Chain of Thought (CoT)

Chain of thought (CoT) means reasoning step by step in writing. A 2022 Google paper showed that simply prompting large models with worked examples of step-by-step reasoning, or even adding “Let’s think step by step”, made them much better at arithmetic and logic puzzles.

From prompt trick to built-in feature

• 2022: CoT is a prompt engineering technique.
• 2024–2025: reasoning models are trained to produce long internal chains of thought before answering.
• 2026: adjustable “thinking” is standard in frontier models.

Why it helps

Writing out steps gives the model more computation per problem and lets it build on intermediate results instead of jumping straight to an answer.

A window into the model?

Visible reasoning helps humans follow and audit what a model is doing, a useful safety tool. But research has shown that a model’s written reasoning doesn’t always faithfully reflect what actually drove its answer. Many safety researchers argue labs should preserve “monitorable” chains of thought as models get more capable. See interpretability.