By Pam Bohline
We’re in the Wild West of artificial intelligence.
From small startups to tech giants, companies are in a frenzied sprint to be first to market, investing some $30 to $40 billion into launching new AI products and features at record pace. All eyes are on the prize: harnessing AI’s immense potential, demonstrating technological prowess, and capturing market share.
But beneath all this enthusiastic innovation lies a growing paradox: while AI capabilities are skyrocketing, user trust is dropping and adoption is slowing. We are developing products faster than people can understand or use them, much less integrate them into their daily lives. The result is a user base that is understandably overwhelmed, wary, and fundamentally unprepared for the technological leap being thrust upon them. This gap between rapid deployment and user trust threatens the deployment, adoption, and long-term success of AI products and experiences.
The "solutionism" trap: Building products for needs that don’t exist
Instead of identifying genuine user pain points and then exploring how AI might address them, many companies start with the impressive AI technology itself ("Look at our cool new large language model!") and then reverse-engineer a "need" for it. There’s a reason why the so-called “smart hairbrush” (which came with a microphone to detect the sounds of hair breakage) never took off.
This solutionism approach bypasses a bedrock principle of good product development: deep user empathy. Without robust UX research (UXR) at the earliest stages, products risk solving problems that don't exist, or worse, creating new ones. Users aren't accustomed to interacting with autonomous agents, especially in complex, multi-step workflows. When confronted with an AI solution in search of a problem, or a tool that adds more friction than it removes, it’s no wonder adoption is lagging and users have grown cynical about AI’s value.
The overwhelm: Users are too often told to just “figure it out”
Imagine being handed the keys to a self-driving car when you're still learning to drive a manual transmission. That's how many users feel about AI. The rapid evolution of AI, from simple chatbots to sophisticated autonomous agents coordinating behind the scenes, far outstrips the average person's ability to comprehend, adapt, and build mental models for these new paradigms.
Users still grapple with basic questions:
- "How does this tool really work?"
- "Can I trust its recommendations?"
- “How do I properly QC its output?”
- "What happens if it makes a mistake?"
- “Who do I go to if I’m having trouble using it?”
- "Who is accountable when something goes wrong?"
When these fundamental questions remain unanswered, or when interfaces are designed without considering this foundational lack of understanding, trust remains low: A recent global survey showed that only 46% of people are willing to trust AI systems. It also showed their skepticism isn’t unfounded: another 56% said they make mistakes in their work due to AI. Why would people trust in tools they don’t know how to use, or that make their work more difficult?
Another survey found that 58% of U.S. companies now require employees to use AI tools, but forcing technology on workers is not the way to build long-term trust or adoption.
UX research: The key to bridging the trust gap
This trust gap is where UX research becomes not just valuable, but indispensable. UXR is the discipline that is singularly positioned to slow down, listen, and translate complex AI capabilities into human-centric experiences that foster understanding, trust, and real value.
Here's how UXR can tackle the AI paradox:
1. Uncovering actual needs, not creating them
- Prioritize exploratory qualitative research. Before a single line of code is written, UXR teams need to conduct deep ethnographic studies, contextual inquiries, and "no-tech" ideation workshops. This process isn't about asking users what AI features they want; it's about understanding their needs, their current struggles, and their underlying motivations. This research ensures that AI solutions are anchored in genuine utility, not technological novelty.
2. Designing for transparency and ease of understanding
- Progressive disclosure of reasoning. Users don't need to understand every neural network layer, but they do need to know why an AI made a suggestion. UXR designs tiered explanations, offering simple summaries by default and allowing users to drill down for more detail, fostering a sense of control and understanding.
- Source citation and audit trails: For complex agent-to-agent-to-human systems, UXR ensures that interfaces visualize the chain of logic, showing which agent contributed what and from where, preventing the "black box problem” and building trust in multi-agent orchestration.
3. Calibrating reliability and managing expectations
- Communicating confidence scores: When users expect perfection from AI, they’re swiftly disillusioned when it fails. UXR leads to designs that clearly communicate the AI's confidence levels or uncertainty. This clarity helps users understand when to fully rely on the AI, what kind of quality checks are necessary at what stage, and when to exercise caution.
- Proactive error highlighting: Instead of hiding potential errors, UXR can identify areas where proactive error handling will help build trust, guiding the user to review or intervene, fostering a sense of partnership rather than blind reliance.
4. Prioritizing user control and agency
- Easy override and flexibility: For every AI suggestion, UXR champions simple, intuitive mechanisms for users to edit, delete, or provide counter-input. This feature reassures users that they are still in command, a critical factor for psychological safety and trust.
- Clear feedback loops: Designing systems that visibly "learn" from user corrections and feedback reinforces the feeling of partnership, ensuring users feel heard and valued
- Emergency stop and human-in-the-loop: For agentic systems, UXR rigorously tests the availability and intuitiveness of "pause" and "override" functions, ensuring humans can intervene effectively when needed.
5. Fostering adoption through progressive AI rollout
- The "Crawl, Walk, Run" approach: Instead of launching fully autonomous, high-stakes AI from day one, UXR and phased deployment lead to higher trust and increased adoption.
- Start with AI as a helper (suggestions only), then move to opt-in automation for low-stakes tasks—only then, with proven reliability, progress to full autonomy for critical functions. This sequence allows users to gradually build their mental model and trust.
The future of AI depends on UX research and human-centered systems design
The current rush to market with AI products risks creating a vast chasm between technological capability and human acceptance. Without a deliberate, human-centered systems design approach, we may end up with powerful AI systems that few people genuinely trust or want to use.
When it comes to real-world AI adoption, UX research is not a nice-to-have: It is a prerequisite for sustainable AI success. By leading with and centering real user needs, fostering transparency, managing expectations, and empowering users with control over the entire autonomous system, UXR can transform the current AI paradox into a virtuous cycle of innovation, trust, and widespread beneficial adoption.
Pam Bohline is Research Director at Blink. She focuses on business purposes and connecting them with consumer and user needs through various research methods, and she's passionate about people, processes, and products.
