Conversation Design Case Study (2025)

Improving AI Reliability Through Intent Mapping, Tone Control, and Safety-Aware Microcopy

Reference: Conversational AI Foundations

• McTear, M. (2023). The Conversational AI Handbook.
  https://arxiv.org/abs/2303.06745

• Google Research. Building Better Bots: Principles of Conversational AI.
  https://research.google/pubs/pub48120/

• Microsoft. Guidelines for Human-AI Interaction.
  https://www.microsoft.com/en-us/research/project/guidelines-for-human-ai-interaction/

• Rasa. Conversation-Driven Development.
  https://cdn2.hubspot.net/hubfs/4984639/Conversation-Driven_Development.pdf

📌 Overview

This project demonstrates how I design AI interactions that remain clear, safe, and trustworthy when user messages are vague or ambiguous.
I built a structured approach combining:

• Intent modeling
• Tone control
• Safety-aware microcopy
• Multi-turn flow patterns

The aim is to reduce model errors, prevent overconfidence, and support dependable user experiences in high-ambiguity contexts.

🧩 Problem

Users often write very short messages with missing context:

• “It won’t let me in.”
• “Fix this.”
• “I don’t see it.”
• “Why did it do that?”

These create three risks:

1. Misinterpreting intent
2. Overconfident or unsafe responses
3. User frustration when the system answers the wrong question

A consistent approach is needed to clarify intent early and manage tone responsibly.

---

✅ Approach

This system has three components:

1. Intent Decomposition

Each ambiguous message is broken into:

• Primary intent
• Secondary possibilities
• Disallowed interpretations
• Required clarification questions

2. Tone Control System

Three tones adapt based on risk:

• Plain Helpful
• Safety-Aware
• Expert-Concise

3. Safety Microcopy

Reusable phrases that expose uncertainty early and keep responses within privacy and compliance boundaries.

---

🗂️ Intent Maps (10 Examples)

Format:
Primary intent → Secondary intents → Clarifying question

1. “It’s not letting me in again.”

• Primary: login/access failure
• Secondary: password issue, MFA failure, permissions
• Clarification: Which step fails — password, code, or permissions?

2. “Why does it keep changing?”

• Primary: inconsistent behavior
• Secondary: UI personalization, layout shift, model variance
• Clarification: What’s changing — the layout, content, or settings?

3. “Fix this, it’s broken.”

• Primary: feature malfunction
• Secondary: error state, crash, unexpected output
• Clarification: What happened right before it broke?

4. “I didn’t ask for that.”

• Primary: incorrect system action
• Secondary: intent mismatch, autofill, hallucination
• Clarification: What did you expect to happen instead?

5. “Where did my stuff go?”

• Primary: missing data
• Secondary: wrong tab, filter, sync delay, wrong account
• Clarification: Were you viewing it in the same place you saved it?

6. “That’s not what I meant.”

• Primary: misunderstood request
• Secondary: vague phrasing, missing context
• Clarification: What outcome were you aiming for?

7. “Why is it taking so long?”

• Primary: performance concern
• Secondary: network latency, heavy processing, large file
• Clarification: Is it stuck loading, processing, or waiting?

8. “I don’t see it.”

• Primary: missing UI output
• Secondary: hidden panel, wrong screen, file not generated
• Clarification: Were you expecting a message, file, or interface element?

9. “Why did it do that?”

• Primary: unexpected system behavior
• Secondary: automation trigger, recommendation logic
• Clarification: Which action are you referring to?

10. “This doesn’t look right.”

• Primary: quality concern
• Secondary: outdated data, formatting issue, wrong version
• Clarification: What seems incorrect — the content, the structure, or the source?

---

🎨 Tone Modes