Style, Tone, and Persona

Imagine a tech startup that has just launched an innovative, user-friendly software product. Their brand is built on being modern, approachable, and helpful. However, when a new user interacts with their AI-powered support bot for the first time, they are met with responses that are robotic, overly formal, and cold. The bot provides accurate information, but the user experience feels impersonal and fails to meet their expectations. The experience doesn’t align with the brand’s intended tone or interaction style.

This scenario highlights a critical aspect of building AI applications: the way a model communicates can be as important as the content it produces. An AI’s voice, encompassing its style, tone, and persona, is not an accidental byproduct. It is a critical, engineerable component of the user experience. A well-crafted voice can build trust, enhance brand identity, and make an application truly delightful to use. A mismatched voice can destroy all of those things.

This lesson extends beyond the logical structure of prompts to encompass the communication and behavioral aspects of model output. We’ll define the concepts of style, tone, and persona, and cover practical techniques for controlling them to match application requirements.

Defining the concepts: Style vs. tone vs. persona

Before we can effectively control the voice of our AI, we must build a shared vocabulary and understand the distinct meanings of these three related concepts. They are the primary levers we will pull to shape the AI’s output.

• Tone: Tone refers to the emotional attitude or feeling conveyed by the text. It is the vibe of the response and dictates how the user feels when they read it. The tone can change from one response to the next based on the context. For example, a support bot’s tone might be empathetic when a user is frustrated but celebratory when their issue is resolved.

  • Examples: Formal, empathetic, witty, urgent, optimistic, professional, ...