Skip to main content

Step 7: Preference Discovery - Research & Best Practices

Last Updated: December 2024 Purpose: Industry research and best practices reference for AI-powered conversational preference discovery

This document captures research on AI-powered conversational preference discovery systems. It serves as a reference for what excellence looks like in this space, independent of our current implementation state.


Table of Contents

  1. Industry Best Practices
  2. Competitor Analysis
  3. Academic Research Insights
  4. Feature Checklist
  5. Sources

Industry Best Practices

1. Conversational UX Fundamentals

Source: Smashing Magazine - Conversational AI Design Guide

PrincipleDescriptionPriority
Design for user intentFocus on what user wants to achieve, not UI tasksP1
Context as goldRemember, reuse, and verify prior informationP1
TransparencyShow what, why, and howP1
User controlConfirm, undo, cancel, override capabilitiesP1
ConsistencyReduce friction with predictable patternsP2
GuardrailsSafe boundaries and ethical defaultsP2

2. Interaction Patterns

Source: WillowTree - 7 UX/UI Rules for AI Assistants

PatternDescriptionPriority
AI proposes, user approvesRecommendations with user confirmationP1
Proactive assistanceAI anticipates needs, user can overrideP2
Multi-turn planningBreak down goals with check-insP1
Mixed-initiativeBoth AI and user can drive conversation flowP2
Loops over linesAllow interruptions and returns to previous topicsP2

3. Visual UI Elements

Source: AIMultiple - Conversational UI Best Practices

ElementPurposePriority
Quick-reply buttonsTap instead of type for common optionsP1
Suggestion chipsShow available options visuallyP1
Typing indicatorsShow "Searching...", "Thinking..." statesP2
Visual formattingLists, bold, icons for readabilityP2
Confidence cues"I'm 80% sure...", "Did I understand correctly?"P2

4. Personalization Approaches

Source: Survicate - User Onboarding Design

ApproachDescriptionPriority
Persona-based flowDifferent question paths for different user typesP1
Experience filteringShow relevant options only based on user contextP1
Adaptive UIAdjust interface based on user preferences/behaviorP2
Progress indicatorsShow completion status and remaining stepsP1

5. Feedback Mechanisms

Source: Mind the Product - UX Best Practices for AI Chatbots

MechanismPurposePriority
Thumbs up/downQuick feedback on AI responsesP2
Regenerate buttonGet alternative response if current one isn't helpfulP2
Edit responsesModify AI's understanding of user inputP2
Explicit confirmation"Did I understand correctly?" for ambiguous inputsP1

6. Confidence Scoring Best Practices

Source: OpenReview - PrefEval

PracticeDescriptionPriority
Multi-factor scoringCombine linguistic, semantic, and contextual signalsP1
Transparent reasoningExplain why confidence is high/lowP2
Clarification triggersLow confidence should prompt follow-up questionsP1
Alternative interpretationsTrack other possible meanings of user inputP2
Conflict detectionIdentify when preferences contradict each otherP1

7. Preference Learning

Source: arXiv - On the Way to LLM Personalization

PracticeDescriptionPriority
Learn from past behaviorUse historical preferences to pre-fill or suggestP2
Smart defaults for new usersResearch-backed defaults based on user type/goalP2
Preference stability trackingKnow which preferences are consistent vs volatileP3
Cross-session memoryRemember preferences across multiple interactionsP2

8. Open-Ended Capture

Source: ACM - Generating Usage-related Questions

PracticeDescriptionPriority
Entity extractionParse free-form text into structured categoriesP1
Impact classificationTag extracted info as critical/important/nice-to-haveP2
Plan generation notesGenerate actionable notes for downstream systemsP2
Prompt hintsProvide examples of what to share in open-ended responsesP2

Competitor Analysis

Fitness App Comparison

FeatureFitbodWHOOPTrainerizeBest Practice Target
Onboarding StyleForm wizardSensor-firstForm wizardAI conversational
Preference CaptureCheckboxes, dropdownsAutomatic from biometricsFormsNatural language + structured
PersonalizationAlgorithm-drivenData-drivenCoach-drivenCoach persona + context
User EducationMinimalMetrics-focusedVaries by coachDeep "why" explanations
Question OrderFixedN/AFixedTier-based with AI flexibility
AdaptationFrom workout historyFrom biometricsManual coach adjustmentReal-time conversation

Fitbod's Approach

Source: Fitbod Blog

"Fitbod begins with a smart onboarding process. It collects details like fitness goals, preferred workout style, equipment available, and experience level."

Key Features:

  • Form-based onboarding (not conversational)
  • ML generates dynamic plans that evolve
  • Learns from workout history and recovery patterns
  • Progressive overload intelligence

Opportunity: Provide educational context during preference collection, explaining why each choice matters. Collect data AND teach simultaneously.

WHOOP's Approach

Source: BuddyX Theme - AI Tools for Fitness Coaching

"WHOOP digs deep into your body's data to tell you when you're ready to train hard or when you need to chill out."

Key Features:

  • Passive data collection via wearable
  • HRV, sleep, and strain analysis
  • Recommendations based on recovery metrics

Opportunity: Capture subjective preferences that sensors can't detect (exercise preferences, training philosophy, lifestyle constraints).


Academic Research Insights

LLM-ConvRec Architecture

Source: GitHub - D3Mlab/llm-convrec

"LLM-ConvRec maintains an explicit internal state that tracks user preferences and constraints. This structure improves response consistency, memory retention, and control."

Four-Stage Pipeline (Recommended Architecture):

  1. Intent Classification - Determine if user is asking, answering, or commenting
  2. State Update - Track answered preferences explicitly
  3. Action Selection - Decide next question based on tiers/priority
  4. LLM-based Response Generation - Generate conversational coach message

Conversational Recommender Systems

Source: ACM - Conversational Style Impact

"The success of these systems is heavily influenced by the preference elicitation process. While existing research mainly focuses on what questions to ask, there is a notable gap in understanding what role broader interaction patterns—including tone, pacing, and level of proactiveness—play."

Key Factors to Consider:

  • Tone: Personality and communication style
  • Pacing: Question ordering and density
  • Proactiveness: Anticipating user needs vs waiting for explicit input

Preference Elicitation Strategies

Source: ACM - Preference Elicitation Strategy for CRS

Key Findings:

  • Start with open-ended questions, gradually elicit specifics
  • Trade-off between conversation efficiency and information accuracy
  • Usage-based questions are effective: "Are you looking for X that is great for Y?"
  • Present trade-offs explicitly: "If you choose X, you get Y but it means Z"

LLM Preference Following Challenges

Source: OpenReview - PrefEval

"In zero-shot settings, preference following accuracy falls below 10% at merely 10 turns (~3k tokens) across most evaluated models."

Mitigations Required:

  • Explicit state tracking (don't rely on LLM memory alone)
  • Confidence scoring to catch misunderstandings
  • Low-confidence should trigger clarification
  • Store preferences in database, not just conversation history

Proactive Suggestions

Source: arXiv - COMPASS: User Preferences with Knowledge Graph

Key Concept: When a user answers one preference, surface related preferences proactively.

Example Relationships:

  • Training split → rest period recommendations
  • Deficit approach → cardio strategy suggestions
  • Intensity techniques → deload frequency implications

Feature Checklist

Use this checklist to compare implementation against research best practices.

Core Architecture

FeatureDescriptionPriority
Goal-specific question bankDifferent questions for different fitness goalsP1
Tiered question systemRequired → Important → Optional flowP1
Educational contextEach question explains "why it matters"P1
Trade-off presentationEach option shows impact/consequencesP1
Multi-preference extractionExtract multiple answers from single messageP1
Intent detectionDistinguish questions from answersP1
Coach personalizationGoal-specific messaging with personalityP1

Confidence & Extraction

FeatureDescriptionPriority
Multi-factor confidence scoring5+ factors: linguistic, semantic, contextual, etc.P1
Confidence thresholdOnly save preferences above threshold (e.g., 0.6)P1
Transparent reasoningExplain why confidence is high/lowP2
Alternative interpretationsTrack other possible meaningsP2
Conflict detectionIdentify contradicting preferencesP1
Clarification suggestionsGenerate follow-up for low confidenceP1

Open-Ended Capture

FeatureDescriptionPriority
Entity extractionParse free-text into structured categoriesP1
Category taxonomyInjuries, equipment, lifestyle, exercises, etc.P1
Impact classificationCritical vs important vs nice-to-haveP2
Plan generation notesActionable notes for plan generationP2
Prompt hintsExamples of what to shareP2

User Control & Revision

FeatureDescriptionPriority
Undo capabilityRevert previous answerP1
Revision detectionDetect "actually I meant..." patternsP1
Inline revision handlingHandle corrections mid-conversationP1
Preference summaryShow what's been answeredP2

Learning & Prediction

FeatureDescriptionPriority
Learn from past plansStore and reuse preferencesP2
Predict for new plansSuggest based on historyP2
Smart defaults (new users)Research-backed defaults by goal/experienceP2
Workout feedback learningAdjust based on completed workoutsP3
Preference stability scoringTrack consistent vs volatile preferencesP3

Feedback & Improvement

FeatureDescriptionPriority
Thumbs up/downQuick feedback on responsesP2
Regenerate capabilityGet alternative responseP2
Feedback storageStore for AI improvementP3
Regeneration trackingTrack original vs regeneratedP3

Proactive Suggestions

FeatureDescriptionPriority
Preference relationshipsMap related preferencesP2
Suggestion generation"Based on X, you might want Y"P2
User engagement scoringAdjust proactiveness by engagement levelP3
Suggestion filteringOnly show high-confidence suggestionsP2

Visual/UI Elements

FeatureDescriptionPriority
Quick-reply chipsClickable option buttonsP1
Progress visualizationTier completion indicatorsP1
Typing indicators"Thinking..." statesP2
Markdown renderingFormat coach messagesP2
Streaming responsesToken-by-token displayP3

Advanced Features (Backlog)

FeatureDescriptionPriority
Knowledge graphLink preferences to exercise databaseP4
A/B testingTest question orders/phrasingsP4
Cross-user patternsLearn from aggregate user behaviorP4

Sources

Industry Best Practices

  1. Smashing Magazine - How To Design Effective Conversational AI Experiences
  2. WillowTree - 7 UX/UI Rules for Designing a Conversational AI Assistant
  3. AIMultiple - Conversational UI: 6 Best Practices
  4. Springs Apps - 10 Chatbot Best Practices In 2025
  5. Exotel - Conversational UX 101 Guide for 2025
  6. Mind the Product - Nine UX Best Practices for AI Chatbots
  7. Botpress - Conversational AI Design in 2025

Fitness App Research

  1. Fitbod - Best AI Fitness Apps 2025
  2. QuickPose.ai - Fitness App Trends 2024
  3. AppInventiv - 15 Use Cases of AI in Fitness Industry
  4. BuddyX Theme - Best AI Tools for Fitness Coaching

Academic Research

  1. GitHub - D3Mlab/llm-convrec: LLM-based Conversational Recommendation Architecture
  2. OpenReview - Do LLMs Recognize Your Preferences? (PrefEval)
  3. ACM - Should We Tailor the Talk? Conversational Styles in Preference Elicitation
  4. ACM - Preference Elicitation Strategy for Conversational Recommender System
  5. ACM - Generating Usage-related Questions for Preference Elicitation
  6. arXiv - COMPASS: Unveiling User Preferences with Knowledge Graph and LLM
  7. arXiv - On the Way to LLM Personalization: Learning to Remember User Conversations

Onboarding Design

  1. Survicate - User Onboarding Design: How to Get it Right
  2. UserPilot - AI User Onboarding: 8 Real Ways to Optimize
  3. Landbot - Onboarding Chatbot Guide
  4. Specific.app - Great Questions for Chatbot Onboarding

This document should be updated as new research emerges. It serves as a reference standard, not an implementation tracker.