Advanced Prompting: Chain-of-Thought & Beyond
Level up your AI skills with advanced prompting techniques: chain-of-thought reasoning, few-shot examples, role-playing, and meta-prompting strategies that produce expert-level legal output.
Step 1: Chain-of-Thought (CoT) Prompting
Chain-of-thought prompting asks the AI to show its reasoning step by step. Instead of 'Is the defendant liable?', use 'Analyze liability step by step: 1) Identify the duty of care owed, 2) Determine if the duty was breached and how, 3) Establish causation between the breach and injuries, 4) Assess damages. For each step, cite the relevant legal standard and apply it to these facts: [case details]'. CoT dramatically improves accuracy for complex legal analysis because it forces the AI to reason methodically rather than jumping to conclusions.
Step 2: Few-Shot Prompting with Examples
Few-shot prompting provides examples of desired input-output pairs before your actual request. Example: 'I will show you how I want case summaries formatted, then you'll do the same for a new case. EXAMPLE INPUT: Rear-end collision, client has whiplash, $15K medical bills EXAMPLE OUTPUT: **Case Type:** Motor Vehicle Accident (Rear-End) **Liability:** Strong — rear-end collision creates presumption of negligence **Injuries:** Cervical strain/whiplash (soft tissue) **Specials:** $15,000 **Estimated Value:** $45K-$75K (3-5x multiplier for soft tissue) Now summarize this case: [your actual case details]'. This technique teaches the AI your exact format and analysis style.
Step 3: Role-Playing and Adversarial Prompts
Use role-playing to stress-test your arguments. Techniques: Devil's Advocate — 'Act as the defense attorney. Read my demand letter and identify every weakness, factual gap, and argument you would use to reduce the settlement value.' Red Team — 'You are the insurance company's claims adjuster. What questions would you raise about this claim? What additional documentation would you request? Where do you see opportunities to deny or reduce the claim?' Mock Judge — 'As a judge, evaluate the merits of this motion. Would you grant or deny it? What additional briefing would you want?'. These adversarial prompts reveal blind spots in your case strategy.
Step 4: Meta-Prompting and Self-Reflection
Meta-prompting asks the AI to improve its own output. Powerful patterns: Self-Critique — after getting output, ask 'Now review what you just wrote. Identify 3 weaknesses and rewrite to address them.' Progressive Depth — 'First, give me a high-level outline. [review] Now expand section 3 in detail. [review] Now add case citations to the liability argument.' Quality Scoring — 'Rate this demand letter on a scale of 1-10 for: persuasiveness, legal accuracy, organization, and completeness. Then rewrite it to score at least 9 in each category.' This iterative approach produces dramatically better output than single-shot prompting.
Step 5: Multi-Model Strategy
Different AI models excel at different tasks. The advanced strategy is to use multiple models: Use Claude for long-document analysis and careful reasoning (medical records, depositions, contracts). Use ChatGPT for creative drafting and role-playing (demand letters, client communications). Use Gemini for real-time research and fact-finding (case law, opposing party research, verdict data). Use o1/o3 reasoning models for complex logical analysis (settlement calculations, multi-party liability allocation). Copy outputs between models: have Claude analyze records, then feed that analysis to ChatGPT for demand letter drafting. The best lawyers use AI like a team of specialists.
Key Takeaways
Related Tools
Video Resources
Advanced tutorial on Chain-of-Thought prompting with real-world applications for LLMs.
Comprehensive guide covering techniques that improve LLM accuracy for complex reasoning.
Comprehensive crash course on advanced prompt engineering including few-shot and chain-of-thought techniques.