Why Most People Still Don’t Trust AI and What Companies Can Do Differently

by Heather Breslow, Senior Director of Research at Superhuman

There’s a well-documented pattern in psychology: The more we encounter something, the more we tend to like it. Songs grow on you. Faces you recognize feel more trustworthy. Familiar brands feel safer.

When Pew Research identified familiarity as the strongest predictor of positive AI sentiment, the finding fit neatly into this framework. The industry drew a natural conclusion: If we can get more people using AI, the skepticism will give way to enthusiasm.

But that pattern has a condition that often goes unmentioned: It holds when the encounters themselves are neutral or positive. However, when experiences are consistently frustrating or disappointing, repeated exposure doesn’t warm people up. Instead, it deepens their skepticism. What the broader data shows is that people with the most AI experience are becoming more critical, not more enthusiastic.

Familiarity is reasonably good at building comfort, the feeling of knowing what to expect. But, it doesn’t reliably build trust, which has a higher bar: believing something will come through when the stakes are high. You can be completely at ease  with your best friend, but if they always run late, you aren’t going to trust them to get you to the airport on time. The tasks that people are actually being asked to use AI for—writing in their voice, analysis under their name, messages to people they care about—require trust over comfort.

AI’s most experienced users are becoming its harshest critics

Gen Z has the highest rates of AI adoption of any generation. They are also becoming more skeptical the more they use it. According to Gallup, in a single year, excitement about AI dropped 14 percentage points among Gen Z users while anger rose 9 points. The same study found that adoption plateaued even as negative sentiment grew.

This pattern holds across the broader American public. Pew Research found that 50% of U.S. adults now say AI makes them more concerned than excited, up from 37% four years ago. A Quinnipiac poll found that 55% believe AI does more harm than good in their daily lives, up 11 points in under a year.

The workplace data tells the same story. A study from WalkMe found that 80% of enterprise workers were either actively bypassing or simply not using their company’s AI tools, even as organizations spend an average of $54.2 million per year on AI transformation. Gartner put the share of employees actually realizing AI’s full potential at 8%. These are people who encounter AI every day and still don’t rely on it for their work.

Behind AI skepticism are distinct, coherent concerns

When we talk about AI hesitancy in research and product circles, we tend to treat it as a single thing: a psychological barrier to be lowered or a knowledge gap to be filled. Our research tells a more nuanced story: The reluctance people feel is a trust problem, and it comes from several distinct and coherent places.

Psychological ownership

In our internal research, people drew sharp lines around the work they weren’t willing to delegate to AI: performance reviews, client-facing writing, personal messages. One participant, a social marketing manager in her late twenties, described using AI to draft a message to her brother and sister-in-law, then deciding against it: “I felt in my heart I should just try to write this myself.” Psychological ownership is the sense that certain work is an extension of who we are—it carries our voice, or our care for someone. Her hesitation reflects what the act of writing means regardless of what the technology can do.

Learning on your own terms

Gallup found that 38% of Gen Z believed AI would do more harm than good for creativity; 42% said the same for critical thinking. In our research, students described deliberately avoiding AI when the goal was to understand something. One put it plainly: “Sometimes it just gives me the answer when I don’t want the answer.” Psychologist Robert Bjork’s research on desirable difficulties showed that the struggle of working through a problem was exactly what makes understanding stick. Getting the answer handed to you short-circuits that process. Students who opt out of AI in learning and creative contexts are protecting an experience they know has value.

Calibrated skepticism

Quinnipiac found 76% of Americans trust AI-generated information only some of the time or hardly ever, including people who use AI every day. In our research on knowledge workers, trust in AI-generated writing ranked near the bottom of all behavioral measures we track, even among daily active users. Calling this distrust misnames it. What the data shows is closer to calibration—an accurate read on the technology, built from firsthand experience with how often it gets things wrong. Among people who use AI regularly, skepticism about its outputs is what you get when familiarity outpaces reliability.

AI products that earn trust center the person, not the capability

So what does it actually take to earn trust from people who have good reasons to withhold it?

There was an account manager in one of our internal focus groups who said: “The stuff that we present to leadership, I need this to be one hundred percent filtered by me. I would not be comfortable putting that in front of leadership with my stamp of approval on it.” Her hesitancy is a precision instrument. Products that treat it as friction to be reduced will, to put it simply, not get used.

In our research on knowledge workers, the single most agreed-upon statement was that people saw AI as a tool that enhances rather than replaces their abilities. That framing holds across comfort level: Among hesitant users, it reduced the sense of threat. Among enthusiastic ones, it accurately described the experience they value most.

Trust means accepting some degree of vulnerability. You’re relying on something whose limitations you can’t fully audit from the outside. It develops through repeated low-stakes moments where people can see how something performs before they have to rely on it for something that matters.

Building products with these qualities—early interactions in which the output is visible and easy to evaluate; controls that make it simple to step back; tools specific enough to the work that you can see what the AI actually did—is how you create the conditions for trust to grow. That is work worth doing.