Persona

On this page you learn about the Persona settings in EbbotGPT Configurations.

Persona

View and edit the Persona in your configs under EbbotGPT Configurations. Your persona is a way to give your chat agent instructions on who it is and how it should behave. Think of a persona as a prompt that directs the chat agent's behavior and answers.

• Chat agent's name and purpose

• Your organisation's name and a short description

• How the chat agent should write its answer

• Specific behaviour: "If asked about X answer Y"

We provide general guidelines because persona design isn't an exact science. With many unique parts such as desired behavior, uploaded knowledge, and tools, finding the perfect balance usually takes some trial and error.

General Guidelines

• Define the interaction style: Who will be talking to the AI agent, employees at your company, customers or someone else? Define if the AI should act as a peer, expert, or facilitator.

• Define who the AI represents: By specify what the bot represents it will have the context to understand the scope of appropriate topics and boundaries. For example, not to recommend competitors or talk about topics unrelated to your organisation.

• Persona length: Keep your persona short and direct. Overly long prompts can distract the AI, making it lose focus on its core tasks. Using headings is one way to manage a long persona; they act like a map, helping the model organize your instructions and follow your rules more reliably.

• Markdown formatting: Clear formatting makes your instructions much easier for the AI to follow. Using Markdown headers and bullet points helps the model see the structure of your persona, separating your identity from your specific rules. This organization acts like a guide, making it much more likely that the AI will stay on track and give you the results you want.

• Give your instructions more weight: To make an instruction more impactful, focus on relevance rather than formatting. Instead of bolding text, use clear trigger-based logic, linking a rule directly to a specific situation or question. By placing instructions under a dedicated heading, you make it a "no-brainer" for the AI to recognize exactly when that information applies.

• Avoid using numbered word or character answer length: By writing, "keep your answers below 150 words" it forces the AI to waste its reasoning on counting instead of focusing on the actual quality of the answer. Instead of strict limits, use descriptive phrases like "keep it concise" or "provide a brief summary" to guide the response length.

• Avoid mixing languages: Mixing languages can confuse the model, making it harder for it to consistently reply in the user's language.

Prompting Examples

• Provide alternative responses: When instructing the chat agent not to respond a certain way, also specify how it should respond instead.

• Concise and Clear Instructions: Shorter, more focused instructions are generally more successful and have less risk of negatively affecting other parts of your AI Agent.

• Use explicit phrasing: For direct instructions, use clear commands like: "Refuse to answer X - Always say Y."

  • It's recommended not to give the LLM an exact sentence to output. Instead, provide instructions on what it should convey. For example, instruct it to 'state that you do not have the information' instead of saying 'answer with: I do not have information about that.'

• Provide exact phrasing in quotes: To guide the AI agent's responses precisely, use quotation marks. For example: "If X happens, always reply: 'I can't help with that, have you tried xyz?'"

  • Using exact phrasing in quotes can cause the LLM to output that exact text, overriding its ability to dynamically translate the response into the user's language.

For more information and best practice regarding the persona and LLM prompts visit the link below:

Persona & prompt guidelines