A journal app with AI: what the AI should see, what it shouldn't, and how to check
Few features split journal users like AI. In App Store reviews this year, one long-time user of a popular journal app wrote that "the one sanctuary I have each day is now infiltrated with AI" and that they had "lost total trust" in the company. A user of a different, AI-first journal wrote the opposite: "the AI helps", it helped them "see patterns" they hadn't noticed.
Both are right. AI in a journal can be genuinely useful, and it can also feel like a stranger reading over your shoulder. The difference is usually not the AI itself. It is whether you chose it, whether you know what it reads, and whether you can say no. Here is what AI is good for, what it should never do, and how to find out where your words go.
What AI in a journal is actually good for
A question instead of a blank page. The hardest part of writing about your life is often the second sentence. You've written what happened, and then you stop. A short question about what you wrote ("What did you do right after?", "Who else was there?") can open up a memory more than any list of generic prompts, because it is about this moment and not about moments in general.
Noticing what repeats. Over months, people tend to write about the same few things without seeing it. An AI that reads a month of entries can point out that the good days all involved being outside, or that a certain kind of week keeps ending the same way. It won't always be right.
Turning a voice note into text. If you'd rather talk than type, transcription makes a spoken story searchable and easier to reread.
What it is bad at, and what it should never do
- It doesn't know you. It reads a few lines of text and makes an informed guess. A pattern it finds in ten entries is an observation, not a fact about your life.
- It is not a therapist and not a doctor. It can't diagnose anything. If you are going through something serious, a person who is trained to help is the right place to go. If you are in danger, call your local emergency number.
- It shouldn't decide for you. A gentle suggestion is one thing. Telling you to leave your job or your partner is another, and no journal app should do it.
- It shouldn't be forced on you. A journal that pushes AI buttons, banners and upsells into every screen with no way to turn them off has forgotten whose journal it is.
Where your words go when an AI reads them
Most AI features in journal apps don't run on your phone. The app sends your text to a large language model on a server, often run by another company, and gets an answer back. To be read, the text has to be readable. If your journal is encrypted, the app has to decrypt the entry before the AI can see it.
Apps that take this seriously say so plainly. Day One's AI features guide, for example, says its AI features "require opt-in and can be disabled at any time", and that entry content is "temporarily decrypted on your device for processing" before it is sent to its servers and then to an AI service. Some features are built differently: Apple says its Journaling Suggestions use "on-device processing" to group moments and suggest what to write about.
It also matters how the app pays for the AI. Google's Gemini API terms are a clear example. On the unpaid tier, Google uses submitted content "to provide, improve, and develop Google products and services", and "human reviewers may read, annotate, and process your API input and output". On the paid tier, Google "doesn't use your prompts... or responses to improve our products", but it "logs prompts and responses for a limited period of time" to detect abuse and for "any required legal or regulatory disclosures". Google's usage policies currently give that period as 55 days.
Seven questions to ask any journal app with AI
- Is AI optional? Is anything sent before I agree?
- What exactly does each feature send, and when: only when I tap, or in the background too?
- Which company's model receives my text?
- Is my text used to train or improve AI models, by the app or by the AI provider?
- How long does the AI provider keep it, and does the app keep the results?
- Can I turn each feature off, and all of AI at once?
- What happens to entries I've locked or encrypted?
A good app answers all seven in plain words, inside the app.
How Lifeline answers them
Here are our own answers, as precisely as we can give them.
- Optional, and nothing before you agree. The first time you use an AI feature, the app asks once: a sheet called "Before the AI reads your memories" explains what is sent and to whom. You confirm that you are 18 or older (Google's Gemini API terms don't allow it to be used in services for people under 18) and allow it, or tap "Not now". Until you agree, nothing from your memories goes to the AI.
- What is sent. Each feature sends only what it needs. Questions to a memory: its title, date, text and emotion, with your own word for the feeling; with Premium, also the dates, titles, emotions and first lines of up to three earlier memories with the same feeling or words. For the second round, also your answers to the first questions. Questions made automatically when you save a new memory: its title, date, text, emotion and your "impact" and "lesson" reflections. Pattern of the month: the dates, titles, emotions and the beginning of that month's memories, with your "impact" and "lesson" reflections. Life Compass: dates, titles, emotions and first lines of your memories. "What followed similar moments": the memory and up to five earlier ones with the same emotion (titles, first lines and short excerpts from reflections). "To text": the recording of that one voice note, only when you tap it. Once you've agreed, a few features also run without a tap: questions for a new memory when you save it, if that switch is on, and with Premium, "What followed similar moments" for a memory you save and a weekly re-check of monthly patterns you've already made. Right after you get Premium, "What followed similar moments" is also prepared for up to 30 of your recent memories.
- Which provider. Google's Gemini model, reached through our own server.
- Training. We use Gemini as a paid service. Under Google's terms for paid services, Google doesn't use the prompts or responses to improve its products. We don't train any model on your memories either.
- How long. Google keeps logs for a limited time to prevent abuse and for legal reasons, as described above. The AI's results (questions, patterns, the compass) are saved in your account so you can see them again. For "To text", our server doesn't keep the recording. When you lock or delete a memory, the questions and predictions about it are removed. A monthly pattern or Life Compass that included it stays until it is rebuilt.
- Switches. In Profile, one switch turns all AI off: then there are no AI buttons, cards or offers anywhere, and your memories and saved answers stay. Turning it off also withdraws your consent, so the app asks again if you turn it back on. There are separate switches for "Questions to memories", "Patterns of the month" and "Compass and similar moments".
- Locked memories. Background features never send the text of a locked memory, and locked memories are never used as context for other questions. If you ask for questions on a locked memory yourself, the app asks first: "This memory is encrypted". Only after you tap "Send once" is its text decrypted on your phone and sent, for as long as you stay on that memory. In your archive it stays encrypted, and so do the questions and your answers. "To text" in a locked memory asks the same way, but what it sends is the recording itself: voice notes are not encrypted, even in a locked memory.
What is free and what is Premium: the first round of questions and "To text" are free. Premium adds the deeper second round, a question that can refer to one of up to three earlier memories with the same feeling or words, Pattern of the month, Life Compass and "What followed similar moments".
The "What the AI sees" screen
In Profile, the "What the AI sees" screen lists each feature (questions to a memory, Pattern of the month, Life Compass, "What followed similar moments", voice notes) and what it sends, followed by the rule for encrypted memories and the daily limits. The master switch sits at the bottom, and the consent sheet links here, so you can read it before you agree.
Frequently asked questions
Is AI in Lifeline turned on by default?
The AI buttons are there, but nothing from your memories is sent until you agree once in the consent sheet. If you don't want to see AI at all, turn off "AI features" in Profile.
Can I use Lifeline without any AI?
Yes. Everything else works without it, and one switch in Profile hides every AI button.
Does the AI read my locked memories?
Not on its own. Background features never send a locked memory's text. If you ask for questions on a locked memory, the app asks you first, and sends the decrypted text only after you confirm, for as long as you stay on that memory. "To text" also asks first; it sends the voice note itself, and voice notes are not encrypted.
Are my memories used to train AI?
We don't train models on them, and under Google's paid-service terms Google doesn't use the prompts or responses to improve its products. It does keep logs for a limited time against abuse and for legal reasons.
Is the AI a therapist?
No. It asks short questions, points out what repeats in your memories and may suggest one small step. It can't diagnose anything, and it isn't a crisis service. If you are in danger, call your local emergency number.
Why do I have to be 18 to use the AI?
Google's Gemini API terms don't allow it to be used in services for people under 18, so the app asks you to confirm your age before the first AI request.
Whatever journal you use, the AI should be a guest you invited, not one who came with the furniture. If an app can't tell you what its AI reads, that is your answer.