How to create effective AI prompts

Proven techniques for getting better results from ChatGPT, Claude and other language models

8 min

The quality of what you get from an AI model depends directly on how you ask. A vague prompt produces a generic response; a well-structured prompt produces results that rival the work of a specialised professional.

Prompting isn’t a mysterious art: it’s a skill that can be learned and systematised. This guide covers the fundamental techniques that transform a casual user into an advanced generative AI user.

Structure of an effective prompt

A well-built prompt has four components: context (who you are and what you need), instruction (what the AI should do), format (how you want the response) and constraints (what to avoid or limit).

Not every prompt needs all four components, but the more complex the task, the more explicit you need to be in each dimension.

  • Context: "I’m the marketing lead at a B2B SaaS company with 50 employees"
  • Instruction: "Write 5 email subject lines for a win-back campaign targeting inactive customers"
  • Format: "Each subject should be under 50 characters. Present results in a table with subject and approach columns"
  • Constraints: "Don’t use exclamation marks or aggressive sales language"

The power of context

Context is the most undervalued prompting component. Language models generate generic responses because most users don’t provide enough context about their specific situation.

The more relevant context you include (industry, company size, target audience, brand tone, specific constraints), the more precise and useful the response will be. The model can’t guess what you don’t tell it.

Role prompting: assigning a role to the model

Role prompting means asking the model to act as a specific professional profile. This adjusts the level of detail, terminology and focus of the response.

  • "Act as a CFO evaluating an investment proposal" → response focused on ROI, risk and payback
  • "Act as a senior copywriter at a creative agency" → response with persuasive, creative tone
  • "Act as a senior DevOps engineer" → technical response with infrastructure considerations
  • Combine role with context: "You’re a UX consultant working with a fashion ecommerce brand"

Few-shot: teaching with examples

The few-shot technique includes 2-3 examples of the expected result within the prompt. This is especially useful when you need a specific format or style that’s hard to describe in words.

Examples act as an implicit template: the model detects the pattern and reproduces it. It’s the most effective technique for classification, reformulation and content generation with a specific style.

Chain-of-thought: step-by-step reasoning

The chain-of-thought technique asks the model to reason step by step before giving a final answer. This reduces errors in tasks requiring logic, calculation or analysis of multiple variables.

  • "Before answering, analyse the pros and cons of each option"
  • "Think step by step: first identify the problem, then the causes, then the solutions"
  • "Show your reasoning before giving the final conclusion"
  • Especially useful in business analysis, problem-solving and alternative evaluation

Iteration and refinement

The best results rarely come from the first prompt. Prompting is an iterative process: you launch a prompt, evaluate the response, identify what’s missing or excessive and adjust.

Refinement techniques include asking for more detail on a specific point, requesting a tone change, adding constraints you hadn’t considered or asking the model to critique its own response and improve it.

  • "Expand point 3 with more detail and concrete examples"
  • "Reframe this for a non-technical executive audience"
  • "Review your response and flag possible weaknesses or blind spots"
  • "Keeping the content, reduce the length by half"

Key Takeaways

  • An effective prompt has context, instruction, format and constraints
  • Specific context transforms generic responses into useful ones
  • Role prompting adjusts the detail level and perspective of the response
  • Examples (few-shot) are the most effective way to teach a style or format
  • Prompting is iterative: refine and adjust until you get the desired result

Want your team to master prompting?

We train your team in advanced prompting techniques to get the most from generative AI tools.