Where AI journaling apps send your entries: we read 13 privacy policies

A journaling app’s AI has to read your entries to do anything with them. So the useful privacy questions are practical ones: where does the reading happen, which companies receive the text, is anything end-to-end encrypted, and does the company say whether your words train a model?

We read the privacy policies, help pages, and store listings of 13 journaling apps and recorded the answers, using only what each company says about itself. This page has the full table, what stands out, a note on each app, and the data as a CSV. We make Merrin, one of the 13. Its row follows the same rules, and two of the findings are about us.

AI journaling app privacy: the full table

Checked November 2, 2026. “Not documented” means we couldn’t find it on the company’s own pages, not that the answer is no.

Download the data as a CSV. It has a few more fields (platforms, the AI features, whether they’re on by default) and the source pages for every row.

What stands out

Most AI journals read your entries in the cloud. Ten of the 13 apps have AI built in. Five of them, Mindsera, Life Note, Reflection, Stoic, and Journey’s Odyssey, send entries to a server for every AI feature. Three mix the phone and the cloud. Apple Journal groups its suggestions on the device but doesn’t say where its Apple Intelligence writing prompts run. Google’s Journal app uses its Private AI Compute, “on your device and in the cloud.” Day One runs some features through Apple Intelligence on Apple devices and the rest on Automattic’s servers or through OpenAI. Merrin and Yaps run their core AI on the phone.

The AI providers are the familiar ones. Leaving out Apple’s and Google’s own apps, three name OpenAI (Mindsera, Day One, Stoic), three name Google (Mindsera, Reflection, Stoic), and two name Anthropic (Mindsera, Stoic). Two apps don’t name anyone: Journey says Odyssey runs on a GPT model without saying whose, and Life Note’s policy doesn’t say which provider powers its mentors.

Encryption and cloud AI don’t overlap. Three apps are end-to-end encrypted by default: Day One on every plan, Apple Journal in iCloud (with two-factor authentication on), and Obsidian for Sync. Journey offers it with a paid membership. None of the cloud-first AI journals does. Mindsera says plainly that it “does not currently use end-to-end encryption,” and Life Note decrypts entries on its servers when an AI mentor replies. Where an app has both, the AI works outside the encryption: Day One decrypts an entry on your phone before sending it for AI, and Journey’s Odyssey isn’t available with encrypted sync.

Every app with AI addresses training except ours. Six apps say outright that entries aren’t used to train AI, and Day One says not without your explicit permission. Apple’s statement is company-wide: it doesn’t use private personal data to train its foundation models unless you opt in. Yaps’s covers content passing through its relay. Merrin’s privacy policy doesn’t address training at all. Merrin’s entries are designed to stay on the phone during normal use, so they don’t reach us, but the policy should say that in plain words.

Dates are patchy. Google’s Journal help page, Diarium, and Stoic show no date, and Daylio’s policy link didn’t load when we checked. Day One’s privacy policy was last revised on April 22, 2024 and doesn’t mention AI; its AI terms are in a help guide. At the other end, four were updated in the last two months: Mindsera’s, Life Note’s, Yaps’s, and Merrin’s.

App-by-app notes

• Merrin (privacy policy): the AI runs on the phone, there’s no account, the database is encrypted on the device, and backups are an encrypted file you keep. One gap: on Android phones without on-device speech recognition, including all phones on Android 11, dictating an entry falls back to Android’s default recognizer, which may process audio on its provider’s servers.
• Yaps (privacy policy): a voice and notes toolkit more than a journal. Core features run on the device, and connecting an outside AI client is opt-in. Its policy discloses that dictation on iPhone and iPad can fall back to Apple’s remote recognizer.
• Apple Journal (Journaling Suggestions privacy): suggestions draw on workouts, music and podcasts, people you contacted, photos, and significant locations, grouped on the device. Apple’s training statement covers Apple Intelligence generally, not Journal specifically.
• Google Journal (help page): Pixel 8 and later. AI features are opt-in and need a Google sign-in. Google notes that dictation through Gboard may be processed on its servers, depending on your settings.
• Day One (AI features guide): AI is opt-in, and Day One says it doesn’t keep AI requests after processing. The privacy policy itself doesn’t cover AI.
• Journey (privacy policy): Odyssey AI is off by default and needs unencrypted sync to work.
• Mindsera (privacy policy): zero data retention covers eligible Anthropic requests. For OpenAI, retention “depends on the endpoint and our account configuration.”
• Life Note (privacy policy): entries are encrypted on its servers and decrypted only when an AI mentor replies. It doesn’t say whether its AI provider keeps data.
• Reflection (terms and privacy): it uses Google Gemini and cites Google’s paid-API terms on training. It shares advertising identifiers with Meta and RevenueCat to measure campaigns, and says journal content isn’t shared with Meta.
• Stoic (privacy policy): AI is an optional paid add-on. Its help center says OpenAI may keep data for up to 30 days for safety, and that no one reviews your content manually.
• Diarium (privacy): no AI and no account. Sync goes through your own cloud storage. It doesn’t say whether the journal is encrypted on the device.
• Daylio (security FAQ): no AI. It says your data is “stored only locally on your phone,” with optional backups to Google Drive or iCloud.
• Obsidian (privacy policy): no built-in AI, and community plugins vary. Sync is end-to-end encrypted by default, but Obsidian doesn’t encrypt local vaults.

What a privacy policy can’t tell you

These are the companies’ own descriptions, not audits. We didn’t test what any app sends over the network, Merrin included. A zero data retention agreement is a contract between an app and its AI provider, and nobody outside it can check it. And policies change, so check the date on a row before relying on it.

If you journal in a general assistant instead of a journaling app, our policy review of ChatGPT, Gemini, Claude, Copilot, and Grok covers training, retention, and ads for each. If you’re choosing an app, our comparison of private AI journaling apps and the guide to Apple Journal alternatives for Android go into features and price.

How we collected the data

Which apps. The AI journaling apps that came up in searches for the best AI journal app and for Apple Journal alternatives on November 2, 2026, plus three well-known journals without built-in AI (Diarium, Daylio, and Obsidian) for comparison. Yaps is included because it shows up in those searches, though it’s a voice toolkit more than a journal.

Which sources. Only each company’s own privacy policy, terms, help center, product pages, and official app store listings. Reviews and “best of” lists weren’t used as evidence. Where a page wouldn’t load, we said so rather than going around it.

What we recorded. Platforms, the AI features and whether they’re on by default, where the AI runs, which AI companies are named, end-to-end encryption, what the company says about training, whether an account is needed, and the policy date. Every row in the CSV lists its sources.

Merrin. Its row comes from our own privacy policy and the app’s code, and we held it to the same rules: where our policy is silent, the table says so.