top of page

Trust Is the New Usability

Writer: Magdalena Rut
Magdalena Rut
Sep 18
4 min read

Every UX conversation in 2026 seems to circle back to the same problem: people don't trust AI outputs, and that mistrust is becoming one of the biggest design challenges of the decade. Progressive disclosure, confidence scores, explainability layers — a whole vocabulary has sprung up around getting users to believe what a model tells them.


I've been sitting with this conversation for a while now, because it isn't new to me. It's the exact problem I've been designing around for years in industrial data platforms — long before "AI trust" was a panel topic.


The problem never was the interface

Early in my time working on discovery for industrial clients, I sat with engineers and operators who had a dashboard sitting right in front of them — clean, well-organized, technically accurate — and they still pulled out a spreadsheet, or picked up a phone, or walked the floor to check for themselves. The dashboard wasn't wrong. It just hadn't earned the right to be believed yet.

That's a distinction most design conversations skip past. We tend to treat trust as a downstream effect of good usability — make the interface clear enough, and trust follows. In high-stakes environments, it's the other way around. An operator won't even engage with a well-designed interface if the numbers behind it feel disconnected from what they know to be true on the ground. Usability without trust just makes it easier to ignore the tool faster.

What we found in that kind of discovery work, over and over, wasn't a UI complaint. It was some version of: "I don't know where this number came from, so I don't act on it." Data pulled from five different systems, contextualized into one dashboard, still gets treated as suspect if the person looking at it can't trace it back to something real. Contextualization is a data engineering achievement. Trust is a design problem, and it's a separate one.


Why AI makes this sharper, not new

AI doesn't introduce the trust problem — it raises the stakes on a problem that was always there. A stale sensor reading is something a domain expert can usually catch and reason about; a model's recommendation often can't be sanity-checked the same way. The person looking at it doesn't have the same intuitive handle on why the system is saying what it's saying.

That's why "communicating uncertainty without alarming the user" has become such a live design question this year. But in a control room, or on a plant floor, alarming someone isn't just a UX inconvenience — an operator who distrusts a signal and ignores it, or one who overtrusts a confident-looking number and acts on it wrongly, both carry real operational consequences. The design bar for AI-assisted decisions in these environments was always higher than in consumer software. AI just made that gap impossible to design around casually.


What design actually has to do about it

A few things I've come to believe, from sitting with this problem directly rather than theorizing about it:


Show provenance, not just output. A number without a traceable source is a number people will quietly discount, no matter how it's styled. The interface needs to make it easy to ask "where did this come from" and get a real answer, not a tooltip that restates the number.


Design for the skeptical expert, not the trusting novice. Most usability heuristics assume a user who wants to be guided. Domain experts want to be convinced. That changes what "good" looks like — density and traceability matter more than simplicity for its own sake.


Uncertainty is information, not noise. Hiding a model's confidence to keep the interface clean doesn't make the system more trustworthy — it just moves the discovery of that uncertainty to a worse moment, usually after something's gone wrong.


Trust is earned over repeated, small correct moments, not one polished screen. This is the part that's easy to forget in a portfolio review and impossible to forget in the field. Nobody trusts a system because it looks trustworthy. They trust it because it's been right enough times, in ways they could verify, that the verifying stops feeling necessary.

None of this is really about AI. It's about what happens when you hand someone a number and ask them to act on it without giving them any way to check your work. AI just made that a mainstream design conversation instead of a niche industrial one — which, honestly, is a good thing. It's a problem worth taking seriously either way.




Sources & Further Reading

On trust as the central AI-UX problem of 2026

On explainability, confidence signals, and uncertainty communication

On designer judgment vs. AI output ("AI as a junior designer")

On adaptive interfaces and designing for different expertise levels

Background on Nielsen Norman Group (for a bio/credibility note if needed)

 
 
 

Comments

bottom of page