ChatGPT Isn’t Broken—You Just Need to Prompt Like a Teacher

Do you know who are the most skilled, most capable people when it comes to getting a good output from AI?

Elementary school teachers.

If you’ve spent more than ten minutes with an LLM, you know one important truth: It takes directions like a 10-year-old. That’s why I fire it almost daily. And it’s why people throw up their arms in frustration and walk away—“it just doesn’t understand what I want!”

But actually, it can. It can give you great outputs…IF you know how to break your ChatGPT prompt down into its simplest parts. And no one does this better than elementary teachers.

Think about how a teacher gives directions for a classroom project. They don’t say, “Go make a diorama about the rainforest.” They break it into steps: pick your animals, choose a shoebox, cut construction paper for trees, draw rivers, glue everything in. They anticipate where a student might get lost and build clarity into the instructions.

AI works the same way.

Consider this scenario: You’re writing an article with ChatGPT and it comes out bland. You ask it to “make it funnier.” Just like a 10-year-old, ChatGPT grabs the lowest-hanging fruit: cheesy one-liners, puns, maybe even a “your mom” joke. You don’t like it. So you push further: “Rewrite it like Jimmy Fallon or Sarah Silverman.” Suddenly you’re holding a comedy set, not an article.

Is that ChatGPT’s fault? Or is it the 10-year-old’s fault for not magically knowing what you meant? Of course not. You gave it vague directions that could mean a hundred different things.

Now you start thinking like a teacher. How would you explain this to someone with little life experience and shaky communication instincts? Instead of a blanket command, you spell it out in your prompt. Something like: “Keep the professionalism of the article for about 80% of it. Use the remaining 20% to add short, funny moments. Pull from things like movie quotes, Gen Alpha slang, or creative word mashups like ‘hangry.’ Blend it so the humor flows with the tone, not against it.”

Now your AI has something it can work with. It won’t be perfect, but it’ll be much closer to what you wanted.

This applies far beyond humor. Say you want a concise summary of a 20-page report. If you ask for “a summary,” you’ll probably get something too long, too short, or too shallow.

Instead, try breaking it down: “Give me three bullet points with key findings, each no more than twelve words, and one sentence at the end that connects the findings to business impact.” That’s the diorama version of the task—the broken-down parts that tell AI exactly how to build what you need.

Or imagine you want help brainstorming social media posts. If your ChatGPT prompt just says, “give me ideas,” you’ll get a random list. But if you explain: “I need five posts for LinkedIn, each written in a confident but casual tone. They should be under 150 words, each focused on a different benefit of our product, and one should include a relevant statistic,” you’ve given AI the scaffolding it needs to deliver something useful.

Does it sound like a lot of work? It is, at first. But just like a teacher gets faster at anticipating what their students will miss, you’ll get faster at anticipating what the AI needs spelled out. And over time, you’ll build a bank of instructions you can reuse and adapt.

Here are a few examples of vague prompts people often use, along with better versions that actually get results:

Vague Prompt	Better Prompt
“Make this more engaging.”	“Pull out three surprising or counterintuitive insights from the text and highlight why they matter to the reader. End each with a one-sentence takeaway that ties directly to something they do every day.”
“Give me a summary.”	“Create five bullet points that highlight the findings most relevant to a marketing director. Show how each finding connects directly to revenue growth, customer acquisition, or cost savings.”
“Write social media posts.”	“Draft five LinkedIn post ideas that surprise my audience by pointing out overlooked, tiny details in daily marketing work—like the hidden cost of broken UTMs, the impact of misaligned campaign dates, or the long-term affects of mislabeled Salesforce fields.”
“Make it sound smarter.”	“Use more precise terminology, establish a unifying framework that ties the points together, and link each section back to strategic implications so the reader feels guided by expertise, not just informed.”
“Brainstorm blog topics.”	“Give me ten blog ideas for IT leaders at mid-sized companies who feel buried in manual reporting tasks. Each topic should highlight micro-frustrations they deal with constantly, like exporting CSVs every Monday or chasing down missing time stamps.”
“Write an email campaign.”	“Draft three email sequences aimed at HR directors evaluating training platforms. Each sequence should spotlight one measurable business outcome (employee retention, faster onboarding, reduced compliance risk) with supporting evidence that draws a direct line to their day-to-day lives.”
“Make it more persuasive.”	“Reframe the argument by focusing on what the reader stands to lose if they ignore this issue. Use small, concrete risks they’ll recognize instantly—like compliance fines from one missing training record, or a single missed escalation that damages client trust.”

So the next time you’re struggling with AI prompts, imagine yourself in a classroom. Imagine you’re explaining it to a 10-year-old who doesn’t yet have context, instincts, or shortcuts. That’s the level of clarity AI thrives on.

And if you’re really stuck? Call up a primary school teacher friend. They’ll know exactly what to do.

Which brings us to a final point—and maybe the most uncomfortable one. The hardest part of this isn’t the AI. It’s you. Breaking your request into pieces forces you to decide what you actually want. And that is where a lot of marketers fall apart.

So yes, it can feel frustrating. But that frustration isn’t because the AI is broken. It’s because you don't actually know what you want. And I hate to say it, but…that one’s on you.