How to Write Better AI Prompts: Prompt Engineering Basics

The gap between a disappointing AI answer and a genuinely useful one is almost never the model. It is the prompt. Most people type a vague one-liner, get a generic reply, and conclude the tool is overrated. This guide shows you how to write prompts that get sharp, usable results, no technical background required.

Why prompting matters so much

A generative AI model responds to exactly what you give it. If your request is fuzzy, the model has to guess your intent, your audience, and your desired format, and it usually guesses toward the safe, average answer. A precise prompt removes that guessing and points the model straight at what you actually want.

If you are new to how these tools work at all, our beginner’s guide to generative AI is a good five-minute primer before you dive in here.

A simple formula that works

You do not need to memorise dozens of tricks. Most great prompts contain four ingredients:

1. Role or context. Tell the model who it is or what situation applies. “You are a patient maths tutor for a 12-year-old.”
2. Task. State the single clear job. “Explain long division using a real-world example.”
3. Format. Say how you want the answer. “Use three short steps and one worked example.”
4. Constraints. Add limits or must-haves. “Keep it under 150 words and avoid jargon.”

Put together, that becomes a prompt the model can nail on the first try instead of the fourth.

Weak prompt vs strong prompt

Weak: “Write about email marketing.”

Strong: “You are a marketing copywriter. Write a 120-word promotional email for a small bakery announcing a new weekend brunch menu. Warm, friendly tone. End with a clear call to action to book a table. Avoid clichés like ‘look no further.’”

The second version tells the model the role, the task, the length, the tone, the goal, and one thing to avoid. The output will be dramatically better, and you will spend less time editing.

Techniques that reliably help

Once the basics are second nature, these upgrades push quality further:

• Give an example. Show one sample of the style or structure you want. Models imitate examples extremely well. This is often called “few-shot” prompting.
• Ask for the format explicitly. Request a table, a numbered list, bullet points, or a specific word count. Vague requests get wall-of-text answers.
• Let it think step by step. For reasoning, planning or maths, ask the model to work through the problem before giving a final answer. This alone improves accuracy on tricky tasks.
• Assign a persona. “Act as a skeptical editor” or “Act as a friendly onboarding coach” shifts tone and priorities usefully.
• Iterate, do not restart. Treat it as a conversation. Instead of rewriting from scratch, reply with “make it shorter,” “more formal,” or “add a concrete example.”

Common mistakes to avoid

Most weak results trace back to a handful of habits:

• Being too vague. “Make it better” gives the model nothing to act on. Say what better means.
• Cramming five tasks into one prompt. Break big jobs into steps and chain them.
• Assuming shared context. The model does not know your company, your audience, or last week’s chat unless you tell it.
• Trusting output blindly. AI can state wrong facts confidently. Always verify anything important, especially numbers, quotes and citations.
• Ignoring follow-ups. The second and third message are where good answers become great ones.

Prompting across different tools

The principles here work almost everywhere, but each tool has strengths worth matching to your task:

• For long, nuanced writing and careful reasoning, Claude responds especially well to detailed context.
• For all-round tasks and a huge ecosystem, ChatGPT is the default starting point.
• For research where you need sources, Perplexity cites as it answers, so your prompts can ask for references directly.
• For fast, Google-connected answers, Gemini is convenient.

You can compare more options in the chatbots category or browse the full directory of AI tools to find one suited to your workflow.

A reusable template

Copy this and fill in the blanks for almost any task:

“You are [role]. Your task is to [specific job]. The audience is [who]. Return the answer as [format and length]. Follow these rules: [constraints]. Here is an example of the style I want: [example].”

Save a few filled-in versions for jobs you do often, such as summarising meetings or drafting outreach. Reusing a proven prompt is faster than reinventing one each time, and it is the quiet habit behind people who seem to “just get” AI.

The bottom line

Prompt engineering is not magic and it is not coding. It is clear briefing: give the model a role, a task, a format and a few constraints, show an example, then refine through conversation. Do that consistently and the same tools that felt underwhelming start producing work you can actually ship.

Want to put these techniques to work? Browse our directory of AI tools and pick one task to practise on this week, then compare notes with our guide to free vs paid AI tools when you are ready to level up.