Instead of relying on the model's training data, provide the source material directly in your prompt. This is the most reliable way to get accurate, specific answers.
Paste the relevant document, data, or context into your prompt and instruct the AI to answer based only on what you've provided: "Answer the following question using only the information in the document below. If the answer isn't in the document, say so. Question: [question]. Document: [paste]"
Ask the AI to explore multiple reasoning paths before converging on an answer. Improves results for complex decisions with multiple valid approaches.
"Consider 3 different approaches to solving [problem]. For each approach: describe the method, pros, cons, and when it works best. Then recommend which approach fits my situation: [context]"
After getting an answer, ask the AI to critique it. This catches errors and gaps before you act on the output.
After any important response, follow up with: "Now review what you just wrote. What assumptions did you make? What might be wrong or incomplete? What important considerations did you miss?"
Use different personas sequentially on the same problem to get multiple perspectives. Ask a pessimist, then an optimist, then a pragmatist.
"Evaluate my business plan from 3 perspectives: 1) A skeptical investor looking for reasons it will fail. 2) An enthusiastic advisor looking for opportunities. 3) A pragmatic operator focused on execution. Plan: [paste]"
Specify exact output structure using examples, schemas, or templates. This is especially useful when you need to feed AI output into another system.
"Output your answer as valid JSON with this exact structure: {'title': string, 'summary': string (max 50 words), 'tags': array of strings (max 5), 'priority': 'high'|'medium'|'low'}. No other text."
Break complex tasks into explicit steps and have the AI complete them sequentially, checking in at each stage.
"Complete this task in 3 steps. After each step, show your output and wait for me to confirm before proceeding. Step 1: [task 1]. Step 2: [task 2]. Step 3: [task 3]."
Chain-of-thought and few-shot work across all major models. Claude responds especially well to structured XML prompts. See our Claude-specific tips and full prompt engineering guide.