How you actually trigger it

There are two common ways to get a chain of thought, both relying on in-context learning, the model's ability to learn from the prompt itself.

The first, from this paper, is few-shot: you include a couple of example problems in your prompt where the worked-out steps are shown, not just the answers. The model sees that pattern and imitates it, producing steps for your new question too.

The second, from closely related work, is even simpler and is zero-shot: you just add a phrase like "Let's think step by step" to your question. That small nudge is often enough to make a large model start reasoning before answering.

flowchart LR
    A[Few-shot CoT<br/>show example problems<br/>with worked steps] --> M[Model reasons<br/>step by step]
    B["Zero-shot CoT<br/>add 'Let's think step by step'"] --> M