Guidelines/Conversation design

More help

This section provides resources and reference materials to support your conversation design work. These tools will help you understand key terms, answer common questions, and improve your conversational experiences.

FAQs

Here are answers to common questions about designing both scripted and generated conversational experiences.

Scripted experiences follow predefined paths, while generated experiences (powered by LLMs) create responses based on what the user says. Generated experiences are more flexible but require stronger error handling.

Use models that recognize multiple intents in one query. For generated experiences, train the assistant on diverse data and fine-tune responses with post-processing.

Personalize interactions using user details like names or preferences. Keep the assistant’s tone and persona consistent with the brand.

Set up clear fallback responses when the system doesn’t understand. For generated experiences, ask clarifying questions to guide the user.

Use emojis sparingly and only when they match a user's joyful tone. Make sure they enhance the message without reducing professionalism.

For scripted experiences, keep responses under 200 characters or 3 lines. For generated experiences, aim for under 300 characters or 4–5 lines, unless more detail is needed.

Give users ways to clarify their input or choose another option. Scripted experiences should use predefined fallbacks, while generated experiences should offer adaptive follow-ups.

Use the same core structure but adapt for each medium. Keep voice responses natural and concise, while text should be scannable. Use implicit confirmations for voice and explicit confirmations for text. Add audio cues for voice and visual formatting for text to improve clarity.

Use buttons for quick selection, images for clarity, and structured layouts for easy scanning. Introduce them for complex information, confirmations, or accessibility but avoid clutter and ensure alternatives for screen readers.

Glossary

These terms explain important concepts in conversation design and help in building better conversational experiences.

  • Connection words: Words that link ideas together, like "however," "first," "then," and "because." They help users understand how different parts of a message relate to each other.
  • Contextual awareness: Using past interactions and current context to give relevant, personalized responses.
  • Cooperative principle: The idea that people in a conversation work together to understand each other. This includes being truthful, relevant, clear, and giving the right amount of information.
  • Error recovery: Also known as contextual repair, this is how the assistant handles misunderstandings by asking clarifying questions or using fallback responses.
  • Fallback: A response used when the system doesn’t understand the user, helping guide them to clearer options.
  • Goal-first structure: Putting what the user wants to achieve before how they should do it. This makes instructions easier to follow.
  • Hallucination: When AI generates incorrect but plausible-sounding information.
  • Intent recognition: Identifying what the user wants from their input so the assistant can respond accurately.
  • Hidden meanings: "Information that's suggested but not directly stated. This lets conversations move quickly without spelling everything out.
  • Large language model (LLM): The AI that powers generated conversations, understanding complex queries and creating natural responses.
  • Multimodal design: Blending text, voice, and visuals into a seamless interaction.
  • Multi-turn interaction: Conversations spanning multiple exchanges.
  • Persona: The assistant’s personality and tone, shaped to match a brand and improve user experience.
  • Post-processing prompt: Backend instructions that refine responses to ensure they match design goals.
  • Prompt engineering: The practice of crafting inputs to shape AI responses.
  • Same pattern options: Presenting choices using the same grammatical structure (like starting all of them with verbs). This makes lists easier to read and remember.
  • Turn-taking: The natural back-and-forth of a conversation, ensuring the assistant and user don’t talk over each other.
  • User variable: A detail like a name or preference that makes responses feel more personal.

Resources

This list of resources will help you learn more about conversation design. It includes guides on design principles, testing tools, and best practices to create better user experiences.