Your users don't trust your AI. They shouldn't have to.
Six months ago we added an AI feature to Murror that we were sure would be a breakthrough. It analyzed journal patterns and surfaced emotional insights -- things like "you tend to feel more anxious on Sundays" or "your mood improves when you write about specific people."
The insight quality was genuinely good. We tested it internally, everyone was impressed.
Usage flatlined within two weeks.
When we dug in, the pattern was clear: users weren't rejecting the insights because they were wrong. They were rejecting them because they felt surveilled. The AI was telling people things about themselves that they hadn't discovered on their own yet. It felt like the product knew them better than they knew themselves, and that's not a compliment -- it's unsettling.
So we tried something different. Instead of the AI declaring insights, we made it ask questions. Instead of "you seem anxious on Sundays," it would prompt: "You've written a lot about weekends lately. What's been on your mind?"
Same underlying data. Completely different relationship.
Usage went up. But more importantly, the quality of what people wrote went up. They were discovering the patterns themselves, with the AI just pointing the flashlight in the right direction.
The lesson we keep coming back to: the best AI features don't prove how smart they are. They make the user feel smarter. The moment your AI sounds like it's performing intelligence, you've lost. People don't want to be analyzed. They want to be understood -- and those are very different things.
The trust didn't come from accuracy. It came from the AI knowing when to step back.
Anyone else navigating this tension between what your AI can do and what it should say?
Replies
WebCurate.co
I like this approach.
Sometimes it's not about making the AI more smart, but making the interaction feel more natural. Letting users arrive at the insight themselves usually builds much more trust than simply telling them what they are feeling.
Murror
Thank you both for the thoughtful responses! @alpertayfurr You nailed it -- "reflection instead of surveillance" captures exactly what we learned. That distinction became our core design principle. Even small language shifts, from declarative ("you are") to curious ("it seems like"), completely changed how people received the same insight. @hosseinyazdi Exactly right. The naturalness of the interaction matters more than the intelligence behind it. When users discover a pattern themselves with a gentle nudge, they own that insight. When the AI announces it, it feels like a diagnosis. Trust comes from making people feel capable, not analyzed.
@monatruong_murror This is a really good lesson. Especially with emotional products, the AI doesn’t always need to announce the insight directly. Sometimes the better experience is helping the user arrive there themselves, so it feels like reflection instead of surveillance.