Chain-of-Thought Prompting

Chain-of-thought (CoT) prompting is a technique where you ask the AI to think through a problem step by step before providing its final answer. Instead of jumping straight to a conclusion, the model works through its reasoning explicitly — and this simple change dramatically improves accuracy on complex tasks.

The power of thinking step by step

Consider this prompt: "A store has 23 apples. They sell 7, receive a shipment of 12, then sell 5 more. How many apples do they have?"

Without chain-of-thought: The model might jump to an answer and occasionally get it wrong.

With chain-of-thought: "Let me work through this step by step. Starting with 23 apples. After selling 7: 23 - 7 = 16. After receiving 12: 16 + 12 = 28. After selling 5: 28 - 5 = 23. The store has 23 apples."

The step-by-step approach forces the model to show its work, catching errors that might otherwise slip through.

Why chain-of-thought works

LLMs generate text one token at a time. When asked to jump directly to an answer, the model has to compute the entire solution in a single step — which is difficult for complex problems. Chain-of-thought prompting gives the model space to break the problem down, with each step informing the next. The intermediate reasoning tokens effectively serve as "working memory."

Research has shown that chain-of-thought prompting can improve accuracy by 20-40% on reasoning-heavy tasks compared to direct prompting.

How to trigger chain-of-thought

Several approaches work:

• Explicit instruction: "Think step by step before answering."
• Structured request: "First, analyse the problem. Then, consider the options. Finally, provide your recommendation."
• Role framing: "You are a senior analyst. Walk me through your reasoning before presenting your conclusion."
• Template: "Problem: [X]. Step 1: [analysis]. Step 2: [evaluation]. Step 3: [recommendation]."

Even the simple phrase "Let's think step by step" has been shown to significantly improve model performance on reasoning tasks.

When to use chain-of-thought

Chain-of-thought is most valuable for:

• Mathematical reasoning: Multi-step calculations, financial analysis, statistical interpretation
• Logical deduction: Arguments with multiple premises, conditional reasoning
• Complex decisions: Evaluating multiple options against multiple criteria
• Code debugging: Walking through code execution step by step
• Strategy development: Analysing situations with multiple factors before recommending action
• Comparative analysis: Systematically evaluating alternatives before making a recommendation

When not to use chain-of-thought

Chain-of-thought adds length and processing time. Skip it for:

• Simple factual questions ("What is the capital of France?")
• Creative tasks where reasoning is not relevant (brainstorming, writing poetry)
• Tasks where speed matters more than accuracy
• Simple classification with clear categories

Chain-of-thought variants

• Zero-shot CoT: Just add "Let's think step by step" to your prompt
• Few-shot CoT: Provide examples that include step-by-step reasoning
• Self-consistency: Ask the model to solve the problem multiple times using different reasoning paths, then take the most common answer
• Tree of thought: Have the model explore multiple reasoning branches before selecting the best path

Practical business examples

• Financial analysis: "Analyse this P&L statement step by step. First identify the key trends. Then compare to industry benchmarks. Finally, recommend actions."
• Risk assessment: "Evaluate this proposal. First list the potential risks. Then assess the probability and impact of each. Finally, provide an overall risk rating with justification."
• Strategic planning: "Consider our market entry options. First analyse each market. Then evaluate our competitive position in each. Finally, recommend the best option with your reasoning."