How to Write an AI Prompt That Actually Works (With Examples)

Why vague prompts give vague answers

An AI assistant answers the prompt you gave it, not the one in your head. “Explain photosynthesis” leaves every decision open: length, level, style, what to include, what to skip. The model fills those gaps with the most average choices possible — which is why the first answer so often reads like a bland encyclopedia entry. A good prompt is simply you making those decisions yourself, in advance, in writing.

The five parts of a prompt that works

1. Role — who should the AI be? “You are an expert tutor…” or “…an experienced copywriter” steers vocabulary, depth and tone more than any other single line.
2. Task — one clear verb-driven job: explain, outline, compare, critique, rewrite. One prompt, one main task.
3. Context — the facts it cannot know: your level, your exam, your audience, the product, the deadline word limit. Context is where most weak prompts are starving.
4. Constraints — the boundaries: length, reading level, what to avoid, “use two examples”, “no jargon”. Constraints prevent the generic filler you would delete anyway.
5. Output format — the shape of the answer: an outline first then detail, bullet points, a table, three short paragraphs. If you do not choose the shape, you get a wall of text.

Example 1: study help, before and after

Before: “Explain the causes of World War One.”

After: “You are an expert tutor and academic writing coach. Topic: the causes of World War One. My goal: understand the causes and plan a structured essay. My level: university undergraduate. Structure your help exactly like this: 1) a short outline of the main points, 2) each cause explained simply with one example, 3) one short model paragraph I can learn the style from, 4) three to five proper references (APA style) I can read and cite.”

The “after” version gets an answer you can actually study from — and note what it does not ask for: a finished essay to copy. Prompts for learning (outlines, explanations, model paragraphs, references) are both more useful and honest; if you are submitting work, run it through the routine in our guide to plagiarism checks before it goes anywhere.

Example 2: an image prompt

Before: “A wallet.”

After: “Create an image of a handmade brown leather wallet on a rustic wooden table, soft window light. Purpose: an eye-catching product photo for an online shop listing. Style: photorealistic. Composition: subject clear and in sharp focus, balanced framing, warm colours, clean background. No text, logos or watermarks in the image.” Subject, setting, light, purpose, style, composition, exclusions — image models respond to exactly these levers.

Example 3: a business caption

Before: “Write an Instagram caption for my page.”

After: “You are an expert copywriter who writes copy that sounds human, not salesy. Write an Instagram caption for an online assignment-support service. Goal: get interested students to send us a message. Tone: persuasive. Must include: a mention of a free first sample; under 80 words; short sentences; one clear call to action at the end; no hype words or fake claims.” The word limit and the “no fake claims” line do real work here — they cut the two most common AI-caption diseases, bloat and exaggeration, before they happen.

Iterate: treat the first answer as a draft

Even a strong prompt rarely lands perfectly first time, and you do not need a new prompt — you need a follow-up. “Make point two simpler, with a daily-life example.” “Half the length, keep the structure.” “Same answer, but as a table.” Each follow-up is cheap and precise because the role, context and format are already established in the conversation. And when you find a combination that works, save it: your best prompts become reusable templates. That is exactly what our prompt generator produces — fill four fields and it assembles the role, task, context, constraints and format for you, with presets for study, image, video and business prompts to start from.