth a thousand adjectives. The AI reads your example, picks up the rhythm, the length, the attitude, the whole thing — and it doesn’t have to guess what “punchy” means to you.
This works for everything, not just writing. Want a specific spreadsheet layout? Describe one row. Want a certain kind of email? Paste one you loved. You’ve probably been wondering why AI keeps missing your style — this is why. Stop describing. Start showing.
3. Describe the Output, Not Just the Input
Most prompts describe the task and then stop. The AI does the task… in whatever shape it feels like. That’s how you get a 900-word essay when you wanted a checklist.
Before: “Summarize this meeting transcript.”
Okay, and then what — a paragraph? A list? For whom? The AI picks, and you get to re-ask. Congrats, you just paid for the same work twice.
After: “Summarize this meeting transcript in three sections: Decisions Made (bulleted), Open Questions (bulleted), and Next Steps (with an owner and a deadline for each). Keep the whole thing under 250 words.”
You just told the AI exactly what the finished product looks like. Format, length, sections — done. The answer comes back ready to paste into Slack instead of ready to be reworked.
The rule is simple: if you know what the output should look like, say so. The AI can’t read your mind, but it follows directions really well when you actually give them. Here’s my bold opinion: most “AI gave me the wrong format” complaints would vanish if people wrote one extra sentence about the output.
4. Break the Big Ask Into Small Ones
This is the one that changed how I work with AI more than any other.
Bad prompt: “Write me a complete marketing plan for my coffee shop.”
That’s not a prompt. That’s a wish. A marketing plan has a dozen moving parts — audience, budget, channels, timeline — and when you ask for all of it at once, the AI gives you a little of everything and not enough of anything.
Better: split it up.
First prompt: “Help me define my coffee shop’s target customer in three personas, one paragraph each.”
Then: “For persona 1, list the five marketing channels most likely to reach them, ranked by cost-effectiveness.”
Then: “Draft a 30-day content calendar for the top two channels.”
Same project. Totally different results. Each answer builds on the last, and you get to course-correct between steps instead of getting one giant document that’s 70% right and 100% annoying to fix.
Think of it this way: you wouldn’t hand a freelancer a one-line brief for a whole marketing plan and expect magic. You’d walk through it with them, piece by piece. The AI works the same way — it just happens to be very fast at each piece.
5. Ask the AI to Argue With Itself
This is my favorite one, and it’s the move most people never try.
You get an answer back, and it’s fine. But fine isn’t what you’re after. So instead of settling, you say: “Now review your own answer. Find the three weakest points and tell me what’s wrong with them. Then fix them.”
The AI re-reads its own work like an editor. It catches the vague paragraph, the missing detail, the assumption it made that you never approved. And then it fixes all of it.
First prompt: “Write a 150-word intro to my newsletter about AI tools for real estate agents.”
Follow-up: “Now act as a skeptical reader. What’s unclear, boring, or unconvincing about that intro? List the problems, then rewrite it fixing all of them.”
The second version is almost always better. Not because the AI is smarter the second time — but because editing is a different skill than drafting, and now you’ve made it do both.
Roxie calls this “trusting the AI to check its own homework,” and she says you should still double-check anything important yourself. She’s right. The critique pass catches style and structure problems, not factual ones. If the answer has numbers, dates, or claims in it, verifying those is still on you.
The Quick-Reference Card
So there they are. Five patterns, one job each:
| Pattern | When to use it | Copy-paste starter |
|---|---|---|
| Give it a job title | The answer feels generic | “Act as a [role]…” |
| Show one example | Style matters and words fail | “Here’s the vibe: [example]. Match it.” |
| Describe the output | You keep getting the wrong format | “Give me [format], [length], [sections]…” |
| Break it into steps | The task is big and the answer is shallow | Ask for step one, then build |
| Critique pass | The answer is “fine” but not great | “Find the three weakest points and fix them.” |
Print it. Stick it to your monitor. Steal it. I did.
Your Turn
If you’re just starting out, go read my prompting guide first — it covers the foundation all of this builds on (How to Prompt AI Tools So They Actually Give You What You Want). These five patterns are the next layer up: the moves that take your prompts from “okay” to “how did it do that.”
And here’s the thing nobody wants to admit: the AI isn’t the bottleneck anymore. You are. Not in a mean way — in an exciting way. The people getting the best results from these tools aren’t prompt engineers with certifications. They’re just the ones who spend an extra 30 seconds thinking about what they’re actually asking for.
You’ve got the patterns now. Go use them.