Intermediate 12 min read

Advanced Prompting Techniques

You've mastered the basics. Now learn the techniques that professionals use to get consistently excellent results from AI models.

1. Chain-of-Thought Prompting

Chain-of-thought (CoT) prompting asks the AI to think through a problem step by step before giving its final answer. This dramatically improves accuracy on maths, logic, and multi-step reasoning tasks.

Without CoT

"If a shop sells 3 shirts at $12 each and 2 trousers at $25 each, and there's a 10% discount on the total, what's the final price?"

AI might jump to an answer and get it wrong

With CoT

"If a shop sells 3 shirts at $12 each and 2 trousers at $25 each, and there's a 10% discount on the total, what's the final price? Think through this step by step, showing your calculations."

AI breaks it down: shirts total, trousers total, subtotal, discount, final

Magic phrases: "Think step by step", "Show your reasoning", "Walk me through your thought process", "Break this down into steps before answering"

2. Few-Shot Prompting

Instead of explaining what you want, show the AI 2-3 examples. This is often more effective than lengthy instructions — the AI pattern-matches from your examples.

Example: Generating Product Descriptions

Input: Red wool scarf, 180cm, handmade in Scotland

Output: Wrap yourself in warmth with this handcrafted Scottish wool scarf. At 180cm, it's generous enough to style multiple ways — whether draped casually or wound snugly against the Highland winds. The rich red hue adds a pop of colour to any winter outfit.

Input: Blue ceramic mug, 350ml, dishwasher safe

Output: Start your morning right with this ocean-blue ceramic mug. Holding a generous 350ml, it's the perfect size for your favourite brew. Best of all? It's dishwasher safe, so you can spend less time washing up and more time enjoying your cuppa.

Input: Black leather wallet, RFID blocking, 12 card slots

Output: [AI generates in the same style]

Notice how the AI picks up on the style (conversational, benefit-focused), length (3 sentences), and structure (feature → benefit → personality) from your examples.

3. Structured Output Requests

When you need data in a specific format, tell the AI exactly what structure to use. This is especially useful when you'll be copying the output into a spreadsheet, document, or application.

Example

"Analyse this job description and extract the key requirements. Return as a table with these columns: | Requirement | Type (must-have/nice-to-have) | Your Match (yes/partial/no) | Base the 'Your Match' column on this CV: [paste CV]"

You can request JSON, CSV, Markdown tables, bullet lists, numbered steps — any structured format. The more explicitly you define it, the more consistent the output.

4. Persona Stacking

Combine multiple perspectives in a single prompt to get richer, more nuanced output.

Example

"Review this business proposal from three perspectives: 1. As a venture capitalist evaluating investment potential 2. As a sceptical customer who has seen similar products fail 3. As a technical architect assessing feasibility For each perspective, give 3 strengths and 3 concerns."

5. Strategic Constraints

Constraints aren't limitations — they're creative tools. The right constraints force the AI to produce exactly what you need.

Length: "In exactly 3 sentences..."

Audience: "Explain to a 10-year-old..."

Style: "In the style of a BBC News article..."

Exclusion: "Without using the word 'innovative'..."

Format: "As a pros/cons table..."

Honesty: "If you're not sure, say so..."

6. Meta-Prompting (Ask AI to Help You Prompt)

One of the most powerful techniques: ask the AI to help you write a better prompt. The AI knows what information it needs to give you the best answer.

"I want to use AI to help me plan a website redesign. Before I give you the task, ask me the 10 most important questions you'd need answered to give me the best possible advice. Ask them one at a time."

This flips the conversation — instead of you guessing what information to provide, the AI tells you what it needs. The result is usually much better than going straight to the question.

Technique Quick Reference

Technique Best For Key Phrase
Chain-of-Thought Maths, logic, analysis "Think step by step"
Few-Shot Consistent formatting "Here are 2 examples..."
Structured Output Data extraction "Return as a table..."
Persona Stacking Balanced analysis "From 3 perspectives..."
Constraints Focused output "In exactly 3 sentences..."
Meta-Prompting Complex tasks "Ask me 10 questions first..."