How to start journaling with AI: a 7-day plan

The order is not arbitrary. Day one asks for nothing but volume, because the single biggest risk in the first week is quitting before day three, and quitting happens when the bar is set by ambition rather than by what is achievable at 11pm on a Tuesday. Day two adds exactly one new behaviour — the mood tap — because introducing two new habits at once is how people abandon both.

By day three the page has already gone blank once, so the prompt arrives exactly when it is needed rather than as advice nobody asked for. Day four and five are deliberately the two hardest asks — writing about someone else, writing what you would not say out loud — placed in the middle of the week rather than the start, because trust in the habit and the tool both need a few days to build before you use them on anything that matters.

Day six and seven are not writing days at all. They are the two days that turn a week of scattered entries into evidence: reading your own account back, and treating the AI's first attempt at pattern-matching as a claim to check rather than a verdict to accept.

Open the editor and write three sentences about something that happened today that you are still thinking about. Not the most important thing. The thing that is still rattling.

Do not summarise, do not structure it, and do not write it for the AI. This is the single most common mistake people make with an AI journal: they produce tidy bullet-point reports because they assume the machine wants data. The opposite is true. A messy paragraph in your own voice carries far more signal than a neat summary, because the way you describe something is most of what a pattern is made of.

Same three sentences, plus one tap to tag how you felt. That tap is doing more work than it looks like: it is what later lets you filter six months of writing down to only the days that went badly, and read them together. Mood tracking attached to entries is worth roughly ten times a standalone mood chart, for exactly this reason.

There will be a day this week when nothing happened and you have nothing to say. That day is the one that decides whether this habit survives, and the answer is not willpower — it is a prompt. Something concrete: what did I avoid today, and what did I tell myself about why?

Two sentences counts. An entry that says "nothing happened, I am tired, I avoided the email again" is genuinely useful data. Three months of those, and the email turns out to have a name.

Pick someone who took up space in your head this week and write about the interaction rather than about them. This is where journals start earning their keep, because relationships are where recurring patterns are most obvious in hindsight and most invisible in the moment.

A journal that only contains presentable thoughts is a diary of your public self, and it will never tell you anything. This is also the day the privacy question stops being abstract, so it is worth having actually checked how your journal is stored before you get here — whether isolation is enforced by the database, whether staff can read entries, whether your writing trains anyone's model.

No writing. Open the archive and read the six entries you have produced, in order. Almost everyone finds one thing they had already forgotten writing, four days later. That experience — the gap between what you remember about your week and what you recorded about it — is the entire case for journaling, and no argument makes it as well as reading it.

Seven entries is thin, and an AI insight drawn from seven entries should be treated as a guess. Read it, click through to the entries it cites, and ask whether the evidence supports it. Often it will not, and dismissing it is the right response.

Getting into that habit now is the point of day seven. In three months the insights will be drawing on ninety entries and will occasionally tell you something genuinely uncomfortable and correct — and the only way to tell those apart from confident nonsense is a habit of checking the citations. An insight you cannot audit is an assertion.

You will miss a day this week. Not might — will, for almost everyone attempting this for the first time. What matters is what happens next, because the thing that kills a journaling habit is never the missed day itself. It is the story you tell about it.

The bad version: you missed Tuesday, so Wednesday needs to be extra good to make up for it, and by the time you have built that pressure the app has become something to dread rather than something to open. The habit dies of guilt long before it dies of forgetfulness.

The version that survives: you missed Tuesday, so Wednesday is exactly the same three sentences it would have been anyway. No catch-up entry, no explanation to yourself, no docked point. A missed day is data — maybe it tells you something about Tuesdays specifically — and the only thing to do with it is notice and continue.

Lower your target. Four entries a week, indefinitely, beats seven for a fortnight and then nothing. Skip days without ceremony; there is no streak here worth protecting at the cost of the habit itself, which is also why streaks in Compline are a display rather than a score.

The insights get interesting somewhere between entry fifteen and entry thirty, and the ones worth the subscription — the seasonal ones, the ones spanning a job change — need months, because they are describing things that took months to happen. Write badly, write often, and let the archive do the remembering.