A company is building a financial analysis tool powered by an AI model. The model sometimes provides risky advice that doesn’t align with business guidelines. To prevent this, they want to set restrictions on the types of responses the model can give. Which prompt engineering best practice would help you establish these restrictions?