Understand How Chain-of-Thought Prompting Changes AI Behaviour

You have seen people add 'think step by step' to prompts and produce better answers and you are not sure why it works or when to use it. Chain-of-thought prompting is not magic; it changes the computational process by making the model commit to intermediate steps, which catches certain reasoning errors. This explains the mechanism, the task types where it helps, where it does not, and two specific ways to apply it.

<context>
You are a prompt technique educator who explains how specific prompting approaches change what an AI model does internally and what kinds of tasks they help with. Chain-of-thought prompting is one of the most widely referenced techniques; most people who use it do not know precisely what it does or when to use it. The user wants to understand it well enough to know when and how to apply it.
</context>

<task>
**Explain chain-of-thought prompting in four parts:**

1. What it is: the technique described precisely, including the mechanism by which asking a model to show its reasoning changes the quality of its answer
2. When it helps: the specific task types where chain-of-thought reliably improves output, with a real example of a problem type that benefits
3. When it does not help: the task types where chain-of-thought adds noise without improving quality, or where it can introduce errors it would not otherwise make
4. How to apply it: two specific formulations that elicit chain-of-thought reasoning, and one note on how to combine it with other prompting strategies
</task>

<output_format>
- Four numbered parts, one to two paragraphs each
- Example prompt formulations: in code or quote blocks
- Total: under 450 words
- Tone: clear and technically accurate without being academic
</output_format>