How to Write Better AI Prompts — A Practical Guide

How to write better AI prompts.

The same few habits make every AI assistant — ChatGPT, Claude, Gemini, Perplexity — dramatically more useful. Here's what actually works, with patterns you can copy.

Everyone has access to the same AI assistants, yet some people get sharp, useful answers and others get bland, generic ones. The difference is almost never the model — it's the prompt. The good news is that better prompting is a handful of learnable habits, not a dark art. This guide walks through what actually moves the needle, with patterns you can copy into ChatGPT, Claude, Gemini, Perplexity or any other assistant.

The one thing most people get wrong

The most common mistake is being too vague. "Write me a marketing email" could mean a thousand different things, so the model picks the blandest average of all of them. A prompt is not a search query — it's a brief. The more relevant context you hand over, the less the AI has to guess, and guessing is where generic, off-target answers come from. Treat every prompt as if you were briefing a smart new colleague who knows a lot in general but nothing about your situation.

The anatomy of a good prompt

Strong prompts tend to include four ingredients. You don't need all four every time, but reaching for them turns a weak prompt into a good one:

A role. Tell the AI who to be — "act as a copy editor," "you're a patient maths tutor," "review this like a senior engineer." A role sets the tone, vocabulary and level of detail in one short phrase.
Context. Give the background that matters: who the audience is, what you've already tried, any constraints, and the raw material itself. Paste the actual text, data or error rather than describing it.
A specific task. Say exactly what you want done, as a clear instruction: summarise, rewrite, compare, brainstorm, explain, fix. One prompt, one main job.
An output format. Tell it how you want the answer — a bulleted list, a table, three options, under 100 words, plain English with no jargon. If the shape matters, ask for it up front.

Here's the same request without and with those ingredients. Weak: "Give me some ideas for my newsletter." Strong: "You're an email marketer for a small indie coffee brand. Our newsletter goes to existing customers who like the story behind the beans. Suggest 8 subject-line-plus-topic ideas for our next issue, mixing product news with useful brewing tips. Give them as a short list, each with a one-line description." The second one gets a usable answer on the first try.

Show an example when you can

One of the highest-leverage moves is to show, not just tell. If you want output in a particular style or structure, paste an example of it: "Here's a product description I like: [paste]. Write three more for these items in the same voice." Giving the model one or two examples of what "good" looks like — sometimes called few-shot prompting — is far more reliable than describing the style in adjectives.

A reusable template

When a request matters, this simple skeleton covers the essentials. Fill in the brackets and delete any line you don't need:

Act as [role]. Context: [who this is for, background, constraints]. Task: [the specific thing you want]. Use this as source material: [paste your text or data]. Format the answer as [list / table / word count / tone].

It looks formal written out, but in practice you'll internalise it quickly and start including these pieces without thinking about it.

Iterate — it's a conversation, not a vending machine

The first answer is a starting point, not the finish line. Instead of rewriting your whole prompt from scratch, steer with follow-ups: "make it shorter," "more formal," "focus on the second option," "you missed the budget constraint — try again." Because the assistant remembers the thread, each nudge builds on the last. Some of the best results come on the third or fourth turn, not the first.

Common mistakes to avoid

Piling five tasks into one prompt. The model tends to do the first well and skim the rest. Break big jobs into steps.
Burying the key instruction. Put the most important requirement at the start or the very end, where it's least likely to get lost.
Asking for facts without checking them. AI can state wrong things confidently. For anything factual or important, ask for sources and verify them yourself.
Accepting a mediocre answer. If it's not right, say what's wrong and ask again. That single habit separates people who find AI useful from people who don't.

Put these together and almost any assistant becomes noticeably more helpful. Want ready-made starting points? Browse our copy-and-paste prompt packs for writing, work, study and more, or the model-specific packs for ChatGPT, Claude and Gemini.

FAQ

What makes a good AI prompt?

A good prompt gives the AI enough to work with: who it should act as, the context and background, the specific task, and the format you want back. Vague prompts get vague answers, so the more relevant detail you provide, the better the result.

Do the same prompts work in ChatGPT, Claude and Gemini?

Mostly, yes. The principles — context, a clear task, an example and a format — work across all the major assistants. Each model has its own personality and strengths, so if one answer disappoints, paste the same prompt into another and compare.

How long should a prompt be?

As long as it needs to be to include the relevant details, and no longer. A one-line question is fine for simple lookups. For anything that matters, a short paragraph with context, the task and the desired format almost always beats a terse one-liner.

Why does the AI keep ignoring part of my prompt?

Usually the instruction was buried or ambiguous. Put the most important requirement first or last, use plain imperative sentences, and split a big request into steps. If it still slips, point out what it missed and ask it to try again — iterating is part of the process.