Chain of Thought (CoT)
A prompting technique where the LLM is asked to reason step-by-step before producing an answer.
What is Chain of Thought (CoT)?
Chain-of-Thought prompting is one of the most-cited findings in LLM research. By asking the model to "think step by step" — or showing it examples of step-by-step reasoning — you dramatically improve accuracy on math, logic, and multi-step questions.
The mechanism is intuitive: forcing the model to write out intermediate steps gives it more "scratch space" to keep track of context. A model that goes straight to an answer has only its hidden state to lean on; a CoT model has its own visible reasoning to refer back to.
Modern reasoning models (OpenAI o-series, Claude with extended thinking, DeepSeek R1) bake CoT into the model itself — they produce hidden reasoning tokens before the visible answer. For everyday LLM use, explicit "think step by step" prompts still help with classical models like GPT-4o and Claude Sonnet.
CoT is the single highest-leverage prompting technique. Indian Gen AI engineers default to it for anything that involves reasoning, math, or multi-step logic.
A Hyderabad fintech using GPT-4o to validate KYC documents added "think through the document step by step before deciding approval" to their prompt. Accuracy on edge cases (poor scans, unusual ID types) rose from 78% to 91%.
Want to master this?
Learn Chain of Thought (CoT) in a structured cohort
3-month live program with mentors, real projects, and 50+ partner placement support.