Techniques for writing effective prompts that get better results from AI models.
10 tips in this topic
Paste 3-5 examples of your writing and ask Claude to analyze the style before writing new content. The model will identify patterns in tone, sentence structure, and vocabulary that it can then replicate.
Start prompts with "You are an expert [role]" to activate relevant knowledge patterns. This framing helps the model draw from specialized knowledge and adopt appropriate communication styles for that domain.
Break complex coding tasks into three parts: context (what exists), goal (what you want), and constraints (requirements and limitations). This structure prevents the AI from making assumptions and produces more accurate code.
Adding "think step by step" or "let's work through this" to prompts triggers chain-of-thought reasoning. This dramatically improves accuracy on math, logic, and multi-step problems by forcing the model to show its work.
When iterating on AI output, structure feedback as: what worked, what to change, then restate what to keep. This prevents the model from over-correcting and throwing out good elements while fixing problems.
Paste relevant documentation, examples, or reference material directly into your prompt before asking questions. AI models work best with explicit context rather than relying on training data that may be outdated or incomplete.
Temperature controls randomness. Use 0-0.3 for factual tasks (coding, math, analysis), 0.5-0.7 for balanced tasks (writing, summarization), and 0.8-1.0 for creative tasks (brainstorming, fiction). Start low and increase if outputs feel too rigid.
Few-shot prompting means giving 2-5 examples before your actual request. Format: "Example 1: [input] -> [output]. Example 2: [input] -> [output]. Now do this: [your input]". The model learns the pattern from examples.
Use GPT-4 for prototyping and diverse tasks. Fine-tune smaller models when you have consistent, narrow use cases with thousands of examples. Fine-tuned models are cheaper to run but expensive to train and maintain. Start with GPT-4, only fine-tune if cost becomes prohibitive.
Put the most important instructions at the start AND end of the system prompt (primacy and recency effects). Use clear formatting like numbered lists. Repeat critical constraints. Test with adversarial inputs to see if the model breaks character.