Self-initiated

Enterprise Product

root, a journal for people who have never kept one

A guided journaling app built around coming back rather than keeping a streak, and around protecting the honesty that makes writing worth doing at all.

Role

Everything, on my own

Disciplines

Product Design, Research, Prototyping

Client

Self-initiated

Platform

iOS and Android, English and Portuguese

Overview

root is a journaling app for people who have never kept a journal. Eight weeks, one small question a night, and every question is about a day that has already happened, so there is always something to answer.

I designed it end to end on my own: the research it stands on, the product decisions, every question in the bank, a working prototype in English and European Portuguese, and the plan for the first round of testing.

Challenge

Two problems, and the second one is quieter

Almost nobody stays. Across more than ninety mental health apps and one and a half million downloads, a median of 3.9% of people were still opening theirs fifteen days after install (Baumel et al., 2019). Journalling apps sit inside that number.

The usual fix is a streak. Streaks work right up until the first missed day, and then they punish the person for the exact thing that was always going to happen. A beginner doesn't need a scoreboard. They need a reason to come back after missing.

The second problem gets less attention. A journal is only worth keeping if it is honest, and honesty dies the moment someone suspects they are being read. Every feature that reads an entry, however helpful, spends some of that honesty. Most products spend it without saying so.

Process

Three directions, one prototype, and a person who approves every question

Three directions went up on the wall: one built around ink and slow handwriting, one around speaking instead of typing, and one around a small plant that grows as you write. I archived two, and they stay visible on the canvas as directions not taken.

Every question in the bank is written against a fixed set of rules, grounded in the expressive writing literature, and then approved one at a time by a person. Nothing reaches the app on its own. Eleven of the first twelve I wrote were rejected.

The prototype is the source of truth. It runs in a browser at phone size, in both languages, with a study mode for handing it to someone else. The design canvas is exported from the prototype, so the two cannot drift apart.

Every screen in these images is the working prototype, exported, not a mockup redrawn beside it.

Decisions

No streak. The count only goes up.

The Tension

Streaks are the most reliable retention mechanic there is, and the most reliable way to make a beginner quit after one missed night.

What I chose

root counts total days written, and that number never goes down. Miss a week and it waits where you left it. The only thing that changes is the plant: the newest leaf wilts, and Today offers a smaller question, labelled Coming back.

What it cost

Day to day pull. There is nothing to protect, so there is nothing to panic about losing. The bet is that the first missed night stops being the end of the attempt.

The AI never reads your journal by default.

The Tension

Everything genuinely useful an AI could do here needs the entries: better questions, patterns across weeks, a word for the thing you keep circling.

What I chose

Three layers, and the person sets each one. The question bank is generated and human approved before anyone installs the app, so it needs no entries at all. On device signals, such as which nights get skipped, can shape tomorrow without any text leaving the phone. Anything that reads actual writing is one entry at a time, asked for each time, on the screen right after saving.

What it cost

The product is less clever than it could be, and the privacy screen had to get longer and more specific rather than shorter and friendlier. Its headline now reads: nothing leaves this phone unless you hand over one entry.

The AI never reads your journal by default.

A question needs a turn in it.

The Tension

A question about a day that already happened is easy to answer. Easy to answer is not the same as worth answering, and a beginner can't tell the difference until they have written the thing.

What I chose

Every question has to have a turn in it: the answer has to arrive somewhere the person did not expect when they started writing. Below is one real batch, judged against that rule.

What it cost

Speed. Eight weeks needs around fifty approved questions, and each one is a small piece of writing that a person has to read and judge. The bank fills slowly on purpose.

Telling people what reads their writing.

The Tension

A journal for beginners will eventually catch somebody on a very bad night. Doing nothing is a decision. So is reading every entry in order to find it.

What I chose

A small safety net, specified in writing and built to run entirely on the phone. When it notices something, the app gets quieter: the saved screen drops its prompts, that entry is never quoted back on Today, and for three days the nightly reminder carries no exercise at all.

What it cost

The privacy promise got harder to write, because something does read the text even though it never leaves the phone. The helpline numbers are still bracketed placeholders in the prototype. A wrong number in a crisis is worse than no number, so they stay empty until a clinician signs off on them, along with the thresholds and the wording.

Telling people what reads their writing.

Nothing private on a lock screen.

The Tension

The most effective reminder is the question itself. A lock screen is also read by partners, parents, and whoever happens to be standing behind you in a queue.

What I chose

By default a notification never carries the question, the theme of the week, a feeling word, or a number of days. There is one switch in Rituals that turns the question on, and it is off until the person turns it on themselves.

What it cost

A blunter reminder, and one more thing to explain. The whole set is deliberately quiet: one a day at most, nothing between 23:00 and 06:00, no counting of absence out loud, and after seven days away it stops on its own.

Solution

One question, one rule, and room to write

A night in root is one question, one short rule about how to answer it, and room to write. The question is always about a day that has already happened, so nobody has to invent anything.

Not every night is writing. Some nights name a feeling and then what put you there. Some ask you to place the people close to you in three circles and show you what that shape means. Some rank what matters against the month you actually had. Some are a pause: a drawing and one sourced line, with nothing to fill in.

The whole app exists in English and European Portuguese, switchable at any point, and on a tablet the writing page opens into a spread.

Validation

HOW IT WAS REVIEWED

Not tested yet. Five moderated sessions and a seven day diary study are planned.

WHAT SHIPPED

Prototype, going into testing

Outcome

Nobody has used it yet, so there is nothing to claim

What exists is a complete prototype in two languages, a question bank being approved one question at a time, a written specification for the safety net, a full set of notification copy, and a first round of research ready to run.

The signal is written down before the study starts, so it cannot quietly move: if fewer than five of the ten people in the diary study write on at least four of seven days, nothing else gets built until I understand why.

Testing. Five moderated sessions with first time journallers, then a seven day diary study with ten people on their own phones, in European Portuguese.

Built so far. Eight weeks of nights, four kinds of exercise, two programmes, a study mode, account recovery, and the full app in English and European Portuguese.

Still open. Clinical sign off on the safety net, its thresholds and the helpline numbers. A native read of the Portuguese. The rest of the question bank.

Reflection

When building gets this fast, judgement becomes the job

The mistake was sequencing. I designed eight weeks of a programme before watching a single person write a single night. The research plan should have come first. It now sits in front of the next build instead of behind it.

This was built with Claude as a collaborator, and being straight about that matters more than looking clever. It wrote most of the prototype code, the design canvas and the written specifications, and it proposed several of the decisions on this page, including the three layer model for what the AI is allowed to read, which I handed over on purpose to see what it would argue for.

My part was deciding and rejecting. Eleven of twelve questions. Two versions of a product film. A home screen that had drifted into feeling like a game. The streak. None of that gets easier when the making gets cheap, and all of it is the part that decides whether the thing is worth anyone opening.

Next Project

Self-initiated

Vovini, for the feedback you keep hearing

Self-initiated

Enterprise Product

root, a journal for people who have never kept one

A guided journaling app built around coming back rather than keeping a streak, and around protecting the honesty that makes writing worth doing at all.

Role

Everything, on my own

Disciplines

Product Design, Research, Prototyping

Client

Self-initiated

Platform

iOS and Android, English and Portuguese

Overview

Challenge

Two problems, and the second one is quieter

Process

Three directions, one prototype, and a person who approves every question

Every screen in these images is the working prototype, exported, not a mockup redrawn beside it.

Decisions

No streak. The count only goes up.

The Tension

Streaks are the most reliable retention mechanic there is, and the most reliable way to make a beginner quit after one missed night.

What I chose

What it cost

Day to day pull. There is nothing to protect, so there is nothing to panic about losing. The bet is that the first missed night stops being the end of the attempt.

The AI never reads your journal by default.

The Tension

Everything genuinely useful an AI could do here needs the entries: better questions, patterns across weeks, a word for the thing you keep circling.

What I chose

What it cost

A question needs a turn in it.

The Tension

A question about a day that already happened is easy to answer. Easy to answer is not the same as worth answering, and a beginner can't tell the difference until they have written the thing.

What I chose

Every question has to have a turn in it: the answer has to arrive somewhere the person did not expect when they started writing. Below is one real batch, judged against that rule.

What it cost

Speed. Eight weeks needs around fifty approved questions, and each one is a small piece of writing that a person has to read and judge. The bank fills slowly on purpose.

Telling people what reads their writing.

The Tension

A journal for beginners will eventually catch somebody on a very bad night. Doing nothing is a decision. So is reading every entry in order to find it.

What I chose

What it cost

Nothing private on a lock screen.

The Tension

The most effective reminder is the question itself. A lock screen is also read by partners, parents, and whoever happens to be standing behind you in a queue.

What I chose

What it cost

Solution

One question, one rule, and room to write

A night in root is one question, one short rule about how to answer it, and room to write. The question is always about a day that has already happened, so nobody has to invent anything.

The whole app exists in English and European Portuguese, switchable at any point, and on a tablet the writing page opens into a spread.

Validation

HOW IT WAS REVIEWED

Not tested yet. Five moderated sessions and a seven day diary study are planned.

WHAT SHIPPED

Prototype, going into testing

Outcome

Nobody has used it yet, so there is nothing to claim

Testing. Five moderated sessions with first time journallers, then a seven day diary study with ten people on their own phones, in European Portuguese.

Built so far. Eight weeks of nights, four kinds of exercise, two programmes, a study mode, account recovery, and the full app in English and European Portuguese.

Still open. Clinical sign off on the safety net, its thresholds and the helpline numbers. A native read of the Portuguese. The rest of the question bank.

Reflection

When building gets this fast, judgement becomes the job

Next Project

Self-initiated

Vovini, for the feedback you keep hearing