Skip to main content
View
project
Design Investigation #01

AI Conversation UX

Design Investigation #01 — a comparative UX investigation of ChatGPT, Claude, Gemini, and Perplexity, asking how conversational AI should build user trust.

9 min read

Timeline

4 Days

Role

UX Researcher

Field

Conversational AI

Products Studied

ChatGPT · Claude · Gemini · Perplexity

4

AI products compared

5

Opportunities mapped

8

Design principles

3

Friction tiers audited

01

Introduction

The way people interact with software is changing. For decades, digital experiences relied on buttons, navigation menus, and search bars. Today, users increasingly interact with AI through natural language — expecting systems to understand intent, retain context, and offer meaningful guidance.

That shift introduces a new kind of UX problem. Unlike traditional interfaces, conversational AI is expected to behave like a collaborator rather than a tool. Users aren't only judging whether an answer is correct — they're judging whether the interaction itself feels trustworthy, transparent, and reliable.

Each assistant approaches this differently. Some prioritize creativity and collaboration, others emphasize citations, long-form reasoning, or ecosystem integration — and these differences shape how confident, credible, and in control a user feels throughout a conversation.

This investigation looks at how four leading AI assistants — ChatGPT, Claude, Gemini, and Perplexity — design conversations, communicate uncertainty, and support user decision-making. Rather than comparing model performance, the study focuses on the interaction patterns that influence trust and overall experience.

Hero — all four AI assistants connected by a shared conversation network
02

Objective

The purpose of this investigation was to understand how conversational interfaces influence user trust — beyond the quality of the generated response itself. The research centered on five questions:

  • How do AI assistants establish credibility during a conversation?
  • How is uncertainty communicated to users?
  • What interaction patterns encourage continued exploration?
  • How do products support verification through citations or references?
  • Which design decisions improve long-term conversational experiences?

Rather than evaluating intelligence or benchmark scores, the study examines how interaction design affects confidence, transparency, and user decision-making.

03

Product Observations

03.1  Understanding the AI Ecosystem

Although conversational AI products often appear similar on the surface, each is built around a distinct product philosophy. Every interaction follows a comparable technical flow:

User Prompt → Intent Interpretation → Context Retrieval → Response Generation → Conversation Continuation

From a user's perspective, though, the experience extends well beyond response generation. Trust is built through the clarity of explanations, the visibility of sources, conversational memory, and a system's willingness to acknowledge its own limitations.

In other words, conversational UX is no longer only about generating answers — it's about designing interactions that reduce uncertainty while maintaining user confidence.

Diagram — the AI ecosystem flow

03.2  Comparing Product Philosophies

Across the comparison, each assistant showed a noticeably different conversational personality.

ProductPrimary Design PhilosophyBest Experience
ChatGPTCollaborative thinking partnerBrainstorming, writing, iterative problem solving
ClaudeThoughtful research collaboratorLong-form writing, analysis, nuanced reasoning
GeminiProductivity assistantGoogle Workspace integration, multimodal workflows
PerplexityResearch-first answer engineFact verification and citation-backed search

Rather than competing on the same strengths, these products optimize for different user expectations. ChatGPT encourages iterative collaboration, Claude prioritizes structured reasoning, Gemini integrates deeply into productivity workflows, and Perplexity leads with evidence-backed responses and visible citations.

03.3  Mapping the Conversation Journey

Across all four products, the conversation follows a similar high-level journey:

User Prompt → Intent Interpretation → Response Generation → Evidence & References → Follow-up Suggestions → Conversation Continuation → Memory & Personalization

Although the flow looks consistent on paper, each product differs in how much visibility it gives at every stage. Some encourage iterative refinement through follow-up prompts, while others prioritize quick, citation-backed answers. Those choices significantly shape how trustworthy the conversation feels.

Diagram — horizontal conversation journey

03.4  Personal Observation

While exploring each assistant, one insight consistently emerged: trust was not created by accuracy alone.

It was influenced by how transparently the system communicated uncertainty, referenced external information, and encouraged users to keep exploring rather than presenting every answer as definitive.

Personal observation

Products that openly acknowledged limitations and provided supporting evidence created stronger confidence than those that relied solely on fluent language.

04

Experience Audit

High Friction

Confidence often appears absolute

AI assistants frequently present uncertain or probabilistic information with the same visual confidence as verified facts. Without visible confidence indicators, users struggle to distinguish established knowledge from generated reasoning.

Impact

Increased risk of over-relianceReduced ability to assess information qualityPotential decision-making errors

Memory is often invisible

As conversational systems get more personalized, users rarely get a clear explanation of what's being remembered, or why it's influencing future responses. Research on conversational memory shows that different memory strategies materially affect personalization quality, and that inappropriate memory use can reduce relevance or expose sensitive context.

Impact

Reduced transparencyLower user controlIncreased privacy concerns

Evidence is inconsistent

Some assistants consistently surface citations, while others prioritize conversational fluency over sourcing. That inconsistency forces users to independently verify information, particularly for factual or research-oriented tasks.

Impact

Extra verification burden on the userInconsistent trust signals across productsHarder to use for research-critical tasks
Medium Friction
  • Context can degrade during very long conversations
  • Branching conversations remain limited
  • Multimodal workflows aren't always consistent across tasks
  • System status during reasoning is often unclear
Low Friction
  • Suggested follow-up prompts
  • Markdown formatting
  • File uploads
  • Voice interaction
  • Quick message editing
Board — High / Medium / Low friction
05

Key Findings

The investigation surfaced three recurring themes.

Transparency builds confidence

Users are more likely to trust systems that clearly explain where information originates.

Conversation should encourage exploration

The strongest experiences treated conversations as collaborative thinking sessions, not single-answer interactions.

Visible system behavior reduces cognitive load

Users make better decisions when they understand why a response was generated and what informed it.

Trust Framework — Accuracy · Transparency · Evidence · Control · Memory
06

Opportunity Matrix

01

Confidence indicators

High

Visually distinguish verified facts from probabilistic or generated reasoning, so users can calibrate trust per-response rather than per-product.

Impact HighEffort Medium
02

Source reliability labels

High

Surface lightweight signals on citation quality — not just that a source exists, but how strong it is.

Impact HighEffort Low
03

Memory timeline

High

Give users a visible, editable record of what's being remembered and why it's shaping the current response.

Impact HighEffort Medium
04

Conversation map

Medium

Let users see and navigate how a long conversation has branched, rather than scrolling linearly to reconstruct context.

Impact MediumEffort Medium
05

Explain response reasoning

Medium

Offer an optional, lightweight explanation of what informed an answer — without turning every response into a research paper.

Impact HighEffort High

These improvements focus less on increasing model capability and more on improving interaction transparency.

07

Design Principles

If I were designing the next generation of conversational AI, these principles would guide every interaction.

01

Design uncertainty instead of hiding it.

02

Show evidence before confidence.

03

Make memory visible and user-controlled.

04

Separate facts from generated reasoning.

05

Encourage collaborative refinement over one-shot answers.

06

Support recovery instead of assuming perfection.

07

Explain why an answer was generated.

08

Design trust as a feature, not an outcome.

08

Final Thoughts

The future of conversational AI won't be defined solely by larger models or faster responses. It will be defined by how effectively products help users understand, verify, and collaborate with AI.

Across all four products, the strongest experiences weren't necessarily the ones that generated the most impressive answers — they were the ones that made users feel informed, in control, and confident throughout the conversation.

As conversational interfaces continue to replace traditional search and navigation, transparency will become one of the most important responsibilities of product design. AI shouldn't only provide answers — it should help users understand why those answers deserve their trust.

More work

Other case studies

View all →
Hospitality Technology

Reservation Management System

Enterprise UX / 2023

Hospitality Tech

Point of Sale — Restaurant Operations

Mobile App / 2023

Hospitality Tech

Central Reservation Office

Enterprise UX / 2023

Have a project in mind?

I'm available for select freelance projects and full-time opportunities. Get in touch to discuss your project.