Chain of Thought Prompting

The Problem

AI models often jump to conclusions when faced with multi-step problems, producing answers that sound confident but miss critical nuances. Without explicit reasoning steps, the model takes shortcuts that lead to wrong conclusions in math, logic, planning, and analysis tasks.

The Prompt

Think through this step by step before giving your final answer.

PROBLEM:
[describe your question, task, or scenario here]

For each step:
1. State what you know so far
2. Identify the next logical question or sub-problem
3. Reason through it explicitly
4. Check if your reasoning contradicts any earlier steps

After all steps, provide:
- **Final Answer**: Your conclusion based on the reasoning chain
- **Confidence**: How certain you are (high/medium/low) and why
- **Assumptions**: Any assumptions you made along the way

Example Output

Step 1: Identifying the core question...
We need to determine whether migrating from REST to GraphQL reduces total API calls for our dashboard.

Step 2: Analyzing current REST usage...
The dashboard makes 6 separate REST calls on load: /users, /projects, /stats, /notifications, /settings, /recent-activity.

Step 3: Modeling the GraphQL alternative...
A single GraphQL query could fetch all 6 data sets in one request, but nested resolvers may still hit 6 database queries on the backend.

Step 4: Evaluating the tradeoff...
Frontend calls drop from 6 to 1, but backend complexity increases. Net latency improves because we eliminate 5 round trips.

Final Answer: Migration reduces frontend API calls from 6 to 1, improving perceived performance by ~400ms on average.
Confidence: High — the reduction in network round trips is well-documented.
Assumptions: Backend resolver performance remains comparable to current REST handlers.

When to Use

Use this skill whenever you face a problem that requires reasoning through multiple steps — debugging complex issues, evaluating architecture decisions, analyzing tradeoffs, or solving logic puzzles. It is especially effective for tasks where the first intuitive answer is often wrong.

Pro Tips

• Be specific about the domain — adding context like “You are analyzing a React 18 app with server components” narrows the reasoning to relevant patterns.
• Ask for counter-arguments — append “Now argue against your own conclusion” to stress-test the reasoning chain.
• Combine with few-shot examples — show one worked example before your actual problem so the model mirrors the depth and format of reasoning you expect.