Designing trust: How UX can make AI products worth relying on

In 2023, two New York attorneys found themselves at the center of an unexpected scandal. Their legal brief — filed in an aviation case — included six citations to court cases that did not exist. The fabricated cases had been generated by ChatGPT, which the lawyers had used to help draft the document. 

When questioned by the judge during a sanctions hearing, they admitted they hadn’t verified the citations. Later, the court sanctioned both attorneys and their firm, calling the episode “an unprecedented circumstance” of lawyers relying on artificial intelligence without oversight.

The story became an instant cautionary tale about misplaced faith in artificial intelligence. It also highlights one of the biggest UX challenges we face right now: how to build trust in AI products.

The trust paradox: Why trust in AI products is so fragile

AI tools are increasingly embedded in everything, from self-driving cars to diagnostic software and personal assistants. They’re fast, confident, and often eerily human. But unlike humans, they don’t have intentions or empathy. Often when users say they “trust” an AI tool, what they really mean is that they rely on it.

And for product teams, that distinction shapes how users build trust in AI products in the first place.

“Trust makes relationships go round — whether it is between people, businesses, or the products we rely on,” says Dr. Janna Lipenkova in the UX Collective. “It’s built on a mix of qualities like consistency, reliability, and integrity. When any one of these breaks, the relationship cracks with it.”

Research from the Software Engineering Institute at Carnegie Mellon University defines trustworthy AI systems by measurable properties like:

• Validity 
• Reliability 
• Safety
• Security
• Fairness
• Transparency. 

In UX design, these technical factors are often translated into user-facing qualities that help people decide when it’s safe to rely on an AI system. These qualities include predictability, controllability, and clarity.

When those elements align, users feel confident enough to depend on the system — the foundation of trust in AI products. When they don’t, the relationship collapses. Either users under-trust (users abandon the tool) or they over-trust (users let it make unverified decisions).

In AI UX design, trust is not a given. It’s engineered through predictability, control, and clarity.

Escaping the “Trust Trap”

Designing trust in AI isn’t about making systems more humanlike. It’s about helping humans understand what’s humanly reasonable to expect from them. The goal isn’t blind faith. It’s calibrated confidence.

Researchers have long described the dangers of miscalibrated trust in automation: when users either rely too little on capable systems or too much on fallible ones.

At Standard Beagle, we call this dynamic the Trust Trap: a mismatch between user confidence and an AI system’s actual performance. When users fall into the Trust Trap, they either under-trust technology that could help them or over-trust algorithms that should still be questioned.

It’s also essential for helping teams build sustainable trust in AI products that users rely on daily.

The ChatGPT court case was a textbook example of over-trust. The attorneys believed the tool’s fluent tone signaled accuracy. It’s a design illusion sometimes called automation bias.

The opposite problem is equally costly. Consider a diagnostic AI that outperforms radiologists at detecting early-stage tumors but is ignored because clinicians find its reasoning inscrutable. Without clear feedback people can interpret, they hesitate to rely on it, even when it’s right.

For UX designers, the challenge is to act as the calibration mechanism between these extremes: creating experiences that continually communicate what the AI knows, what it doesn’t, and how confident it is.

The architecture of trust

Designing reliable AI means shifting from abstract notions of “trust” to a concrete framework of trustworthiness and reliance.

• Trust lives in the user’s mind: a psychological state based on expectation.
  
• Trustworthiness belongs to the system: the demonstrable qualities that make it reliable.
  
• Reliance is behavior: the act of using the system in real decision-making.
  

Designers can’t install trust directly into users. What they can design is trustworthiness: the evidence that convinces users it’s safe to rely.

That evidence often comes down to how information is surfaced through the interface. These include visual cues, feedback loops, and signals of reliability. When these are missing or misleading, trust erodes. When they’re clear and consistent, users can make informed judgments about when to lean on the AI and when to intervene.

How AI UX design builds user confidence

Explainable AI (XAI) and transparency: Making the invisible visible

Transparency is the cornerstone of building trust in AI. Industry leaders and researchers agree that people are far more likely to rely on a system when they understand how it works.

According to IBM, transparency means clearly communicating how and why an AI system reaches its conclusions. We’re not talking about exposing the code. Instead, it means making the reasoning accessible to users. Gartner’s AI governance framework also says that transparency is a prerequisite for responsible adoption. It helps users fine-tune their confidence in the technology.

The World Economic Forum puts it even more simply: trust in AI can’t happen without transparency. When organizations explain how data is used, how models make decisions, and what safeguards are in place, users get a clearer mental model of the system.

In UX design, that understanding translates into confidence. It’s not blind trust but informed reliance. This clarity is what ultimately strengthens trust in AI products.

Transparency happens on three levels:

1. Algorithmic transparency – revealing the logic, data sources, and processes behind the model.
   
2. Interaction transparency – showing how the AI uses input and feedback during live use.
   
3. Social transparency – acknowledging the ethical, societal, and fairness implications of its design.

Each level contributes to what users perceive as honesty. And that honesty is quickly becoming a legal requirement.

Under the EU’s Artificial Intelligence Act, providers of AI systems that interact with humans have to tell users when they are interacting with AI. Systems that generate or manipulate synthetic content (such as deepfakes) must mark that content as artificially generated or manipulated. The Act also requires transparency obligations for certain high-risk AI systems, including documentation and user-notification requirements.

Explainable AI (XAI) takes transparency a step further by answering why. Why was your loan application denied? Why did the system recommend this medication instead of another?

Techniques like LIME and SHAP allow designers to translate machine reasoning into human terms. A good explanation doesn’t bury users in math; it connects cause and effect. When users can see that cause-and-effect reasoning, trust in AI products increases significantly.

IBM defines explainability as helping people understand and trust AI outputs by making the reasoning behind results clear and accessible. Instead of just presenting data, effective explanations communicate insight. It becomes a transfer of understanding between the system and the user.

Designing for control, not surrender

AI products succeed when users feel in control… even if the system is autonomous. That sense of agency is the difference between collaboration and dependency.

In Human–AI Interaction research, users are shifting from traditional operators to orchestrators. That means they are guiding intelligent systems instead of issuing rigid commands. This framing stresses collaboration and oversight.

User agency has to be preserved. Microsoft’s Human-AI Interaction Guidelines say to give users ways to “pause, correct, or dismiss” AI actions at any time. Google’s People + AI Guidebook has a similar principle: “Support efficient correction.”

This design philosophy is visible in tools like Figma’s AI “assist” features or GitHub Copilot’s code suggestions. The AI proposes, but the human can always decide to trash it. Users can reject, rewrite, or refine outputs, which keeps the creative authority where it belongs.

A healthy AI relationship feels collaborative, not prescriptive. UX designer Dan Saffer argues that AI should act as a collaborator or co-author, not a replacement for human creativity.

The value of visible fallibility

Traditional UX tries to minimize error. But in AI design, showing how a system handles mistakes is often more important than preventing them.

Candace Wilson notes that transparency about uncertainty is key to maintaining trust. Users feel safer when AI systems acknowledge limitations rather than pretending to be infallible.

That’s why the most reliable AI interfaces visibly indicate uncertainty — tools like Google’s search experiments or Perplexity’s AI browser. Phrases like “I’m 70 percent confident” or “I might be wrong” help users gauge when to double-check.

Even simple transparency cues can be transformative. When AI systems provide source links or evidence for their claims (like the reference links now visible in ChatGPT) users get clearer context for evaluating accuracy and reliability.

Designers should treat every system failure as a moment to demonstrate integrity. A well-designed fallback, such as “I didn’t understand, can you rephrase?” keeps users in control. The system’s errors aren’t hidden. They are acknowledged and recoverable.

Ethics isn’t optional

No design can compensate for unethical AI. Bias undermines every other dimension of trust, whether in data or decision logic.

IBM’s Principles for Trust and Transparency identify fairness as one of five pillars of trustworthy AI, alongside explainability, robustness, transparency, and privacy. In practice, fairness starts with inclusive research and representative datasets. But it also extends to UI design: how results are framed, which defaults are presented, and whether users can flag unfair outcomes.

UX can’t fix systemic bias alone, but it can make bias visible. A feedback tool that allows users to report discriminatory behavior turns ethical accountability into a user-facing feature.

Patterns for designing trust

We’re starting to see “trust patterns” emerge across products. They’re the repeatable UX solutions that keep humans and algorithms in healthy balance:

• Label AI clearly – IBM’s Carbon for AI uses consistent “AI labels” to identify algorithmic outputs and link to more detail.
  
• Set expectations early – Google and Microsoft recommend onboarding that states what the AI can (and cannot) do.
  
• Communicate uncertainty – Confidence scores and probabilistic phrasing help calibrate reliance.
  
• Provide contextual explanations – Offer just enough reasoning, at the right time, with the option to learn more.
  
• Keep humans in the loop – Always allow overrides, corrections, or alternative workflows.
  
• Encourage co-creation – Present multiple options, not single answers, to foster exploration and critical thinking.

These aren’t cosmetic touches. They’re structural reinforcements for the bridge between user and machine.

Lessons from the road, the screen, and the clinic

Real-world examples show how trust design plays out in radically different contexts.

• Waymo’s self-driving taxis use transparent visuals to show passengers what the car “sees” on a dashboard display — things like other vehicles, pedestrians, and stoplights. This visualization of the AI’s awareness demystifies its decisions and builds passenger confidence.
  
• Netflix’s recommendation engine relies on clear explainability (“Because you watched…”) and feedback loops (thumbs up/down) to help users fine-tune results and, in turn, improve the model itself.
  
• Medical diagnostic AIs like Google’s DeepMind tools for eye disease detection combine visual heatmaps and confidence indicators to show clinicians why a diagnosis was made, keeping human judgment at the center.

Each case has the same principle: trust isn’t granted. It’s earned through visibility, agency, and accountability.

The metacognitive era of UX

For decades, UX design has prized frictionless experiences. But in AI, a little friction can be healthy.

When a chatbot asks a clarifying question instead of guessing, or when a recommendation tool shows confidence scores instead of pretending to be certain, it’s inviting the user to think. That’s metacognitive design — interfaces that encourage awareness, skepticism, and critical engagement.

In this era, the designer’s job isn’t to make AI invisible. It’s to make its reasoning legible.In the end, trust in AI products isn’t about making technology more human—it’s about designing systems that make humans more informed and confident.