Conversation design

Who this is for

This is for anyone involved in designing conversational experiences, including conversation designers, UX designers, product managers, and engineers.

It provides best practices for both scripted and AI-generated conversations, ensuring clear, effective, and on-brand interactions. Whether you're creating rule-based flows or AI-powered assistants, this guide helps maintain consistency, improve usability, and enhance user satisfaction.

How it will help

Improve interactions: Help users easily move through conversations with simple, useful messages.

Create consistency: Keep the same style and persona, no matter where the conversation happens.

Enable growth: Expand conversational experiences to new environments and situations.

The basics

Conversation design helps people communicate with machines. It blends UX, linguistics, and human-computer interaction to help users complete tasks.

Here are the key aspects:

• Intent recognition: Understanding what users need.
• Interaction flow: Mapping the user's journey from start to finish.
• Contextual awareness: Using old chats to help with new ones.
• Language: Ensuring clarity, inclusivity, and proper grammar in responses.
• Voice and tone: Creating a natural, on-brand persona.
• Turn-taking: Managing the back and forth between users and the assistant.

The theory

How language works

Language is always changing and growing. We use it every day to share ideas, ask questions, and connect with others. When we design conversations for assistants, we need to understand how people naturally talk to each other.

People learn to have conversations from a very early age. Even before we know many words, we learn the back-and-forth rhythm of talking with others. These natural conversation patterns are what we want to use when designing how assistants talk.

The cooperative principle

When people talk to each other, they usually try to help the conversation move forward. A researcher named Paul Grice noticed this and called it the "cooperative principle." He found that most conversations should follow four simple rules:

1. Be truthful: Don't say things you know aren't true.
2. Give the right amount of information: Share what's needed—not too little, not too much.
3. Stay on topic: Keep your responses relevant to what's being discussed.
4. Be clear: Use simple language and organize your thoughts in a way that's easy to follow.

People follow these rules naturally, and they expect assistants to follow them too. When an assistant breaks these rules—like giving too much information or going off-topic—it makes the conversation feel unnatural.

Hidden meanings

Sometimes we say one thing but mean more. For example:

The second person isn't just sharing random information. They're actually suggesting, "You could go there to get food." This happens all the time in human conversations, and it helps us communicate quickly without spelling everything out.

Understanding these hidden meanings helps us design more natural conversations for assistants.

Core principles

These principles help make conversations easy, quick, and engaging, while meeting users' needs and maintaining a brand's identity:

• User-centered design: Put users' time and effort first.
• Clarity and brevity: Use simple, direct language.
• Consistency: Keep a recognizable persona.
• Error recovery: Give users ways to fix mistakes.
• Guided interactions: Make it obvious when and how to respond.

Types of conversations

Understanding the differences between scripted and generated conversations will help you choose the right approach for your use case.

1. Scripted

Rule-based conversation flows that follow predefined scripts and decision trees, ideal for handling predictable, repetitive tasks efficiently.

2. Generated

AI-powered conversations that create custom responses in real-time, designed to handle complex interactions and adapt dynamically to user needs.

Designing each type

While the same core principles apply, each approach requires a different design strategy.

Scripted design:

• Create clear, predefined user paths
• Plan for common user inputs
• Provide structured choices
• Use fallback options to handle errors

Generated design:

• Continuously understand what the user needs and why
• Adapt to a wide range of inputs
• Learn from past interactions to improve responses
• Handle errors dynamically with clarifying questions

Both approaches aim to guide users efficiently, but generated interactions offer more flexibility for handling complex, multi-turn conversations.

The persona

The assistant’s persona defines how it communicates, ensuring a consistent, natural, and engaging experience. A strong persona helps users feel comfortable and confident while interacting with the assistant.

Persona is made up of two key elements:

Voice

The assistant’s foundational personality—it stays the same across all interactions.

  A good assistant voice is:

• Consistent: Maintains the same personality no matter the scenario.
• Recognizable: Feels distinct yet aligns with the brand.
• Approachable: Communicates clearly and naturally.

💡 A friendly, professional assistant will always sound direct and supportive.

Tone

Adjusts based on the user’s situation. It ensures the assistant responds appropriately in different contexts.

  Common tone variations include:

• Welcoming: Friendly and warm for greetings.
• Supportive: Calm and understanding for troubleshooting.
• Neutral & informative: Clear and concise for technical guidance.
• Encouraging: Positive and motivating when acknowledging progress.

💡 The assistant should sound more empathetic when a user is frustrated but neutral when providing general instructions.

Creating personas

To design an effective persona, you’ll need to understand your users and their expectations.

• User demographics: Who will interact with the assistant?
• Use case and industry: What tasks will the assistant handle?
• Brand identity: How can the assistant reflect the company’s brand?
• Emotional expectations: What emotional state will users be in?

💡 A healthcare assistant should sound calm and empathetic, while a travel booking bot can be more upbeat and conversational.

Finding the voice

Identify the key personality traits that will shape the assistant’s voice.

Formal — Casual
"Hello, how can I assist you?" — "Hey! Need a hand?"

Serious — Playful
"Your balance is $500." — "Cha-ching! 🤑 You have $500. "

Concise — Descriptive
"Your flight booking is confirmed." — "Great news! Your flight is booked and ready to go."

Setting the tone

Define how the assistant will adapt based on the situation and user needs.

Here's a solid foundation for any assistant's tone:

Welcoming
Friendly for greetings: "Hi Alex, how can I help today?"

Supportive
Calm and reassuring for troubleshooting: "I know that’s frustrating. Let’s get it fixed."

Neutral and informative
Direct for technical tasks: "Go to Settings and select the Account option to update your details."

Encouraging
Motivational for progress: "Okay, we’re almost there. Just one last question.”