There are about 10,000 “top AI prompts” lists on the internet right now. I know because I’ve read an embarrassing number of them.
Here’s what they all get wrong: they give you fish instead of teaching you to fish. You copy a prompt, it works once, then you face a slightly different situation and you’re stuck again. Back to Google. Back to the list.
What you actually need isn’t 500 prompts. It’s 7 patterns. Patterns you can adapt to any situation, any tool, any task. Once you learn them, you’ll never need a “top prompts” list again.
Why Prompts Fail (And It’s Not the AI’s Fault)
Before the patterns, you need to understand why most prompts don’t work.
When you type “write me an email” into ChatGPT, you’re giving the AI almost nothing to work with. It doesn’t know who the email is for, what tone you want, how long it should be, what you’re trying to accomplish, or what “good” looks like to you.
So it guesses. And it guesses generically — because generic is the statistically safest prediction when you have no context.
The fix isn’t a better prompt template. It’s understanding what AI needs from you to produce something specific. If you’re curious about why AI works this way, we explain the mechanics in How AI Actually Learns.
Every good prompt gives the AI three things:
- Context — who you are, what situation you’re in
- Instructions — what you want, specifically
- Constraints — what “good” looks like, what to avoid
The 7 patterns below are different ways of providing these three things. Some work better for certain tasks than others. All of them work across ChatGPT, Claude, Gemini, and basically any AI tool that takes text input.
Pattern 1: The Role Assignment
What it does: Tells AI who to be before asking it to do anything.
The pattern:
You are a [specific role] with [specific experience]. Your style is [characteristics]. [Now here’s what I need.]
Example:
You are a senior marketing copywriter who specializes in B2B SaaS. Your style is clear, conversational, and avoids buzzwords. Write a LinkedIn post announcing our new feature that saves teams 5 hours per week on reporting.
Why it works: AI’s predictions change dramatically based on “who” it thinks it’s being. A response from a “senior marketing copywriter” is structurally different from one by a “helpful assistant.” You’re steering the entire prediction space with one sentence.
Use this for: Writing tasks, analysis, advice, reviews — anything where perspective matters.
Pattern 2: The Example Shot
What it does: Shows AI 2-3 examples of what you want instead of describing it.
The pattern:
Here are examples of [what I want]: Example 1: [paste it] Example 2: [paste it] Now create one for [new situation].
Example:
Here are examples of our weekly client update emails:
Example 1: “Hi Sarah, Quick update on Project Atlas: we finished the homepage redesign (3 days ahead), started the API integration, and flagged a potential delay on the mobile build. Happy to jump on a call if you want details. —Tom”
Example 2: “Hi Marcus, Project Beacon update: content audit is complete (47 pages reviewed), we identified 12 pages to consolidate, and the new sitemap is ready for your review. Link attached. —Tom”
Now write one for a client named Lisa, project called Horizon, where we completed user research (interviewed 15 users), started wireframes, and discovered the client’s existing analytics are broken.
Why it works: Examples are worth more than pages of instructions. The AI picks up on patterns in your examples — length, tone, structure, what you include and skip — without you having to articulate those patterns explicitly.
Use this for: Anything where you have existing examples of “good” — emails, reports, social posts, code comments.
Pattern 3: The Chain of Thought
What it does: Makes AI show its reasoning step by step, so you can catch mistakes.
The pattern:
[Your question or task]. Think through this step by step. Show your reasoning before giving your final answer.
Example:
Our Q1 revenue was $2.3M (up 15% from Q4), but our customer count dropped from 450 to 420. Marketing spend increased by 22%. Should we be worried? Think through this step by step. Show your reasoning before your conclusion.
Why it works: When you make AI “think out loud,” it catches its own errors. Without this, it jumps to a conclusion. With it, it reasons through intermediate steps — and you can see where the logic holds or breaks.
Use this for: Analysis, decisions, problem-solving, math, anything where the answer depends on reasoning through multiple factors.
Pattern 4: The Constraint Box
What it does: Tells AI what NOT to do, which is often more useful than telling it what to do.
The pattern:
[Task]. Requirements: [what to include]. Constraints: [what to avoid, format limits, tone rules].
Example:
Write a product description for our new project management tool. Requirements: mention the 3 key features (AI task assignment, deadline prediction, team workload balancing), include one customer quote, end with a clear CTA. Constraints: under 150 words, no buzzwords like “revolutionary” or “game-changing”, no exclamation points, professional but warm tone.
Why it works: AI defaults to generic, positive, slightly overwritten content. Constraints force it away from defaults. “No buzzwords” alone improves output quality dramatically.
Use this for: Any task where you’ve been disappointed by AI output before. Add the constraints that address what was wrong last time.
Pattern 5: The Persona Interview
What it does: Makes AI pretend to be your target audience so you can test ideas.
The pattern:
Act as [specific persona with details]. I’m going to pitch you [something]. React honestly — tell me what concerns you, what excites you, and what you’d ask before saying yes.
Example:
Act as a 42-year-old marketing director at a mid-size insurance company. You’ve been in your role for 6 years. Your team is 5 people. Your biggest frustration is reporting — it takes 2 days every month. You’re skeptical of new tools because the last three your company bought went unused. I’m going to pitch you our reporting automation tool.
Why it works: The more specific the persona, the more realistic the reactions. Vague personas give vague feedback. Detailed personas push back on the specific things a real person in that role would push back on.
Use this for: Testing pitches, messaging, product ideas, email drafts before sending them to real people.
Pattern 6: The Iterative Refinement
What it does: Uses a multi-step conversation instead of trying to get everything in one prompt.
The pattern:
Step 1: [rough request] Step 2: “Good, but change [specific thing]” Step 3: “Almost — make it [more specific adjustment]” Step 4: “Perfect. Now do the same for [variation]”
Example conversation:
You: Write a cold email to a VP of Engineering about our developer productivity tool.
AI: [produces draft]
You: Good structure, but the tone is too salesy. Make it sound like a peer sharing something useful, not a vendor pitching. Cut the first paragraph — start with the value.
AI: [produces better draft]
You: Better. But the CTA feels pushy. Change “book a demo” to “happy to show you in 15 minutes if it’s relevant.” Also the subject line is boring — give me 5 options.
Why it works: Nobody gets a perfect result on the first try. But most people give up after one prompt. The people who get great output treat AI like a conversation — iterating 3-5 times is normal and expected. We cover this in depth: The Follow-Up Trick.
Use this for: Everything. Seriously. Iteration is the single biggest gap between people who think AI is “meh” and people who get incredible results.
Pattern 7: The Format Specification
What it does: Tells AI exactly what shape the output should take.
The pattern:
[Task]. Output format: [specific structure with headers, bullets, tables, or code blocks].
Example:
Analyze the pros and cons of switching from Slack to Microsoft Teams for a 50-person company. Output format:
Recommendation (1 sentence)
Key Factors
Factor Slack Teams Winner Migration Risks (bullet list, max 5)
Estimated Timeline
What I’d Watch Out For (honest concerns)
Why it works: AI defaults to paragraphs. But most people need structured output — tables, bullet lists, headers they can scan. Specifying the format prevents the “wall of text” problem and makes the output immediately usable.
Use this for: Reports, comparisons, analysis, meeting summaries, decision briefs — anything that needs to be scannable.
How to Practice
Don’t try to memorize all seven. Instead:
This week: Pick Pattern 1 (Role) and Pattern 4 (Constraints) and use them in every AI interaction. They’ll improve your output the most with the least effort.
Next week: Add Pattern 6 (Iteration). Stop accepting the first draft. Push back 2-3 times on every response.
After that: Try the rest as situations come up. You’ll naturally reach for the right pattern once you’ve used each one a few times.
If you want a structured path through these techniques:
And if you want the advanced techniques after that — Advanced Prompts goes deeper into chaining, multi-step workflows, and system prompts.
The bigger picture: prompts are just one of 5 AI skills that separate the top 5% of workers. Once you’re comfortable with these patterns, the next skill to learn is applying them to real work — writing, analysis, and automation.
Frequently Asked Questions
Do these patterns work for all AI tools?
Yes. ChatGPT, Claude, Gemini, Copilot, Llama, Mistral — they all respond to the same principles because they’re all based on similar architectures. Some tools handle certain patterns better (Claude is particularly good with long context and examples), but the patterns themselves are universal.
What’s the most important pattern if I only learn one?
Pattern 6 (Iterative Refinement). Most people send one prompt and accept whatever comes back. The simple habit of saying “good, but change X” two or three times will improve your results more than any other technique.
How long should my prompts be?
As long as they need to be. A prompt with good context, clear instructions, and specific constraints might be 100-200 words. That’s fine. AI handles long prompts well. The problem is never “too long” — it’s “too vague.”
Should I use the same prompt structure every time?
No. Different tasks need different patterns. A quick email needs a Role + Constraints. A complex analysis needs Chain of Thought + Format Specification. Match the pattern to the task.
This post is part of our Learn AI series — a practical guide to AI skills for non-technical workers.
