Our most active users aren't our healthiest users. We had to reckon with that.
Last month I pulled our engagement data expecting to celebrate. Daily active users up. Session length up. Entries per week up. Every growth chart pointed right and up.
Then I read what our most active users were actually writing.
One user was journaling 6 times a day. Not because they were building a healthy reflection habit — they were spiraling. Writing the same anxious thought over and over, looking for the AI to say something that would make it stop. Our product was doing exactly what it was designed to do: reflecting their emotions back to them. But reflection without boundaries isn't support. It's a mirror in a hall of mirrors.
We had to ask ourselves a question most product teams never face: what if our best engagement metrics are a sign that something is going wrong?
So we made changes that would horrify a traditional growth team:
We added gentle cooldown prompts after the third entry in a short window. Not blocking access — just a pause that says "you've been here a lot today. want to try something offline first?"
- We stopped celebrating streaks. Murror has no streak counter, no daily login rewards. Coming back should feel like a choice, not an obligation.
- We built a "quiet mode" that reduces AI reflections to simple acknowledgments when it detects repetitive emotional loops.
The result? Our DAU dropped 8%. Our average session count per user dropped.
But the users who stayed started writing differently. More depth. More honesty. More willingness to sit with uncomfortable feelings instead of seeking reassurance from the AI.
Our therapist advisor told us something that stuck: "The goal of a good therapeutic tool is to be needed less over time, not more."
I think this is the hardest product lesson in emotional AI: growth and health aren't always the same direction. Sometimes the right metric is one that goes down.
Anyone else building products where you've had to choose between engagement and user wellbeing? Curious how others navigate this.
Replies
the hard part here isn't deciding to add the cooldown, it's telling apart the two people hitting the same usage number. someone journaling 6 times a day could be spiraling like you described, or could just be having an unusually eventful day and processing it in real time. does the quiet mode key off entry count alone, or is it actually looking at repetition/sentiment in the content itself before it decides someone's in a loop rather than just busy?
Murror
really good question — and honestly one we're still refining. entry count alone would be too blunt, you're right. someone processing a breakup and someone who just had the best day of their life can both hit 6 entries.
right now quiet mode looks at a few signals together: repetition in language patterns (are they circling the same phrases?), emotional valence staying flat or negative across entries, and shrinking reflection depth — like when responses get shorter and more reactive instead of reflective.
it's not perfect. we've had false positives on people who were just venting productively. that's why we kept it as a gentle nudge rather than a hard gate. if someone says no, I want to keep writing, we let them. the goal is to interrupt the loop, not block the tool.
the 8% DAU drop is the part I'd want to hear more about - not the product decision, that part's clearly right, but how did that number land internally? did you have to actively defend it to a cofounder or investor who saw the chart go down, or was everyone already aligned once you showed them what the power users were actually writing?
Murror
@galdayan honestly it was uncomfortable. our first reaction as a team was denial — we kept looking for a bug in the analytics. when we confirmed the drop was real, there was a moment where it felt like we'd made a mistake.
what helped was showing the qualitative data side by side with the numbers. we pulled actual journal entries (anonymized) from before and after the changes. the before entries had this frantic, repetitive quality. the after entries were shorter but more honest. when the team could see the difference in what people were actually writing, the DAU drop stopped feeling like a loss.
we're lucky to be a small team where everyone believed in the mission enough to stomach a dip. I think if we'd had external pressure from investors optimizing for growth metrics, it would've been a much harder conversation.
@monatruong_murror the before/after journal entries side by side is a smart way to make an abstract metric feel real to the team. easy to rationalize a chart, much harder to rationalize actual words getting healthier. and yeah, the investor-pressure point is real, that's probably the scenario where most teams would've quietly reverted the change
Murror
@galdayan exactly — reading someone's actual words is so different from reading a chart. numbers abstract away the human experience, which is exactly what we're trying not to do. and yes, we got lucky on the investor side. I think that tension — between what grows and what heals — is going to define the next wave of mental health products. the ones backed by growth-first VCs will probably look very different from the ones that aren't.
"a mirror in a hall of mirrors" is a really sharp way to put it. most metric dashboards can't tell the difference between someone who loves your product and someone who's stuck in it, they just count events. curious what the actual intervention looked like once you noticed, did you build something that detects the pattern and nudges them elsewhere, or was it more of a manual outreach thing at first?