Let's be honest about the current state of digital mental health tools: most of them are drowning in AI slop.
If you download a top-charting wellness app today, you'll likely be greeted by a chatbot masquerading as a therapist, automated journal entries that read like generic horoscopes, and push notifications guilt-tripping you into hitting your "daily reflection streak." It's an ecosystem built on frictionless, soulless content—what UX researchers are starting to call "the output of averaging."
At ViviDiary, we build a modular mood and life tracker. Our core promise is "Your day, in moods, emojis, and patterns." Notice what isn't in that sentence? Artificial Intelligence.
We use AI, but we treat it as a highly restricted, opt-in utility rather than the star of the show. Today, I want to walk through our product decisions regarding AI, why we actively design against AI slop in wellness apps, and the hard truths about privacy architecture that the industry is trying to ignore.
!UX designer crossing out auto-generated AI text on a mobile wireframe
The Rise of AI Slop in Wellness Apps
When Large Language Models (LLMs) became accessible via API, product teams across the wellness sector lost their minds. The mandate was clear: Put AI everywhere.
The rationale made sense on paper. Journaling is hard. Staring at a blank page causes friction. If an AI can look at your step count, screen time, and a few tapped buttons, and automatically write a beautiful, three-paragraph summary of your day—isn't that magic?
No. It's slop.
We define AI slop in wellness apps as any auto-generated content that removes user agency and replaces genuine self-reflection with algorithmic assumptions. It leads directly to wellness app fatigue, where users feel burdened by the constant need to evaluate whether the AI accurately captured their feelings. Instead of offloading mental load, you're just shifting it from "writing" to "editing a robot's guesses."
Why We Killed Fully Automated Journal Entries
In early 2023, we tested a feature internally. You would tap a few emojis—say, 🌧️ (Rain), 💻 (Work), and 😫 (Exhausted)—and our AI would instantly draft a journal entry: "Today was a rainy day. I felt exhausted from work, but I'm doing my best to push through."
The Approaches We Considered 1. The Fully Automated Diary: The AI writes everything based on passive data and minimal input. User just hits "Save." 2. The Conversational Coach: The AI acts as a proactive chatbot, asking you probing questions about why you felt exhausted. 3. The Human-Led Archive (Our Choice): The user logs their mood and emojis. If they want to write, they write. The AI only steps in if explicitly summoned to help expand on a thought, and the user must manually review and confirm every word.
What We Chose (and Why) We decided the AI should only draft from a conversation you review. It was one of the easiest product decisions we've made.
When we put the prototype in front of a beta group of 50 users, the feedback was brutal. 82% of them reported feeling "disconnected" from their own diary. One tester told us, "It feels like someone else is living my life and just CC'ing me on the summary."
This is exactly why we killed the fully automated mood tracker. If the app does the feeling for you, you aren't tracking your mood—you're just training an algorithm. We realized that the friction of self-reflection is actually the feature, not the bug.
Emojis First: Why AI is Strictly Opt-In
To combat AI slop, we had to rethink the core input loop. We stripped ViviDiary down to its absolute essentials.
New users start with exactly one module turned on: Mood. It's a simple, 5-level scale (Great, Good, Okay, Low, Rough). That's it. It takes under 30 seconds. If you want more detail, you can toggle on 22 manual emoji categories or 4 HealthKit auto-categories.
We rely on a single "Warm (따뜻하게)" tone for the app's interface. It sits beside you like a companion; it doesn't coach you.
If a user wants to write a deeper entry, they can. On those specific days, if they want help articulating a complex feeling, they can tap an opt-in AI helper. But the AI never saves or confirms without user review, and it never provides therapy or diagnoses.
More importantly, we stripped out all pressure-style UX. There are no panic-inducing streaks. There are no completion percentages. Our Focus module (Routines + Todos) is strictly opt-in. A Routine is just something you want to notice—it keeps a gentle personal-best count, never a streak-freeze guilt trip. We learned early on that we stopped telling users how they feel and started letting them just be.
!Mobile app interface showing a simple 5-level mood scale and opt-in emoji modules
Privacy Architecture: De-Identification and Data Minimization
If you're going to design against AI slop, you also have to design against the marketing slop surrounding AI privacy.
Right now, there is a massive trend of wellness apps claiming to be "privacy-first" or boasting about "de-identified AI processing." They tell users, "Your diary text is de-identified before any external processing!"
Let me be direct: for 95% of apps using high-quality LLMs, this is a lie. True offline LLMs are currently too large, battery-intensive, and slow for a lightweight mobile app experience. If an app is generating high-quality text, it is almost certainly pinging a cloud API, which is why our privacy approach relies entirely on data minimization and strict de-identification.
At ViviDiary, we refuse to play that game. We are transparent about our architecture: ViviDiary is cloud-stored using Supabase.
We don't rely on the false comfort of offline storage claims. Instead, our privacy comes from strict data minimization and de-identification.
How We Built It When you use our opt-in AI helper, your diary text is de-identified before any external or AI processing occurs. We strip out personally identifiable markers locally, send the anonymized payload to the cloud for processing, and return the result.
We chose privacy-first, de-identified cloud processing because it allows us to provide a fast, reliable experience without draining your battery or lying to you about where your data goes.
Furthermore, your archive is safely backed up in our privacy-first cloud architecture. We don't hoard data we don't need. The AI doesn't train on your personal trauma. It's a stateless transaction.
The ROI of Keeping It Human
When you refuse to use AI slop to artificially inflate engagement metrics, you have to build a genuinely good product.
ViviDiary is Free for all input modules, unlimited mood and emoji logging, a 3-month calendar archive, and up to 3 Routines and 5 Todos. Our Premium tier ($2.99/mo or $11.99/yr) unlocks deeper historical archiving and unlimited Focus items.
We don't paywall the AI because the AI isn't the product. The product is your self-awareness.
What We Learned By keeping AI strictly opt-in and removing pressure UX (like streaks), we expected a drop in daily active users (DAU). The opposite happened.
Our Day-30 retention for users who only use the 3-second emoji check-in is 40% higher than the industry average for journaling apps. Why? Because we don't exhaust them. We don't make them read a robot's summary of their own life. We just let them tap "Okay," tap a ☕ (Coffee) emoji, and get on with their day.
What's Next We are currently refining our "Mirror" feature (our pattern discovery engine). Mirror looks at your data across Time, Activity, People, and Focus domains. It is purely observational. It will tell you, "You tend to log 'Low' moods on days you don't complete your Morning Walk routine." It will never say, "You should walk more to be happier!"
Designing against AI slop in wellness apps isn't about hating AI. It's about respecting the user's mental space. In a world where every app wants to talk at you, sometimes the best feature you can build is an app that simply listens.
