Which statement about chain-of-thought prompting is correct?

• A. It always improves factual accuracy.
• B. It eliminates the risk of hallucination.
• C. It can improve accuracy on multi-step tasks but may reveal reasoning steps.
• D. It shortens outputs.

Answer: (C)

Chain-of-thought prompting tests how guiding a model to show its step-by-step reasoning can help with problems that require several moves or logical deductions. When a task is multi-step, breaking it down and laying out intermediate steps lets the model examine its own logic and catch mistakes before reaching a final answer. That can raise accuracy on those kinds of tasks because errors are more likely to be spotted and corrected in the process.

At the same time, making the reasoning visible means the model’s internal thought process is exposed in the output, which has safety and privacy implications and can be undesirable in many settings. Outputs also tend to be longer when reasoning steps are included, and on simple or factual tasks the extra steps may not help and can even mislead. It doesn’t necessarily eliminate hallucinations—the model can still produce incorrect steps or made-up information within the reasoning chain.

So the best statement captures both the potential gain in accuracy for multi-step tasks and the fact that those reasoning steps may be revealed.