Most people who say they “use ChatGPT” are typing one loose sentence, taking whatever comes back, and either accepting it or giving up. That’s not a skill problem so much as a technique problem. Prompt engineering sounds like a specialized job title, but the actual skill underneath it is closer to learning how to ask a very capable, very literal assistant for what you want. This guide covers what that skill actually involves, five techniques worth learning first with plain examples, and where to practice and study for free.

What prompt engineering actually is

Prompt engineering is the practice of writing instructions for an AI model in a way that reliably gets you a useful result. It is not a programming language and it doesn’t require code. It’s closer to being a clear, specific communicator with someone who has read almost everything but has no idea what’s in your head, what you already tried, or what “good” looks like to you unless you say so.

The reason this became its own skill worth naming is that the same underlying model can produce wildly different answers depending on how a question is framed. Two people can ask ChatGPT for “a marketing email” and get a generic paragraph or a sharp, on-brand draft ready to send, and the difference is almost never the model. It’s the prompt.

Five prompting techniques worth learning first

You don’t need a long list of tricks. A small set of habits covers most of what separates a flat, generic ChatGPT answer from one that actually saves you time.

1. Be specific about format, length and audience

The single highest-leverage change most people can make is simply saying what shape the answer should take. Instead of “explain photosynthesis,” try “explain photosynthesis to a curious 10-year-old in three short paragraphs, no jargon.” The model isn’t guessing at your intent anymore, it’s following a spec.

2. Give examples, also called few-shot prompting

If you want a specific tone, structure or style, show the model one or two examples of it rather than describing it in the abstract. Paste in a sample of your own writing and say “write the next paragraph in this voice,” or give two example product descriptions before asking for a third. Models are very good at pattern-matching to examples, often better than they are at following an abstract description of the same pattern.

3. Break a big task into steps

Asking for a whole finished report in one shot tends to produce something shallow and generic. Asking for an outline first, reviewing it, then asking for one section at a time, produces something you actually shaped. This also applies inside a single prompt: telling the model to “first list the key points, then write the summary” often produces a better summary than just asking for the summary directly, because it gives the model room to reason before it commits to final wording.

4. Ask it to critique or check its own answer

A simple, underused move is asking a follow-up like “what’s weak about that answer?” or “check this for factual errors and flag anything you’re unsure about” before you accept a first draft. This won’t catch everything, ChatGPT can still be confidently wrong, but it often surfaces gaps, and it builds the habit of treating the first answer as a draft rather than a final product.

5. Give it a role or persona, when it genuinely helps

Telling the model “you’re an experienced editor reviewing this for clarity” or “answer as a skeptical software architect” can sharpen the angle of a response, because it narrows what kind of feedback or voice you’re asking for. This is easy to overuse, though. A role only helps when it changes what a reasonable answer looks like; slapping “act as an expert” in front of every prompt out of habit does little on its own.

The habit underneath all five: iterate instead of expecting a perfect first answer

This is arguably the most important shift for beginners. ChatGPT works less like a search engine and more like a conversation. If the first answer is close but not right, telling it exactly what’s off, “too long,” “too formal,” “missing the pricing section,” almost always gets you closer than starting over with a longer, more complicated prompt. People who feel like “AI doesn’t really work for me” are frequently stopping after one exchange when a second or third message would have closed the gap.

How to use AI tools effectively, beyond ChatGPT specifically

The techniques above transfer across tools. Anthropic’s Claude, Google’s Gemini and ChatGPT all respond to the same underlying habits, being specific, giving examples, breaking work into steps, because they’re all large language models trained on similar principles, even though each has its own strengths and quirks. Learning to prompt well on one makes you noticeably faster at picking up another. If you’re deciding what to use AI for in the first place, not just how to phrase requests, our AI skills and career roadmap walks through where prompting fits alongside other AI skills worth building.

One habit worth building early: keep a running note of prompts that worked well for tasks you repeat, an email tone, a summary format, a way of asking for feedback on your writing. Reusing a prompt that already worked is faster than re-explaining yourself from scratch every time, and it’s the closest thing to a personal template library.

Where to practice and learn for free

You genuinely do not need to pay for this. The most effective free practice is inside ChatGPT’s own free tier: open a chat, try a real task you actually have this week, and apply one technique from above at a time so you can feel the difference it makes. That said, a structured course helps if you want the concepts named and organized rather than picked up piecemeal, and a few are worth knowing about specifically because they’re free and well built.

• DeepLearning.AI’s short courses: including one built with OpenAI specifically on prompt engineering, free to work through, code-light, and organized around real use cases like summarizing and rewriting text.
• Free, audit-only courses on Coursera: several universities and companies, including ones on Google Cloud and Vanderbilt-built prompting courses, can be audited for free, meaning you get the video lessons and readings without paying; a certificate usually requires payment or financial aid unless the provider explicitly offers one at no cost.
• OpenAI’s own prompt engineering guide: a free, text-based reference straight from the company that builds ChatGPT, useful once you’ve done some hands-on practice and want the underlying reasoning spelled out.
• Free introductory courses with a certificate of completion: a number of platforms now offer short, no-cost prompt engineering courses aimed at complete beginners that issue a free certificate at the end; treat the certificate as a nice bonus for a resume or LinkedIn profile, not a substitute for actually practicing.

If you want the broader picture of how prompting fits alongside other free ways to learn AI and large language models generally, see our guide to free generative AI and LLM courses, or start from the beginning with how to learn AI for free. You can also browse a running list of current free course listings on our free courses page.

A realistic first week

Pick three real tasks you’d normally do yourself this week, drafting an email, summarizing a long document, planning something. For each one, write a first prompt the way you normally would, then rewrite it applying one technique from this guide, being specific about format, giving an example, or breaking it into steps. Compare the two outputs side by side. This single exercise, repeated a few times, teaches the skill faster than reading about it does, because you’re seeing the exact effect of the change on a task you actually care about.

• Start with tasks you already do, not invented examples. The stakes feel real, so you’ll notice the difference.
• Change one thing at a time at first, so you know what actually moved the needle.
• Save prompts that worked. You’ll reuse more than you’d expect.
• Treat the first answer as a draft. A short follow-up almost always improves it.

Common mistakes beginners make

• Writing one vague sentence and judging the whole tool by the result: a one-line prompt with no format, audience or example gives the model very little to work with.
• Giving up after one exchange: treating the first response as final instead of refining it with a follow-up message wastes most of what makes chat-based AI useful.
• Collecting techniques without practicing any of them: reading a list of five prompting tricks does nothing until you’ve used at least one on a real task.
• Assuming a role or persona fixes everything: telling the model to ‘act as an expert’ rarely helps on its own; being specific about the actual output usually matters more.

Frequently asked questions