Most people use AI the way they’d use a search box: type a few words, hope for the best, and blame the tool when the answer is vague. Then they watch someone else get a brilliant result from the same tool and wonder what the secret is. Usually, there is no secret. There’s a better prompt.
Prompt engineering basics are just that — the basics of talking to AI clearly enough that it can actually help you. You don’t need a course, and you don’t need jargon. You need to stop treating the chatbot like a mind reader and start treating it like a very capable assistant who has just walked into the room and knows nothing about your situation.
Table of Contents
What a prompt really is
A prompt is everything you give the model before it answers: the question, the context, the format you want, and any examples. The model can only work with what you give it. Give it “write about dogs” and you’ll get generic fluff, because generic is the safest guess. Give it “write a 150-word introduction about golden retrievers for first-time owners, friendly tone, include exercise needs and one common mistake” and you’ll get something you can actually use.
That gap isn’t the model being lazy. It’s the model doing exactly what you asked — which, in the first case, wasn’t much.
The 4 ingredients of a good prompt
Think of a strong prompt as having four ingredients. You won’t need all four every time, but missing the first two is where most bad answers come from.
1. Role. Tell the AI who to be: “Act as a patient maths tutor” or “Act as a friendly email writer.” A role sets the tone and the level of detail before a single fact appears.
2. Task. Say exactly what you want, with a verb: summarise, compare, draft, list, rewrite, explain. “Tell me about climate change” is a topic, not a task. “Explain climate change in 5 bullet points for a 14-year-old” is a task.
3. Context. Who is this for? How long should it be? What should it avoid? What do you already know? Two sentences of context often do more than ten extra adjectives.
4. Format. If you want a table, say table. If you want steps, say steps. If you want a script, say script. Models follow format instructions remarkably well — when you actually give them.
Try it once, side by side, and the difference will convince you faster than any article can.
Prompt engineering basics: 9 tricks that work in practice
1. Be specific about the audience
“Explain photosynthesis” and “Explain photosynthesis to a 10-year-old who loves football” produce completely different answers. The second is better, not because the model got smarter, but because you narrowed the target.
2. Give one example
If you show the model the style you want — one headline, one paragraph, one email — it will match it far more reliably than any description of the style. “Write headlines like this: …” beats “write catchy headlines” every time.
3. Break big jobs into steps
Don’t ask for a perfect 2,000-word guide in one prompt. Ask for an outline first, pick the sections you like, then ask it to expand one section at a time. You’ll get better text, and you’ll catch errors earlier.
4. Ask it to think step by step
For maths, logic or planning, add “think step by step” or “show your working.” It genuinely improves accuracy, because the model works through the problem instead of jumping to a plausible-sounding answer.
5. Set limits
Word counts, number of items, tone, things to avoid — limits are your friend. “Give me 5 ideas, each under 20 words, no jargon” is a prompt you can judge. “Give me ideas” is not.
6. Iterate, don’t restart
When an answer is close but not right, don’t start a new chat. Say what’s wrong: “shorter,” “less formal,” “add an example about students,” “remove the last section.” Each correction teaches the model your taste, and the second draft is usually much closer.
7. Ask for alternatives
Stuck on one answer? Ask for three versions — formal, friendly, and funny — and pick the best parts of each. AI is cheap at generating options; your judgement is what makes the final choice good.
8. Check, don’t just accept
For anything factual, follow up with “what are your sources?” and then open them. This matters enormously, because fluent AI text can be wrong — see our guide on AI hallucinations: why AI makes things up and how to check facts.
9. Save prompts that work
When a prompt gives you a great result, save it. A small personal library of proven prompts — for emails, summaries, lesson plans, captions — will save you more time than any new tool subscription.
Mistakes beginners keep making
The most common mistake is the one-word prompt, or close to it. “Essay.” “Ideas.” “Help.” The model has nothing to work with, so it guesses — and guesses generically. The second mistake is the opposite: a prompt so long and tangled that the actual request gets lost in the middle. Keep the request at the start or the end, not buried in paragraph three.
The third mistake is trusting the first answer. The first answer is a draft. Good prompt engineering includes a second step: edit, question, or regenerate. And the fourth mistake is pasting private information — customer lists, passwords, unpublished work — into a public chatbot. Use tools with clear privacy policies for sensitive material, and keep personal data out of prompts entirely.
A template you can steal
Here’s a template that works for most writing tasks. Fill in the brackets and you’re already ahead of most users:
“Act as [role]. Write [what] for [audience], in [tone], about [topic]. Keep it to [length]. Include [one specific thing]. Avoid [one thing to avoid].”
Example: “Act as a friendly study coach. Write a revision plan for a biology student, in an encouraging tone, about cell biology. Keep it to 10 bullet points. Include one break schedule. Avoid medical advice.” Try it — then change one ingredient and watch how the answer changes. That’s prompt engineering, learned by doing.
For the concepts behind all this, read what is a large language model and what is generative AI. For tools that answer with sources you can check, see our Perplexity AI guide. Outbound reading: OpenAI’s prompting guidance at https://openai.com and Anthropic’s prompt library at https://www.anthropic.com.
Frequently asked questions
Is prompt engineering a real job?
Yes, in some companies — but for most people it’s a skill, not a career. Learning the basics makes every AI tool you already use work better.
Do I need to learn coding for prompt engineering?
No. Good prompts are written in plain language. Coding helps for automation, not for everyday prompting.
Why does the same prompt give different answers?
Models generate probabilistically, so wording, temperature settings and even conversation history change the output. That’s why saving good prompts and iterating matters more than finding one magic prompt.
Prompt engineering basics come down to one habit: say what you want, for whom, in what form — then
refine. Do that, and AI stops feeling like a lottery and starts feeling like a tool you actually control. These prompt engineering basics improve with practice, so keep testing prompt engineering basics in your daily work. For more on prompt engineering basics, see the prompt engineering basics guidance at https://openai.com and https://www.anthropic.com.
Once your prompt is clear, you can also show AI what you mean: our guide to multimodal AI explains how AI that sees, hears and speaks works.
Good prompting also pays off at work, so see which AI and jobs skills become more valuable, not less, as AI spreads.
