Prompt Engineering for Beginners: 10 Techniques That Actually Work

“Prompt engineering” is a ridiculous-sounding term for something you already know how to do: ask for things clearly. There’s no certification, no secret syntax, and you definitely don’t need those 500-word “mega-prompts” people sell for $9 on the internet. I’ve tested a lot of that stuff. Most of it is theater — long, rigid templates that fall apart the moment your actual task doesn’t match the template.

What actually moves the needle is a handful of small habits. Below are the ten I reach for every day, each with a quick before-and-after so you can see the difference instead of taking my word for it. They work on whatever AI you use — ChatGPT, Claude, Gemini — because they’re about clear thinking, not one tool’s quirks.

1. Give it a role

Tell the AI who it’s being. “You’re a copywriter for small restaurants” puts it in a completely different headspace than a cold start.

• Before: “Write a tagline for my bakery.”
• After: “You’re a seasoned brand copywriter. Write five taglines for a small neighborhood bakery known for sourdough. Warm, a little playful, under six words each.”

Same request, night-and-day output. The role acts like a filter on everything that follows.

2. Spell out the audience

The AI writes for whoever you tell it to. Skip this and you get mush aimed at no one.

• Before: “Explain compound interest.”
• After: “Explain compound interest to a 15-year-old who finds math boring. Use one everyday analogy and keep it under 120 words.”

3. Say how long you want it

This one saves me constantly. AIs love to over-write — they’ll hand you five paragraphs when you wanted two sentences. Put a leash on it up front: “In two sentences…”, “In a 5-bullet list…”, “Under 100 words…”. You’ll stop deleting filler.

4. Show an example (this is the big one)

If you want a specific style or format, paste an example of it. This is the single most powerful trick on the list, and almost nobody uses it. The AI is superb at pattern-matching — so give it a pattern to match.

• “Here’s a product description I like: [paste]. Write three more for these items in that exact voice and length.”

I use this for everything: email tone, how I format reports, the rhythm of my blog intros. One good example beats a paragraph of instructions.

5. Give it the raw material

Don’t make it guess facts. Paste your notes, the email you’re replying to, the actual data. The AI is a brilliant editor and a terrible mind-reader.

• Before: “Write my out-of-office reply.”
• After: “I’m out Aug 4–12, back the 13th. Urgent stuff goes to maya@. Write a short, friendly out-of-office, no exclamation marks.”

6. Ask for options, not an answer

When I’m stuck, I never ask for the answer. I ask for five and pick. It’s faster, and the third option is usually the keeper.

• “Give me 5 different angles for this blog intro, each one sentence.”

This works because you stop gambling on the AI reading your mind and start choosing from a menu.

7. Let it ask you questions first

Flip the script. This one feels strange until you try it, then you can’t stop.

• “Before you write the proposal, ask me any questions you need to do a great job.”

It’ll surface things you forgot to mention — budget, deadline, who’s reading it — and the final result is dramatically better because it’s working from real information instead of guesses.

8. Tell it what NOT to do

Negative instructions work. If the AI keeps doing a thing you hate, name it explicitly.

• “No corporate buzzwords. Don’t use ‘leverage,’ ‘seamless,’ or ‘unlock.’ No exclamation marks. Don’t start with ‘In today’s world.’”

I keep a little list of my banned words and paste it in. The drafts come back noticeably more human.

9. Build on the conversation

You don’t have to cram everything into one perfect prompt. Start rough, then steer: “shorter,” “more formal,” “now turn that into an email.” The chat remembers everything earlier in the thread, so each instruction stacks. I almost never nail it on message one — the good version shows up after two or three nudges. (I built a whole beginner’s guide to ChatGPT around this idea.)

10. Ask it to fix its own work

When something’s almost right, hand it back to itself: “Critique this draft like a tough editor, then rewrite it addressing your own notes.” You’ll be surprised how good the second pass is — it catches weak spots it wrote thirty seconds earlier.

A simple template to start from

When in doubt, this skeleton covers most of the techniques above:

“You’re [role]. I need [what], for [audience], in [length/format]. Here’s the raw material: [paste]. Avoid [things you hate]. Ask me anything you need before starting.”

You won’t use every slot every time. But run a weak prompt through even half of it and the jump in quality is obvious. Keep this somewhere you can paste it — it’ll become muscle memory within a week.

The mistake to unlearn

Beginners think better prompting means longer prompts. It doesn’t. It means clearer ones. A sharp two-line request beats a rambling paragraph almost every time. Add detail where it removes ambiguity — audience, length, examples, raw material — and ruthlessly cut everything else. The goal isn’t to sound technical. It’s to leave the AI with no room to guess wrong.

One more habit worth building: when an answer disappoints you, resist the urge to blame the tool and reach for a different one. Nine times out of ten, the fix is in the question. Re-read what you asked, find the ambiguity, and tighten it. That instinct — “what did I leave vague?” — is the whole of prompt engineering in one sentence.

Frequently asked questions

Do I need to learn special prompt “syntax” or commands? No. Plain, clear language is the whole game. Anyone selling you secret syntax is selling you something.

Do these work on Claude and Gemini too, not just ChatGPT? Yes. These are about clear communication, not one tool’s quirks. They transfer everywhere, including image tools.

Are paid “prompt packs” worth buying? Rarely. Once you internalize the ten habits here, you’ll write better prompts for your own situation than any generic pack can.

What’s the fastest way to improve? Steal good prompts and tweak them. Start a notes file, and every time something works, save it. Within a month you’ll have a personal library that beats anything off the shelf.

Where to go next

The quickest way to make these stick is to start from prompts that already work. My next piece, 40 ChatGPT Prompts for Everyday Tasks, is a copy-paste starter set — grab the ones that fit and reshape them with the techniques above.

Leo Tan writes hands-on AI guides at AI Play Books. Everything here comes from prompts I actually use. Something out of date? Tell me and I’ll fix it.