All articles

Prompt Engineering

How to Write Better ChatGPT Prompts (With Real Examples)

Most people use ChatGPT like a search engine and get mediocre results. Here's how to write prompts that actually work — with practical examples you can use today.

January 15, 20257 min readBy Rebus Labs

---

If you've typed a question into ChatGPT and gotten a generic, half-useful answer, you're not alone. Most people treat AI like a search engine — a one-shot query that either works or doesn't. But that's not how it works best.

ChatGPT responds to the quality of your input. The clearer, more specific, and more structured your prompt, the better the output. This guide covers the techniques that make the biggest difference.

Why Most Prompts Fail

The most common mistake is being too vague. "Write me a marketing email" will produce a generic email. "Write a short marketing email for a B2B SaaS tool that helps HR teams automate onboarding, targeting a VP of People at a 200-person company, with a casual but professional tone and a call-to-action to book a 20-minute demo" will produce something you can actually use.

The other mistake is treating a single prompt as a complete interaction. ChatGPT works best as a back-and-forth conversation, not a vending machine.

The Four Parts of a Good Prompt

Every strong prompt has some combination of these four elements.

Role — Tell it who it is. Giving ChatGPT a role activates a different mode of response. "You are an experienced HR business partner with 10 years in tech companies" produces a fundamentally different output than a default response.

Context — Give it background. What's the situation? Who's the audience? What's already been tried? The more relevant context you give, the less it has to guess.

Task — Be specific about what you want. Not "write something" but "write a 3-paragraph email" or "create a bulleted list of 5 options" or "summarize this in 2 sentences."

Format — Tell it how to structure the output. Specify length, tone, and constraints: "in a table", "as a numbered list", "in plain language", "in under 150 words."

5 Prompt Techniques That Work

1. Use the "Act as" prefix

Act as a career coach helping a mid-level software engineer prepare for a senior engineering interview. Give me 5 behavioral interview questions I should be ready for, with a brief note on what each question is really testing.

This produces more focused, expert-quality output than "give me interview questions."

2. Give it a before/after example

I want to rewrite this job description to be more inclusive. Here's the original: [paste text]. Rewrite it keeping the core requirements but removing unnecessary credential gatekeeping and softening the language.

Examples are the fastest way to align the AI's output with your expectation.

3. Ask it to think step by step

Walk me through step by step how you would analyze a 20% drop in employee engagement scores at a 500-person company. What data would you look at first? What questions would you ask? What hypotheses would you test?

"Step by step" triggers more methodical, structured reasoning. It works especially well for analysis tasks.

4. Constrain the output

Give me exactly 3 options. No more. For each, write a single sentence summarizing the approach and one potential downside.

Constraints prevent AI from padding output with filler. Be specific: "exactly 3", "under 100 words", "no bullet points."

5. Iterate, don't restart

If the first response isn't quite right, don't start over. Refine it:

That's good but too formal. Rewrite the second paragraph in a more conversational tone, as if explaining it to a colleague over coffee.

This is where most of the real value in ChatGPT comes from — the iteration, not the first shot.

Common Mistakes to Avoid

Being too open-ended. "Help me with my business" gives it nothing to work with. "Help me write a 60-second pitch for a bootstrapped SaaS tool that reduces time-to-hire for startups" is workable.

Accepting the first response. The first output is rarely the best. A quick follow-up — "make it shorter", "give me 3 alternatives", "what are the downsides of this approach" — almost always produces something more useful.

Not specifying the audience. The same content written for a technical audience looks completely different from content written for executives. Always specify who the final reader is.

Ignoring the system prompt. In ChatGPT, you can set a custom system prompt that persists across your session. Use it to set your role, company context, or preferred format once so you don't repeat it every time.

Getting Better at This

Prompt engineering isn't a skill you learn by reading about it — it's built through practice. The fastest way to improve is to pick one task you do repeatedly (drafting emails, summarizing documents, preparing for meetings) and spend a week experimenting with how you phrase prompts for that specific task.

At Rebus Labs, we've built courses specifically around this: turning AI tools from novelty into a core part of your actual workflow. Our courses are designed for people who learn by doing, not just reading.