Finding breakthroughs through AI-assisted journaling
The Brief
Mindbloom is the leading platform for at-home ketamine therapy, facilitating hundreds of thousands of sessions across 36+ states. Their clinical team knew two things from their data: clients who set intentions before sessions rated their experiences 22% higher, and clients who journaled after sessions were 17% more likely to continue their program. But the app wasn't helping with either. Intention setting was loosely guided, and journaling was analog — physical notebooks mailed to new clients. The brief was to bring both into the app and use AI to help clients get more out of each step. Three-person team: me, an engineer, and a PM. Eight weeks.
Intention Setting
Most clients didn't know what a "good" intention even looked like. Too open-ended and they'd freeze. Too prescriptive and it wouldn't feel personal. We landed on a three-question conversational flow where the AI built up context about what the client wanted from their session, then suggested three tailored intentions to choose from — or they could write their own. Text or voice input. Multi-tap options kept it moving; freeform input kept it personal. We prototyped extensively in Figma — testing whether it should feel like one long chat or one question at a time, whether clients should be able to ask the AI questions back, how many questions was right before it felt like a chore. Three was the sweet spot.
Post-Session Journaling
Post-session journaling had a completely different set of constraints. User research showed session memories were most vivid immediately after — clients who closed the app and came back later usually didn't come back at all. So the journaling flow triggered right when the session ended, using the same background so it felt like a continuation rather than a separate task. The other constraint was physical. Post-session, some clients had blurry vision or reduced fine motor control. We doubled the typography and button sizes compared to the intention-setting UI and made the entire flow voice-only — no typing. One question at a time instead of a scrollable chat. The design had to meet people where they physically were.
Results
Intention quality improved by 23% after release, beating the target. But a vocal group of clients pushed back — not on the output quality, but on the idea of AI being involved in something so personal. We removed all explicit mentions of AI while keeping the functionality identical. Completion rates spiked immediately. The feature worked better when people didn't know how it worked. That was the most interesting finding of the whole project. Post-session journaling was harder to benchmark since it replaced physical notebooks with no tracking. The first release didn't hit internal goals, and the team was experimenting with making the flow shorter while acknowledging that some clients want to journal later.
Looking Back
If I had more time, I would've pushed harder on the journaling side. The intention setting flow was dialed in, but journaling needed more iteration on timing and length — and eight weeks wasn't enough to get both right. The constraint of "journal immediately after" made clinical sense but created a real tension with the user's emotional state that we never fully resolved.
