Framework
Trust Journey Framework
How designers and users build — and rebuild — trust with AI tools over time.
What is the Trust Journey Framework?
Quick Answer: The Trust Journey Framework maps the five stages designers and users move through when building trust with AI tools — from first use through to mature, resilient relationships. Developed by Riley Coleman from 240 designer interviews across IBM, Microsoft, CSIRO, and Atlassian.
- Trust is not binary — it evolves through specific touchpoints and experiences
- Five sequential stages: First Use, Early Usage, Regular Usage, Error Encounters, Continued Use
- Each stage carries a distinct primary risk and a set of targeted design actions
- Error encounters are the critical moment — how a system responds determines whether trust survives
- Forward-thinking design teams map trust journeys to identify moments where intervention matters most
A designer at a mid-sized consultancy adopts a generative AI tool. The first three sessions go well (Stage 1 win), she starts testing edge cases in week two (Stage 2), integrates it into her daily workflow by week four (Stage 3), then the tool produces an incorrect brief summary in a client meeting (Stage 4). Because the system immediately explained what went wrong and offered a corrective pathway, she returned to active use within a day — entering Stage 5 with calibrated, resilient trust rather than abandonment.
Understanding the Trust Journey
One of the most powerful reconceptions in this new paradigm is understanding that trust isn't a binary state but a relationship that evolves over time through specific touchpoints and experiences.
Forward-thinking design teams are now mapping trust journeys to identify critical moments that shape the user's confidence in AI systems.
Stage 1
First Use — First Impressions & Expectation Setting
Users approach new AI tools with what I call "curious scepticism": they want to believe, but they've been burned before. The first 3–5 interactions determine whether users will invest further exploration or immediately dismiss the tool.
Key Characteristics
- Users test basic functionality, compare outputs to their existing workflows, and look for quick wins that demonstrate immediate value.
- They're not yet emotionally invested, making abandonment costless.
Primary Risk
- Immediate dismissal if early interactions fail to show clear benefits. Many designers in my research abandoned tools within the first week if they didn't see immediate value.
Design Actions: Transparent Onboarding
- Capability Disclosure: Clearly communicate what the AI can and cannot do from the first interaction
- Limitation Transparency: Explicitly state boundaries and limitations upfront, not buried in settings
- Privacy Controls: Provide meaningful privacy settings with clear explanations of data usage
- Value Demonstration: Show immediate value with simple, successful first interactions (quick wins)
- Expectation Setting: Set realistic expectations about AI performance and learning curves
Stage 2
Early Usage — Testing Capabilities & Limits
Users who pass the first stage move into active testing mode. They are deliberately trying to break the system, comparing it against edge cases, and evaluating whether it fits their actual workflows (not just demo scenarios).
Key Characteristics
- Users develop mental models of what the AI can and cannot do.
- They start identifying specific use cases where it excels and scenarios where it fails.
- Trust builds through successful repetitions.
Primary Risk
- Inconsistent performance that contradicts the user's emerging mental model. "It worked yesterday, why not today?" is the death knell of trust.
Design Actions: Verification Support
- Progressive Disclosure: Gradually introduce complexity as user confidence grows, not all at once
- Low-Stakes Testing Grounds: Create safe spaces for experimentation where failures don't matter
- Visible Confidence Indicators: Show how confident the AI's decision is, so users know when to verify
- Feedback Collection: Actively solicit user impressions during early interactions to calibrate expectations
- Quick Wins: Design initial experiences that showcase reliability in controlled contexts
Stage 3
Regular Usage — Building Consistent Reliability
Users who successfully validate the tool move into routine integration. The AI becomes part of their daily workflow, trust deepens through consistent performance, and users begin championing the tool to colleagues.
Key Characteristics
- Users develop muscle memory around the tool, stop consciously evaluating every output, and begin relying on it for progressively higher-stakes tasks.
Primary Risk
- Over-reliance blindness. Users stop critically evaluating outputs because "it's always worked before." This complacency sets up the conditions for trust collapse when errors eventually occur.
Design Actions: Deepening Confidence
- Consistency Patterns: Establish predictable interaction rhythms that users can rely on
- Performance Dashboards: Show improvement metrics over time to reinforce trust with evidence
- Personalisation Evidence: Make adaptation to user preferences visible ("I learned you prefer...")
- Progressive Controls: Offer increasing customisation options as expertise grows
- Relationship Memory: Reference past interactions to build continuity and demonstrate learning
Stage 4
Error Encounters — Handling Unexpected Behaviours
This is the critical moment. Eventually, every AI system will produce an error, unexpected behaviour, or outright failure. How the system handles these moments determines whether trust survives or collapses.
Key Characteristics
- Users experience emotional betrayal ("I trusted you"), professional embarrassment (if the failure was visible), and cognitive dissonance (reconciling past success with current failure).
Primary Risk
- Complete abandonment. Users who experience a significant failure without effective system response rarely return to full trust levels.
Design Actions: Graceful Recovery
- Transparent Explanations: Explain what went wrong and why in plain language, not technical jargon
- Corrective Pathways: Provide clear actions to remedy the situation immediately
- Learning Signals: Show how the system is improving from this specific mistake
- Human Backstops: Offer access to human assistance for critical failures
- Apology Mechanics: Acknowledge failure with appropriate tone, not defensive explanations
Stage 5
Continued Use — Rebuilding After Challenges
Users who successfully navigate error encounters enter a new phase of mature trust. They understand the AI's limitations, have experienced recovery, and have developed a realistic, resilient relationship with the system.
Key Characteristics
- Users return to active use but with calibrated expectations.
- They maintain appropriate verification habits while still leveraging the AI's capabilities.
Primary Risk
- Permanent scepticism or anxiety. Some users never fully recover, using AI reluctantly and always waiting for the next failure.
Design Actions: Relationship Building
- Evolving Capabilities: Demonstrate new competencies based on usage patterns and feedback
- Trust Reinforcement: Periodically remind users of system reliability statistics and improvements made
- Collaborative Memory: Build shared history that deepens value over time ("We've worked together on 47 projects")
- Value Tracking: Help users see the cumulative benefit of their AI collaboration with concrete metrics
- Agency Calibration: Adjust autonomy levels based on re-established trust, gradually returning control
Related Frameworks
Frequently Asked Questions
What is the Trust Journey Framework?
The Trust Journey Framework is a model developed by Riley Coleman at AI Flywheel that maps the stages designers and users move through when building trust with AI tools. Based on 240 designer interviews conducted across organisations including IBM, Microsoft, CSIRO, and Atlassian, it identifies five stages: first encounter, active scepticism, controlled experimentation, calibration, and confident integration. It is used by design leaders to diagnose where their teams and users sit in the AI adoption curve.
How is the Trust Journey Framework different from other AI adoption models?
Most AI adoption models — including technology acceptance models and diffusion of innovation frameworks — focus on whether people adopt AI, not on how the quality of their trust develops over time. The Trust Journey Framework specifically maps the psychological and professional calibration process: how a designer learns when AI output can be relied upon, when it needs checking, and when it should be overridden. It is grounded in designer-specific research, not general technology adoption theory.
What is trustworthy AI design?
Trustworthy AI design is the practice of creating AI-powered products where users can develop appropriately calibrated trust — not blind reliance, and not unnecessary avoidance. It requires designing for transparency (users understand what the AI is doing), explainability (users understand why), control (users can override the AI), and graceful failure (the AI recovers from mistakes in ways that do not damage trust). The Trust Journey Framework provides a diagnostic lens for assessing whether an AI product supports or undermines this calibration process.
What is human-centred AI design?
Human-centred AI design is an approach that places human needs, values, and capabilities at the centre of AI system design — as opposed to designing AI-first and fitting humans around it. It draws on human-centred design (HCD) principles — research, iteration, and feedback loops — and applies them to the specific challenges of non-deterministic, probabilistic systems. The Trust Journey Framework is one tool within the broader human-centred AI design discipline.
How do I apply the Trust Journey Framework to my design team?
Start by mapping where each team member currently sits on the Trust Journey — which stage describes their current relationship with AI tools. Then design targeted interventions: for those in the scepticism stage, structured experimentation with low-stakes tasks; for those in calibration, peer review sessions that surface when AI outputs should and should not be trusted. The Trust Journey Framework is taught in depth in the Trustworthy AI for Designers course (ai-flywheel.com/course/trustworthy-ai, AU$750).
Why do users lose trust in AI products?
Users lose trust in AI products when the AI fails in ways they did not expect and were not prepared for — what the Trust Journey Framework calls the Confidence Cliff. This happens when AI adoption programmes skip the calibration stage, giving users access to powerful AI tools before they have developed accurate mental models of when those tools fail. Trustworthy AI design prevents this by designing explicit trust calibration into the product experience.