AI Prompt Engineering 101: Get Better Results Every Time

The difference between a mediocre AI output and an extraordinary one is almost always the prompt. Prompt engineering is the skill of communicating clearly and precisely with AI systems — and the good news is that it can be meaningfully improved in under an hour of practice.

The 5-Part Prompt Formula

The best prompts across all AI tools share five elements. You do not need all five for every prompt, but including more of them consistently produces better output.

1. Context — Who you are and why you need this. "I'm a marketing manager at a SaaS startup" gives the AI more to work with than nothing.

2. Task — What specifically you want. "Write a 3-sentence product description for our landing page" is better than "write something about our product."

3. Format — How you want the output structured. "Give me 5 bullet points" or "write this as a numbered list with subpoints" or "respond in JSON format."

4. Tone — The voice, register, and style. "Professional but approachable" or "technical and precise" or "conversational and Gen-Z friendly."

5. Constraints — What to avoid. "Don't use jargon", "keep it under 100 words", "avoid passive voice", "don't mention competitors."

Prompting for Image Generation (Midjourney, DALL-E, Stable Diffusion)

Image generation prompts reward extreme specificity. The more visual detail you include, the more control you have over the output.

Bad prompt: "a woman in a garden"

Good prompt: "Portrait photograph of a woman in her 30s with dark curly hair, sitting in a lush English cottage garden in golden hour light, shallow depth of field, Canon 5D, 85mm lens, soft bokeh background of roses and lavender, warm colour grading, National Geographic style"

Key modifiers to learn: photography style (portrait, editorial, street), lighting (golden hour, studio, dramatic side lighting), camera/lens references (85mm f/1.4, wide angle, fisheye), artistic style (photorealistic, oil painting, watercolour, anime), quality markers (8K, highly detailed, professional photography).

For Midjourney specifically: Use --ar 16:9 for widescreen, --v 6 for the latest model, and --s 750 to increase stylization. Reference specific artists whose style you want: "in the style of Gregory Crewdson" or "Studio Ghibli aesthetic."

Prompting for Writing AI (ChatGPT, Claude, Gemini)

Give the AI a role. Start your prompt by assigning a persona: "You are a senior copywriter with 15 years of experience writing for luxury brands." Role assignment consistently improves output quality because it loads relevant context into the model's generation process.

Show, don't just tell. Provide an example of what good looks like. "Write a product description in this style: [paste example]" is far more reliable than describing the style in abstract terms.

Use negative instructions. Tell the AI what NOT to do. "Do not start sentences with 'I'", "avoid clichés like 'game-changer' and 'cutting-edge'", "do not use em dashes" — negative constraints significantly improve the naturalness of AI writing.

Ask for options. "Give me 5 different versions of this, each with a different tone" produces more usable material than hoping the first output is perfect.

Prompting for Code AI (Cursor, GitHub Copilot, Claude)

Specify your stack explicitly. "Using React 18, TypeScript, and Tailwind CSS" prevents the AI from defaulting to older patterns or different technologies.

Describe the input and output. "This function takes an array of user objects and returns a new array filtered by active status, sorted by creation date descending." Precise input/output descriptions produce more accurate function implementations than vague descriptions.

Ask for tests. "Write this function and then write three unit tests for it using Jest" produces better code than just asking for the function, because the AI has to reason about edge cases to write the tests.

Chain Prompting for Complex Tasks

Complex tasks almost always produce better results when broken into steps. Instead of one massive prompt, use a sequence:

Step 1 — Research/outline: "Give me an outline for a 1,500-word article about [topic] targeting [audience]."

Step 2 — Expand sections: "Using this outline, write section 2 in full. Tone: [specify]. Include [specific requirements]."

Step 3 — Refine: "Rewrite the second paragraph to be more concise and to lead with the most important point."

This chain approach gives you more control, reduces hallucinations, and produces dramatically more consistent results than trying to generate everything in one shot.

Common Mistakes to Avoid

• Being too vague: "Write something good about my product" — the AI has almost no information to work with.
• Not iterating: Treating the first output as final. Prompt engineering is a conversation — refine, redirect, and improve.
• Ignoring the system prompt: In tools that support it (Claude, ChatGPT with Custom Instructions), a well-crafted system prompt permanently improves all outputs.
• Asking too many things at once: "Write a blog post, suggest 5 social media captions, create 3 email subject lines, and give me a content calendar" — split these into separate prompts for better results on each.
• Not providing context: The AI does not know your audience, your brand, your previous content, or your goals unless you tell it.