Chain-of-Thought Prompting

Chain-of-thought (CoT) prompting is a technique that instructs an AI to show its reasoning before giving a final answer. Instead of jumping to a conclusion, the model works through intermediate steps — like showing maths workings. The technique was popularised by a 2022 Google Brain paper and consistently improves accuracy on complex reasoning tasks across ChatGPT, Claude, and Gemini.

Use CoT for tasks involving multi-step reasoning: maths and calculation, logical puzzles, business decisions with multiple variables, debugging, and analysis that requires weighing competing factors. It adds little value for simple factual lookups or creative tasks where reasoning is not a bottleneck. The improvement is largest on tasks where the AI would otherwise skip over important intermediate steps.

The most effective trigger phrases are: "think through this step by step", "let's think step by step", "show your reasoning before answering", "break this down stage by stage", and "explain your logic as you go". Research shows "let's think step by step" is one of the highest-performing triggers across models. You can also provide an example of reasoned output (few-shot CoT) for even stronger results.

CoT prompting works well with large language models (GPT-4, Claude, Gemini). Smaller models benefit less because they lack the capacity to produce coherent extended reasoning chains. For best results, use CoT with the most capable model available to you. Claude responds particularly well to CoT because of its instruction-following training. GPT-4o and Gemini 1.5 Pro also show strong CoT performance.

Zero-shot CoT adds a simple trigger like "think step by step" without providing examples. Few-shot CoT provides 2-3 worked examples showing the reasoning pattern you want before your actual question. Few-shot CoT consistently produces more reliable and structured reasoning but requires more prompt writing. For most everyday tasks, zero-shot CoT ("think step by step") is sufficient. Use few-shot CoT when you need a specific reasoning structure or format.