Is It Safe to Journal With AI? A Privacy Checklist

It can be — if you can answer two questions about the app you’re using: who can read your entries, and does anything you write train an AI model. Most privacy policies bury both answers. This post is how to dig them out in about five minutes, plus the checklist I’d run any AI journal or AI diary app through — including my own, since I develop one and have obvious biases you should account for.

The two questions that matter

Everything in an AI journal’s privacy story reduces to these:

Who can read it? Not in theory — in practice. Which companies receive your text? Do their employees ever see it (some services use human reviewers for quality checks)? Is it stored on their side, and for how long?

Does it train anything? When your entries are used to train a model, fragments of what you wrote can influence — in rare cases, surface in — what the model says to other people. For a to-do list, who cares. For the thing you wrote about your marriage at 1 a.m., you care.

A journal is the most sensitive writing most people produce. The bar should be correspondingly high.

A general chatbot is not a journal

Plenty of people journal by talking to a general-purpose chatbot. The conversation is genuinely good — it asks decent follow-up questions. But understand the deal you’re making.

Consumer chatbots are typically conversational products first: as of this writing, several use your conversations for model training by default unless you find the setting that opts out, keep chats in a history designed for browsing rather than retrieval, and remember your past selectively at best. There’s a meaningful difference between a product where your text is the product and a product you pay to store your text. Business and API access to the same AI models usually comes with no-training terms — which is why a dedicated journal app calling an AI provider through its API can honestly offer a stronger deal than chatting with that provider’s own app.

The practical advice: if you journal with a general chatbot, check its training toggle today. If you use a dedicated app, the next section is for you.

Five things to check in any AI journal’s privacy policy

1. Named providers. Does the policy say which AI companies process your text, by name? “Trusted third-party providers” with no names is a yellow flag — you can’t evaluate a deal made with someone unnamed.
2. A plain training sentence. Look for an explicit “your content is not used to train models” — and check whether it binds the providers too, not just the app developer. The honest version mentions data-processing agreements that forbid providers from using your text for their own purposes.
3. Real deletion. Can you delete a single entry, and your whole account, from inside the app — and is deletion described as permanent? “Contact us to request deletion” is a worse answer than a button.
4. Consent before processing, not after. Were you asked before your first entry was sent to an AI provider — and can you withdraw that consent without losing the app entirely?
5. What’s retained where. When your entry is processed by the AI provider, does it come back and get discarded on their side, or does it linger in their systems? Policies that address provider-side retention at all are already ahead of most.

Bonus tell, free to check: how does the app make money? A subscription means you’re the customer. “Free forever” with AI processing means ask harder questions — inference costs real money, and something is paying for yours.

On-device vs. cloud AI

On-device AI — where the model runs on your phone and your entries never leave it — is the strongest privacy position, full stop. If a fully on-device journal does everything you want, that’s a great choice.

The honest tradeoff is capability. The models that fit on a phone today are far more limited than cloud models at exactly the things that make an AI journal worth using: understanding a rambling voice entry, answering questions across years of entries, noticing the pattern between March-you and November-you. Cloud-based apps can do those well, at the cost of requiring the trust chain this post is about. Neither side gets to claim the other is wrong; it’s a real tradeoff, and you should pick it consciously rather than by default.

How I handle this in my own app

I build Second Brain, an AI journal where diary-style notes share a chat with everything else you capture — and if it’s going to call itself the best AI diary for private thoughts, it has to win on exactly this dimension: the privacy deal. So here are the actual answers, from the person who wrote the policy:

Before your first note is processed, the app asks for explicit consent to AI processing — nothing is sent until you agree, and you can withdraw consent in Settings. The providers are named in the privacy policy: Anthropic processes notes to generate answers, OpenAI handles transcription and the embeddings that make notes searchable. Both act under data-processing agreements that forbid using your content to train models for other customers. Your notes are stored isolated per user, you can delete any note or your entire account from inside the app, and deletion is permanent. We don’t sell data and don’t run ads; the business model is a subscription, which is the whole reason the incentives line up.

Don’t take this paragraph’s word for any of it — that’s the point of the checklist. Read the policy and check the boxes yourself; it’s short on purpose.

The checklist

Before you pour your heart into any AI journal:

• AI providers named in the privacy policy
• Explicit “not used for training” statement that covers the providers
• In-app deletion of entries and account, described as permanent
• Consent asked before first processing, withdrawable later
• Provider-side retention addressed
• Business model that doesn’t need your data to be the product

Six checks, five minutes. If an app passes, journaling with AI is about as safe as anything you do on a phone gets — and the format genuinely helps people who’d never keep a paper journal, especially the ones who hate writing. If an app fails the checks, it doesn’t deserve your 1 a.m. thoughts. Keep them somewhere that’s earned them.