What to write in an empty journal?
An empty screen is never empty. What sits on it decides the register of your answer before you’ve touched the keyboard.
By Anastasiia Nedbaliuk, 21 August 2026, 16 min read
- design
Evening. You open a journaling app because the day was the kind of day that has to go somewhere. The cursor blinks. You write: “Day was fine, just tired,” close the app, and go to bed.
The same day could have come out as three paragraphs.
Which of the two happens is decided before you touch the keyboard, by what sits on the screen while you still haven’t written anything.
The empty state is not empty
Interface design has a term for it: the empty state, the screen with no data in it yet. It usually gets designed last, on the leftover principle: something has to be shown while there’s nothing to show.
But in an app people write about themselves in, the empty state is not an interim condition before the real interface. It is the interface. It opens every session, the one element of the product every user sees, every single time. And everything that happens next, the text you leave and what the app comes to understand from it, starts here.
Our own spec called this task the blank page problem. The framing isn’t ours: Apple’s own description of Journal promises that with its suggestions you’ll never have to start with a blank page. The industry has agreed that the page is the problem and the prompt is the solution.
Accept that frame, and the only question left is what to fill the page with. What the frame conceals is that the page was never blank in the first place.
Three ways to fill it
If the page is the problem and the prompt is the solution, the design space turns out to be small. Every journaling app we looked at fills the empty screen in one of three ways, and you have seen all of them.
Ask about the day. “How was your day?” and its variants. A question creates two people: the one who asked and the one who answers. You’re placed in the position of reporting, and a report demands completeness and a summary. A summary of the day at eleven at night is “fine.” The most common answer to a direct question about your day is one word, and not because the person had nothing to say.
Give a topic. Prompt of the day, theme of the week, list of categories. This works better than the report, because it takes the choice of subject off the person: you don’t have to decide what to write about, only to write. The cost is that the subject is set from outside, and whatever doesn’t fit it won’t get written today. Ask someone about gratitude on a day when they feel awful and they’ll either lie or close the app. Worth noticing, too, where these prompts tend to pull: Apple describes its reflection prompts as focused on gratitude, kindness, purpose and the like. What’s set isn’t only the topic. It’s the intent.
Ask nothing. A blank field and a cursor. That’s how most text editors behave, and it’s how Day One behaves: the daily prompt exists, but it doesn’t sit on the writing screen. You have to go to a separate view for it. A blank field gets either everything or nothing. It works beautifully for people who already know what they want to say, and not at all for everyone else, which is most people on most evenings.
Three different decisions, and what separates them isn’t topic. It’s who they propose you should be for the next five minutes: the person filing a report, the person writing an essay on an assigned subject, or the person left to manage alone.
What we had and what we have now
In the spec written before release, our home screen was a list of topics. Six tiles: Personal growth & learning, Goals & aspirations, Life balance & priorities, Relationships & connections, Work & career, Challenges & obstacles. Above them a heading: “Choose a topic to explore, or start with your own thoughts.” Below them a field: “What’s on your mind?”
At first glance, this would seem like the right framing. Almost every journaling app does something like this, and the logic is clear: we take the choice of subject off the person, give structure, don’t dictate content. The categories were broad, too: not “three things you’re grateful for” but whole areas of a life. And the way out was on the screen from the start, in the heading and in the field. It seemed like a frame that couldn’t do any harm.
We changed that screen before the app shipped. No user ever saw it. Why the second version works better isn’t immediately obvious.
Here’s why we ended up never shipping this screen.
To press any of those tiles, you would first have to classify what is happening to you. And at the moment you actually need to write, that is usually the exact thing you can’t do. Is a fight with your mom “Relationships & connections” or “Personal growth & learning”? Is waking at four for the third week running “Life balance & priorities” or “Work & career”? The tangle is the reason you came. A list of topics asks you to take first the step that should have been the outcome.
And the category wouldn’t let go once you’d pressed it. Choosing “Work & career,” you’d have promised to write about your work. If it turned out by the third sentence that this was actually about your father, you’d already be in the wrong room, and either you’d keep writing about the wrong thing, or you’d abandon it and start over. People generally don’t start over.
The free text area at the bottom was our answer to exactly this, and it wasn’t enough. Six tiles hold the center of the screen; the field is a gray line beneath them. Formally the choice is equal. Nothing stops you from ignoring the list. But what’s set large reads as the offer, and what’s set small reads as permission. By the time you notice the difference, you’ve already picked a drawer.
Instead of a list of topics, the first screen now opens with three unfinished sentences:
I want to make sense of… / I’m trying to decide… / I’ve been noticing…
On the very first journal, they sit under a greeting from Alma, our reflection companion, the AI you’ll be writing with. She introduces herself, says she helps you think through whatever is on your mind, and asks what you’d like to start with. Under the three sentences there’s a line: tap one to start, or just type what’s on your mind. And there’s a fourth button, “I’m not sure where to start,” doing a different job than the three.
So there is a question on the screen after all, and it’s worth being exact about what each part does. Alma’s line sets the frame: someone is here, and they read what you write. The three sentences are something else: not what she asks you, but ways of beginning to speak. You don’t answer them. You continue them.
None of them is a topic. They are the openings of first-person sentences, and what they sort is intent, not subject: to make sense of something, to decide something, to notice something. What it’s about gets settled while writing, not before it.
The difference isn’t cosmetic. A topic asks you to classify your life; an intent only asks you to know what you want from yourself right now, and people usually do know that, even when they can’t name the subject. An intent also doesn’t hold you in the room: “I want to make sense of…” works just as well when the third sentence turns out to be about your father.
And the grammar does one more thing, less obvious. A question puts an interlocutor in front of you, someone who has to be answered. An unfinished sentence puts no one there: your own voice has already started, and all that’s left is to continue it. Nobody asked you anything, so there’s nobody to stay consistent for. Alma’s greeting does ask, and that’s the price of having someone in the room at all: a companion who reads you has to say hello. But what she hands you isn’t a question to answer. It’s three ways to begin.
Picture one evening and one event: you’ve had a fight with your mom.
To “How was your day?” you’ll most likely write: “Fine, argued with my mom, it’s OK now.” Report, summary, closed.
A “Relationships & connections” tile will produce something tidy and on-theme, a text about your relationship with your mother in general, because you’ve already promised to write about relationships.
And “I’ve been noticing…” will produce something more like: “I’ve been noticing that after conversations with my mom I need an hour to come back to myself.” Not a summary and not a topic: an observation, and one with somewhere to go.
Same evening, same event, three different texts. When the goal is open ended, the starting point decides how far you can reach.
The fourth button is a different story. It came from a pattern we kept seeing: people installed the app, opened it, and never wrote anything at all. The three sentences assume you know what you want from yourself, even when you can’t name the subject. That assumption holds more often than a list of topics does, but it doesn’t hold for everyone, and it holds least well on the first evening, when the app is still a stranger. For those people the screen was asking a question they had no answer to, and the honest answer, “I don’t know,” had nowhere to go.
So we gave it somewhere. Choosing it doesn’t open a different journal; it just means Alma speaks first. She says there’s no right way to do this, that nothing you say needs to be important or well put, and asks what’s been taking up more space in your head than it deserves lately. Two or three words are enough.
Early signs are good: people who tapped it started writing. It’s worth saying what that costs us to admit: our own answer to the blank page needed a fallback for the people it didn’t reach.
The conversation before the journal
Both versions so far are about the first arrival, the screen a person meets when the app still knows nothing about them. The blank page itself comes back every session, and the standing answer to that comes later in this essay. But somewhere along the way to making it easier to get started, we also passed through an interim solution, and it ran into something else.
Having dropped the list of topics, we first decided not to drop people straight into a journal. Before the first entry there was a separate conversation. The AI explained that it wanted to get to know you a little, and asked what had brought you here.
The version testers saw put an Understanding meter and a list of topics in a left sidebar, with Skip and a Finish button at its top. Four core topics (Values, Goals, Struggles, What Brings You Here), each with its own fill and a line of what the system had understood from your words: takes on too much at work, fixes it at night. Below them a second group, Also emerging (Life Domains, Reflection Experience, Current Capacity), the things you hadn’t been asked about but had said anyway. The AI introduced itself as our AI Companion; the name Alma came later. At 55% the information was considered sufficient, and you could move to the journal or keep talking.
Only testers saw this version.
It’s worth looking at closely precisely because it was made conscientiously. The boundary was visible: unlike most onboarding, this conversation showed you how much was left. The exit stood on the screen from the first second. The system showed not only its own progress but its reading of you.
And the meter still does something we hadn’t planned for. It measures not your answer but the understanding of you: “Understanding: 0%” stands in front of you as a statement about you, and how much 55% is, and of what, doesn’t follow from anything on screen. More importantly, a percentage that rises with what you write gives you a reason to write for the percentage. Confiding quietly turns into completing, and that works against the thing the conversation exists for.
In the end we removed the stage entirely. We didn’t improve the meter. We removed it along with the conversation. The questions the conversation asked didn’t go anywhere: they’re answered by the three unfinished sentences and by the first entry itself, from which the system extracts what it needs anyway. The answer to “when do I move to the journal” became “you’re already in it.”
The principle
The empty state is not the absence of design. It’s the hardest question in the product to answer, and it gets asked every single time someone comes in to write what’s on their mind.
This carries beyond journaling. A feedback form, a support ticket window, a “tell us about yourself” field on an application: anywhere an interface asks a person to formulate something of their own, the first line decides whether you get a brush-off or an answer.
When the question is drawn from you
Everything above is about day one. The question is the same for everyone, because the app knows nothing about you.
After that, something else begins. A system that reads what you write can ask a question assembled from your own entries. In our case that’s the Seed, a journal starter that sits on the empty screen of a newly created journal. It’s assembled from Highlights, the key phrases the system pulls out of your entries, and Glimpses, observations that connect different journals to one another. The spec states the Seed’s purpose like this: to open territory the user hasn’t yet explored.
Note the verb. Not “suggest a topic” and not “remind you what you wrote about.” Open the unexplored.
It looks like this:
Between work going quiet and pulling back from family, which one is actually easier to imagine facing first?
Beneath the question is a Reasoning block you can expand. The idea of showing not only the result but the reading it was made from appeared back in that first conversation, and outlived the conversation itself:
You named both as stalled at once, but they may not carry equal weight. Separating them could show whether one shutdown is masking a harder truth about the other.
And a caption: inspired by three items.
We’re not the only ones with personalized prompts. The closest example is Go Deeper in Day One: if the entry is empty, it offers general starter questions; if you’ve already written something, the questions are drawn from what’s there, and the more text, the closer they land. The feature is paid and has to be switched on separately. The difference comes down to two things. Go Deeper reads the current entry; the Seed reads the history. And Go Deeper is a button you have to go and find; the Seed sits on the screen by itself.
Both differences favor power at the expense of caution. Which brings us to the cost.
What it costs
Yesterday’s you sets today’s agenda. A question assembled from previous entries leads you further along a track already laid. A blank field lets you arrive with something that has nothing to do with anything before it; the Seed, slightly less. We frame Seeds around the unexplored for exactly this reason, but that’s a mitigation, not a solution.
The observation was made without your asking for it. “You named both as stalled at once” is not a prompt. It’s something the system noticed in your words and told you, unasked. It’s traceable to what you wrote and mostly useful. But sometimes a person didn’t ask to be read that closely, and finds out at the moment they already have been.
The Reasoning block shows the basis but not the alternatives. You see what the question was assembled from. You don’t see what else the system thinks about you and decided not to ask today.
The exit is quieter than the offer. Nothing is blocked. You can refresh the Seed and get a different one, you can ignore it and start with something of your own: the field at the bottom accepts anything. But the refresh button gives you a new question rather than none, and the Seed occupies the center as a prominent tile while the field stays a gray line at the bottom. Formally the choice is equal. Visually it isn’t, and the default is set not by a toggle in settings but by whichever thing is bigger on the screen. On a wide screen the balance shifts: the field stretches into a pale band across the bottom and holds its own against the tile. Which only sharpens the point: what you’re being nudged toward isn’t a setting anyone chose, it’s a consequence of how much room each thing got.
Three questions for any app you write about yourself in
Everything above is one product’s path through one problem. The same audit works from the outside, on any app you write about yourself in. You don’t need access to anyone’s spec; three things visible on the screen are enough.
Was I asked something, or was I given a way to start talking? Look at the punctuation of the first line. A question mark creates an interlocutor you have to report to. A category name asks you to sort what’s happening to you into a drawer before you’ve described it. An unfinished first-person sentence does neither. Three seconds on the punctuation, and you know who you’ve just been asked to be.
Can I see why I was asked this particular thing? A personalized question is never neutral: it rests on something the system noticed in your words. So look for the way back, whether you can get from the question to what it was made from. Sometimes there’s nothing. Sometimes there’s a general formula like “based on your activity,” which names the source without showing it. And sometimes you can see the actual phrases from your entries that the question stands on. The difference here isn’t politeness: a basis you can’t see is a basis you can’t disagree with.
Is there an exit, and is it as visible as the offer? There’s almost always an exit: nobody forbids you from writing your own thing. The question is what it costs. First check whether it actually exits: a “show me another” button gives you a second offer rather than none, and leaves you inside the same frame. Then look at what occupies the center of the screen and what the edge, what’s set large and what’s set gray. Permitted is not the same as offered.
Our answers
Was I given a way to start talking? Can I see why I was asked this? Is there an exit, and is it visible? Here are our answers, and where they fall short.
The first is answered by the whole section above: three unfinished sentences instead of a list of categories. We chose intent over topic so as not to ask people to classify themselves before they’ve managed to say anything.
The second is answered by the Reasoning block under every Seed. It’s always there, and it shows not only what the system asked but what the question rests on. The spec puts this as a commitment: “The AI never makes assumptions or draws conclusions that aren’t traceable to the user’s own words.” The Reasoning block is what that line looks like in pixels.
But it’s collapsed, and that’s a decision rather than an oversight. The question has to land first: read the explanation before the question and the question turns into a conclusion drawn from something you’ve already read, and stops being a question. An empty screen also shouldn’t become a screen for reading: an expanded block doubles the text in the place where a person is being invited to write. And most of all: the explanation is an observation about you, and showing it every time unasked would be doing exactly the thing named as a cost a paragraph above. Collapsed means “here when you want it,” not “here, look at this.”
We see the cost of that too. Most people will never expand it, which means the basis is one tap away rather than immediately visible. By our own second question, that’s an answer of “yes, but.”
The third: the Seed can be refreshed, it can be ignored, the field at the bottom accepts anything. And the tile is still more prominent than the line. That’s a “yes” with the qualification we named above and haven’t yet removed.
What’s missing: a way to say “no question today.” You can always write your own, but the refresh button offers a different Seed rather than none, and there is no empty screen in ours; its place is always occupied by something. In Day One, by comparison, the shuffle button likewise gives another prompt instead of none, but next to it the menu offers “Dismiss Today’s Prompt” and “Don’t show Prompts”. This is consistent with everything above: we think an empty screen asks something regardless, and we’d rather the question were a deliberate one.
There is no neutral empty screen. Every one of them asks you something, and you answer it, even when nobody put a question mark there. The only thing that really separates one product from another is whether the person who built it knew what it had just asked.