Asking the model to show intermediate reasoning steps before an answer—improves some math and logic tasks; still needs verification.
Chain-of-thought (CoT) means eliciting stepwise reasoning (“think step by step”) rather than jumping straight to a conclusion. It differs from the ReAct loop, which couples reasoning with external actions via tools.
Caveats
CoT can increase hallucination fluency: longer reasoning is not guaranteed correct. Use checks, tests, or retrieval when stakes are high.