How to Talk to AI: Prompt Engineering for Non-Developers

You’ve been asking AI the same question three different ways, getting three different answers, and wondering why. You’re not bad at this. You’re just writing prompts like you’re texting: casual, half-formed, hoping the AI will guess what you mean.

It won’t. The good news is that prompting is a learnable skill, and it has nothing to do with code. It’s about being clear — and it’s fast becoming part of the basic AI literacy employers screen for.

Why this matters

Gallup’s 2026 workplace survey shows AI use rising fast — leading in tech, but spreading into marketing, ops, HR, and most knowledge work roles. Most readers here are non-technical, and most are leaving output quality on the table by not telling AI what they actually need.

The gap between people who use these tools well and people who don’t is already widening into a real career divide, and prompting is where it starts.

The pattern is consistent across anyone who uses AI regularly: clearer, more specific prompts produce better outputs. Not marginally better, noticeably better. The difference between a vague prompt and a structured one is often the difference between output you can use and output you have to rewrite from scratch.

The biggest win is speed. Instead of six iterations to get something usable, you get it right in two.

The frameworks that work

Forget complexity. The best frameworks for non-developers are simple checklists that make you think through what you actually want.

COSTAR is the workhorse. It stands for Context, Objective, Style, Tone, Audience, Response. Developed by Sheila Teo, who won GovTech Singapore’s first GPT-4 prompt engineering competition with it, it works across every job type.

Say you’re in HR and need to draft a rejection email:

• Context: “I’m sending a rejection to a candidate who made it to our final round. The role was senior engineer. The company is a fast-growing fintech startup.”
• Objective: “Draft a professional, warm rejection that explains why we didn’t move forward without crushing their confidence.”
• Style: “Direct, honest, brief.”
• Tone: “Encouraging. Respectful.”
• Audience: “A smart person who didn’t get the job.”
• Response: “Give me an email I can send in under 2 minutes.”

You’ll get something thoughtful and ready to send. Without COSTAR, you’d probably write “write me a rejection email” and get generic corporate-speak.

BAB (Before, After, Bridge) is built for strategy and change communication. It asks: what’s the current state, what’s the desired future, and how do we get there?

Marketing example — you’re pitching a shift to AI-assisted content creation to your team:

• Before: “We’re spending 40 hours/week on first drafts. The process is bottlenecked on writer availability. Ideas take weeks to ship.”
• After: “We’re using AI to generate first drafts in 2 hours. Writers spend their time editing and refining, not staring at blank pages. We publish 3x more content at higher quality.”
• Bridge: “We introduce AI tools, set clear guardrails, run a 2-week pilot, then expand.”

That’s a compelling narrative. Try writing it without the framework and you’ll probably ramble.

CARE (Context, Action, Result, Example) is your friend for narrative-heavy work: case studies, project retrospectives, pitch decks.

Operations example — you need a case study of how you streamlined vendor onboarding:

• Context: “We receive 20-30 vendor applications per month. Onboarding took 3 weeks average. Bottleneck: manual document review and email follow-up.”
• Action: “Implemented AI-assisted document review to flag missing items. Set up automated templates for follow-up.”
• Result: “Onboarding now takes 4 days. 92% of vendors submit complete applications on first try. Team reclaims 12 hours/week.”
• Example: “Last month, we onboarded Acme Corp in 3 days vs. 21 days previously.”

The structure forces you to think like a storyteller instead of a list-maker, and people remember stories.

The one rule that matters most

All three frameworks follow the same principle: specificity beats cleverness.

You don’t need poetic language. You need context. The AI can’t read your mind. It knows your company’s culture if you tell it. It knows your audience if you describe them. It knows your constraints if you name them.

The most common mistake isn’t poor wording, it’s missing context. “Summarize this meeting” is vague. “Summarize this 45-minute meeting about Q3 hiring plans for our engineering team. I need 3 bullet points of decisions made and action items with owners” is sharp.

Specificity saves you time. Anyone who’s used AI regularly will tell you: the more precise the constraint, the less cleanup you do on the other side.

Common mistakes to skip

Skipping examples. If you want output formatted a certain way, show the AI what you mean. “Format as: Name | Role | Key Achievement” is clearer than “make it easy to scan.”

Assuming context. You know your company and your role. The AI doesn’t. Spend 15 seconds explaining what industry you’re in, what your team does, what success looks like.

Asking for everything at once. If you need a 10-point analysis, ask for it in stages. First pass: rough outline. Second pass: develop the risky sections. You’ll catch more and learn what the AI does well.

Using “best” or “innovative” or “game-changing.” These words sound impressive but give no direction. Use specifics: “better for remote teams,” “under 200 words,” “suitable for non-technical readers.”

When the output improves, that’s not the AI getting better. That’s you getting clearer. And clarity compounds: it makes you sharper in meetings, in briefs, and in every request you make of another human, too.