The performance of an AI assistant depends largely on the quality of its initial setup. In this sense, the prompt acts as the foundation on which the agent's behavior is built: it determines what it should know, how it should respond and how it should interact with customers.
At Prometheo we recommend following a series of best practices to design prompts that are clear, consistent and scalable. Below, we present some key guidelines.
1. Start with clear context
Every prompt should start with a context section. There you describe the company, its products or services, the communication tone and the assistant's main objectives. A well-defined context ensures the AI understands the framework in which it should operate. For example:
- “We are a company dedicated to selling laminate flooring and porcelain tiles…”
- “The assistant should respond briefly, professionally and warmly…”
2. Break it into sections with headers
To structure the prompt and make it easier for the model to understand, we recommend dividing it into sections using headers like ## Products, ## Pricing policy, ## Sales procedure, etc. This organization helps the assistant locate the relevant information more easily and avoids repetition.
3. Specify style and tone instructions
The prompt should clarify how the assistant is expected to respond:
- Message length (brief, detailed, formal, informal).
- Whether or not to use emojis.
- Language or regional variants (neutral Spanish, Argentine, Mexican, American English, etc.).
An example: “Use a neutral, formal tone. Keep responses under 50 words.”
4. Include interaction examples
Question-and-answer examples work as a practical guide for the model. You can indicate frequent cases, customer objections or specific scenarios. For example:
- Customer: “Do you have 60x60 gray porcelain tile in stock?”
- Assistant: “Yes, we have 60x60 gray porcelain tile available. Would you like me to prepare a quote?”
5. Define what it should and shouldn't do
A good prompt clarifies not only what the assistant should answer, but also what it should avoid. This reduces errors and ensures consistency in the customer experience. Examples:
- “Don't make up information that isn't in this prompt.”
- “Don't share long lists; instead, ask questions to better understand the need.”
6. Keep it updated and review it periodically
The business evolves, and the prompt should reflect those changes. Reviewing the content from time to time ensures the information stays accurate and the assistant remains aligned with the commercial strategy.
Conclusion
A well-designed prompt is the foundation of a successful AI assistant. By offering clear context, organizing information into sections, defining the tone, including practical examples and setting limits, companies ensure Prometheo provides more precise and relevant responses.
Investing time in writing the prompt not only improves the quality of attention, but also boosts the impact of all the integrations and automations available on the platform.