Research
/
2025
AI Trust: What Makes Users Adopt or Reject AI
A two-market study, across the U.S. and India, of the psychology behind AI adoption, distilled into a framework teams can design against.
Project details
Company
Meta
Role
Research lead, end to end
Methods
Unmoderated mixed-methods studies: surveys, ratings, card sorts, and video responses (n=39, U.S. + India)
Year
2025
Overview
The Problem: Tech leaders were calling AI “the biggest shift of our lifetimes,” but everyday people still found it unreliable and hard to trust. Meta was building AI into its products without a clear picture of why people embrace it or walk away.
The Method: Five quick-turn studies in the U.S., one for each barrier to adoption, then a deeper multi-part study with everyday AI users across India: ratings, rankings, a tone card sort, and selfie-style video responses.
The Core Insight: AI adoption is a perception problem, not a technology problem. People adopt AI when it feels right, not just when it works.
The Impact: A five-barrier framework and a portable design principle for how AI should sound and behave, built to guide Meta’s Business AI direction across two of its biggest markets.
What was this?
Everyone said AI was the future. Users weren’t so sure.
CEOs called AI “the electricity of our age.” Meanwhile, surveys showed most people didn’t trust businesses to use it responsibly. That gap mattered for Meta, which was putting AI in front of billions of people. I set out to understand the psychology behind it: what makes someone trust an AI tool, and what makes them reject it.
How did I study it?
Five barriers in the U.S., then a deep dive in India.
Inspired by Harvard Business Review’s research on the consumer psychology of AI, I broke adoption into five barriers and ran a focused study on each one with U.S. participants aged 21 to 78. Then I took it to India, one of the world’s most active AI markets, with 10 everyday AI users from Agartala to Hyderabad. They used their own real tools, with no fake prototypes, and answered through ratings, rankings, a card sort, and 7 selfie-style video responses.
What did people tell me?
Five barriers stand between people and AI.
Trust: AI for speed, humans for judgment. People trusted AI to find the fastest route (83%) but not to check health symptoms. Comfort with AI replacing humans averaged 3.5 out of 5.
Control: “Let me decide, not the AI.” People wanted AI to ask first, let them override it, and let them turn it off, especially for money and health.
Personalization: AI adapts “just okay,” at 3.0 out of 5. “It never learns. I keep correcting it, and it keeps messing up.”
Emotion: Tone matters, at 4.0 out of 5. “It was too robotic. I didn’t feel heard.”
Transparency: People want the why, at 3.8 out of 5. “If I don’t know why it’s making that decision, I’m not going to trust it.”
Did it hold up in another market?
In India, feeling known mattered most.
8 of 10 people said their AI tools felt like they were learning them, and remembering past preferences ranked as the most valued feature. But the line was thin. “It feels exciting and scary at the same time.” When AI forgot what people told it, like serving non-vegetarian options to a vegetarian, or couldn’t understand their accents, it felt careless and exclusionary.
How should AI sound?
Let the task dictate the tone, not the style guide.
In both markets, I asked people to match AI tools to the tone they should use, and both markets gave the same answer: friendly for dating apps, formal for customer service, health, and money, and clear and neutral for writing tools. A casual tone in a serious moment broke trust. Tone isn’t flavor. It’s function.
What did I recommend?
Sell what works, not what sounds human.
Marketing: don’t sell “humanlike.” Lead with utility, because personality-first messaging confuses people from the start.
Creative: friendly is good; too friendly is not. Match tone to the task, the moment, and the mood, not just the brand voice.
Product: show the why behind every suggestion (“Because you watched X”), let people override and undo, and make memory visible so people feel known.
Strategy: clarity beats clever. The more plainly we explain what AI can and can’t do, the more likely people are to try it again.
Where does it fall short?
I was upfront about its limits.
Small samples, by design. About six people per barrier study and ten in India gave me fast, directional signal, not population-level proof.
General consumers, not just business owners. I studied everyday AI users to understand the underlying psychology, then applied it to Meta’s Business AI work. A follow-up with business owners would confirm the fit.
Two markets. The U.S. and India differ, and other markets will too.
What came of it?
Outcome
A five-barrier framework for designing trustworthy AI experiences: trust, control, personalization, emotion, and transparency
A portable design principle: let the task dictate the tone, not the style guide
A two-market research program across the U.S. and India on AI voice and personality
Built to guide Meta’s Business AI work, including onboarding and dashboard priorities
Next project
