Most mental health and journaling apps suffer from what I call the "Dashboard of Doom" problem. You log your feelings for a month, open the app, and it cheerfully presents you with a giant chart proving you've been miserable for three straight weeks.
I'm Ethan Cole, Head of Product at Vividiary. Today, I want to talk candidly about how we tackled one of the hardest problems in our space: figuring out how to design proactive AI companions that actually make users feel better, rather than just acting as passive, depressing data loggers.
We made a lot of mistakes early on. We built features we thought were brilliant, only to watch them crash and burn in user testing. Here is the story of what we built, what we killed, and how we eventually found our footing in emotion AI UX design.
The Trap of Passive Tracking in Mental Health Apps
When we first started building Vividiary, we knew we wanted a frictionless entry point. We designed a rapid, 3-second 5-grade mood logging system (Best to Worst) with an optional emoji multi-select for specific emotions and activities.
It worked beautifully. It was low-barrier, and users were logging their moods consistently. But we quickly realized a painful truth: passive tracking isn't enough. Collecting emotional data without providing immediate, empathetic utility is like taking someone's temperature when they have a fever and just saying, "Yep, you're sick," before walking out of the room.
Users don't just want to know how they feel; they want to feel heard, and they want to know what to do next. We needed to shift from passive data collection to real-time emotional intervention.
What We Rejected: The Mood Calendar Heatmap UI (And Why It Failed User Testing)
Our first attempt to make this data useful was a classic product management trap: we built a dashboard. Specifically, we built a highly polished mood calendar heatmap UI, heavily inspired by GitHub's contribution graph.
The idea was simple: lighter colors for good days, darker colors for bad days. We thought users would love seeing their emotional trends at a glance.
We were wrong.
In our beta testing, the feedback was brutal. 7 out of 10 users told us the heatmap made them feel worse. One user's feedback stuck with me: "Seeing a sea of dark red squares on my screen just validates that I'm having a terrible month. It's overwhelming."
- Easy to build.
- Visually striking data density.
- Backward-looking.
- Lacks context (why was it a bad day?).
- Induces shame and anxiety during depressive episodes.
We realized that a raw mood calendar heatmap UI is a terrible default screen for someone in a vulnerable state. We killed it as the primary interface. (We eventually brought a modified, context-rich version back for our Premium tier's advanced analytics, but only when paired with actionable insights).
The Pivot to Emotion-Driven Product Design
We needed a completely different approach. Instead of showing users a sterile graph of their past, we wanted to build a system that adapted to their emotional state in real-time. We needed to focus on emotion-driven product design.
We asked ourselves: How can we create an "Emotion-Recognition-Emotion-Elicitation loop"? How can logging a bad mood actually trigger a comforting experience?
Our solution was gamification, but not the cheap "earn badges" kind. We introduced a clay character companion. When you log your mood, this character reacts. Over 30 days, based on your emotional data and interactions, the character evolves into one of 8 final forms.
This simple shift changed everything. Users stopped feeling like they were reporting to a database and started feeling like they were caring for a companion. It built emotional attachment and softened the clinical edge of mood tracking.
How to Design Proactive AI Companions That Actually Help
While the clay character provided emotional warmth, we still needed a way to help users process their feelings. This led to the core of Vividiary: our AI conversation mode.
When figuring out how to design proactive AI companions, the biggest risk is the AI feeling intrusive or preachy. Nobody wants a robot telling them to "just smile" or "try deep breathing" when they are venting about a terrible day at work.
The Approach We Chose:
Instead of the AI acting as a therapist, we designed it to act as a reflective middleware. You can talk to the AI via voice or text. The AI listens, validates your feelings, gently probes for context, and then—crucially—drafts a first-person diary entry for you.
You review it, edit it, and confirm it. This keeps the user entirely in control of their narrative while removing the friction of the "blank page." We dive deeper into why this specific workflow works so well in our post on automated mood insights.
What We Learned:
User testing showed that drafting the entry for the user reduced journaling abandonment by 60%. When you are exhausted, writing is hard. Talking to a sympathetic AI that does the heavy lifting for you feels like a superpower.
Rethinking Sentiment Dashboard Design: Forward-Looking vs. Backward-Looking
Even with a proactive AI companion, users still need to understand their long-term patterns. But we had to fix our sentiment dashboard design to avoid the "Dashboard of Doom" effect.
We shifted from backward-looking to forward-looking analytics. Instead of just showing what happened, we focus on why it happened and what to expect.
For users on our Free tier, we provide unlimited mood logging, 3 AI conversations a day, and basic analytics. But for our Premium tier ($2.99/mo or $11.99/yr), we unlock advanced analytics that use bubble charts to connect specific activities to mood shifts. We are also experimenting with predictive AI mood tracking to gently prompt users when they are entering a historically stressful period (like finals week or end-of-month reporting).
The key to successful sentiment dashboard design is ensuring every data visualization is paired with an actionable, empathetic insight.
Our Privacy-First Cloud Architecture: Protecting Vulnerable Data
You cannot build an emotion AI UX design without addressing the elephant in the room: privacy. Affective computing deals with the most sensitive data imaginable.
We are incredibly strict about our privacy-first cloud architecture. I want to be completely transparent here: Vividiary is a cloud-based app. To power the complex LLM interactions and ensure your diary is synced and safe if you lose your phone, your data is stored in the cloud.
However, "cloud" doesn't mean "exposed." We built our backend using React Native, Supabase, and Firebase Auth. Every piece of emotional data and every diary entry is encrypted in transit and at rest. The AI processes your context to help you, but your personal narrative is locked down. We don't sell your data, and we don't use your personal diaries to train public LLMs.
Building trust isn't just about writing a good privacy policy; it's a core component of UX design. If a user doesn't feel safe, they won't be honest. And if they aren't honest, the AI companion can't help them.
What's Next
We are currently testing new ways for the AI companion to recognize cognitive distortions in real-time (like "all-or-nothing thinking") and gently challenge them during the conversation phase. It's a delicate balance between being supportive and being overly analytical, but the early beta metrics are promising.
Building Vividiary has taught our team that the future of mental wellness apps isn't in better charts. It's in building systems that actively, empathetically participate in the user's emotional journey.
