What is a prompt?

01 / The basics

So, what actually counts as a prompt?

If you’ve ever typed into ChatGPT “Can you make this email sound less awkward?”, or asked Google “Does freezing peanut butter sandwiches count as meal prep?” and ended up chatting with its AI response — congratulations. You’ve written a prompt.

Found a “must use” prompt from a YouTube video that was too boring to finish, but you copied the 2,000-word specification telling AI exactly how to architect a new app? Still a prompt. Just one that arrived with a laptop, three coffees, and a detailed ticket.

OpenAI describes prompt engineering as writing effective instructions so a model can generate results that meet your requirements — but “effective instructions” is the important part. You’re not learning how to speak Robot. You’re learning how to explain what you want clearly.

Think of it like ordering food. “Give me a sandwich” will probably produce a sandwich. “Give me a toasted chicken sandwich with avocado, no tomato, lots of chilli, and please don’t make it enormous because I have to eat it at my desk” has a much better chance of producing the sandwich you actually wanted. AI works surprisingly similarly.

Prompts aren’t just something you type into ChatGPT, either. You can prompt Claude, Gemini, and plenty of other AI tools — including AI built directly into coding software and IDEs. IDE stands for Integrated Development Environment: the workspace developers use to build software. Think Microsoft Word + a workshop + a mechanic’s diagnostic machine, but for code. A developer might tell an AI inside their IDE “Make this button larger, keep the existing styling, and don’t change anything else on the page.” Still a prompt. Different surroundings, same basic idea.

02 / Reusing what works

Prompt templates are just prompts you don’t want to keep rewriting

A prompt template is a prompt designed to be reused. Instead of typing this every Friday —

Write a friendly weekly update.

You might save something like:

The bracketed bits are simply the pieces you replace each time — the AI version of an email template. You wouldn’t rewrite “Hi Susan, please find attached…” from scratch 487 times if you didn’t have to.

The same goes for ordinary prompts. Eventually you’ll accidentally write one that works really well, then three weeks later remember approximately none of it and type “do that good thing you did last time.” The AI will respond with something sort of related, but definitely not as good.

OpenAI recommends treating important production prompts a bit like application code: stored, tested, and improved systematically. Your grocery-list prompt probably doesn’t need version control — but saving it is still useful if you’re tired of explaining that you hate mushrooms and already have rice. A simpler rule works fine here: if you’ve asked AI to do the same thing a few times and found a version that works, save it.

03 / A handful of habits

How to write a better prompt, without earning a degree in prompting

You do not need a 47-page MASTER SUPER PROMPT. OpenAI’s guidance for reasoning models specifically recommends keeping instructions simple and direct. Elevated prompting comes down to a handful of very normal communication skills.

1. Say what you want, and give the context that matters

   Not your whole autobiography — just what another person would need to know before they could do this properly.

2. Tell it who it's for and what the result should look like

   Audience and format change the answer more than almost anything else. Say “five bullets” if you want five bullets.

3. Give it tone, boundaries, and examples when they matter

   Tone is information too. And if a style is hard to describe, paste an example instead of writing paragraphs about it.

4. Make big prompts easy to understand

   Once you're pasting documents or data, separate your instruction from the material clearly, e.g. with a heading like “CUSTOMER FEEDBACK:”.

5. If accuracy matters, say so — and ask for evidence

   Ask it to use current sources, cite them, and say clearly when something can't be verified. Then check anything important yourself.

6. Most importantly: keep talking to it

   Your first prompt sets the starting context. It doesn't need to be perfect — you can always say “shorter”, “less formal”, or “that's not what I meant”.

Compare “Explain mortgages” with “Explain mortgages to someone buying their first home who knows absolutely nothing about finance.” Same topic, very different answer. And imagine walking into a meeting and saying “So what should we do?” — you’d probably get blank faces. Now try: “We’re launching our bakery delivery service next month. We have a $2,000 marketing budget, most customers come from Instagram, and I need three realistic ways to generate our first 100 delivery orders.” Much better starting point.

OpenAI’s own beginner guidance describes the first prompt as a starting point you refine through follow-up messages. Prompting isn’t a university entrance exam where your first answer is permanently submitted to the robot university.

04 / Across tools

Will the same prompt work in ChatGPT, Claude, Gemini, and everything else?

Usually, the fundamentals travel extremely well. A good prompt tells the AI what you want, gives it useful context, explains any important constraints, and describes what a good result looks like. Those same principles show up repeatedly in official guidance from OpenAI, Anthropic, and Google — so you generally don’t need to learn an entirely new language every time you open another AI tool.

There is, however, a catch. Different models have different capabilities, tools, and ways of interpreting instructions. A prompt for writing an email will naturally look different from one for generating an image, researching the web, editing software, analysing a spreadsheet, or having a voice conversation.

Think of it like giving instructions to a designer, an accountant, and a chef. If you tell all three “Make this better,” you may receive three fascinating interpretations of “better.” The clearer your goal is, the more portable your prompt becomes — then you adjust it for the particular tool when you need to.

05 / The recap

The beginner prompt formula

For example:

That’s already a perfectly respectable prompt. No XML. No secret commands. No “ACT AS THE WORLD’S GREATEST OMNIPOTENT EMAIL ARCHITECT.” Just good instructions.

A prompt is something you ask or tell AI. A good prompt makes what you want clearer. A prompt template is a useful prompt you save so you can reuse it. Your opening prompt gives the conversation direction, but you can refine it as you go, and the same basic communication skills work across most major AI tools. The great secret of prompt engineering is therefore slightly disappointing: clear communication works. Turns out your Year 8 English teacher was playing the very long game.