Reasoning Prompts (Chain of Thought)

Disclaimer. Modern reasoning-tuned models (Claude extended thinking, OpenAI o-series, DeepSeek-R1, Gemini thinking) handle multi-step reasoning natively. When using one, prefer the provider's reasoning mode and reserve these patterns for shaping the visible output. The provider's documentation is authoritative.

Chain-of-thought prompting asks the model to make its reasoning explicit before producing the final answer. This trades tokens for accuracy on tasks where the right answer depends on intermediate steps the model would otherwise skip.

When to use

CoT helps:

• Multi-step math and unit conversions.
• Logic puzzles and constraint satisfaction.
• Code debugging by tracing execution.
• Planning with dependencies.
• Decision-making with explicit trade-offs.
• Extraction from unstructured text where the answer requires linking multiple sentences.

CoT does not help (and wastes tokens):