Journaling apps have been adding AI features at a rapid pace. Mood tracking that charts your emotional patterns over time. Semantic search that finds entries not by keyword but by meaning. Conversational AI that lets you query your journal in natural language. Weekly insights that surface connections between your entries you hadn’t noticed yourself.
These are genuinely useful features. They are also fundamentally incompatible with real encryption — and most apps that offer both are quietly doing one of two things: compromising the encryption, or running the AI on your device in ways that may not match what the marketing suggests.
Understanding the technical reality helps you make a clear-eyed choice about which matters more to you: the AI features, or the guarantee that your entries remain private.
Why Encryption and AI Processing Cannot Coexist on the Server
Encryption works by transforming readable text into scrambled ciphertext. Without the key, the ciphertext is useless. With the key, it becomes readable again.
For a cloud-based journaling app, there are two fundamentally different encryption models:
Server-side encryption. The service provider encrypts your entries before storing them. The provider also holds the encryption keys. This protects your data from outside attackers who might breach the database — they get ciphertext, not your entries. But the service itself can decrypt your entries at any time.
Client-side (end-to-end) encryption. Your device encrypts entries before sending them to the server. The encryption key exists only on your device. The service stores ciphertext and has no key. Even in a breach, even under a legal order, the service cannot give a readable version of your entries to anyone.
The difference matters enormously when it comes to AI features.
If the service can decrypt your entries — server-side encryption — it can run AI processing on the plaintext. Mood analysis, semantic indexing, pattern detection: all of this requires readable text. The server processes that text, generates insights, and sends them back to you. The AI features work. Your entries are also fully accessible to the service, their employees, and anyone who obtains a legal order compelling disclosure.
If the service cannot decrypt your entries — genuine end-to-end encryption — it cannot run server-side AI. The ciphertext that arrives at the server is noise from the perspective of any AI model. No mood analysis. No semantic search built server-side. No insight generation.
On-Device AI: The Middle Path
There is a technically honest version of AI journaling that preserves encryption: running the AI processing entirely on your device.
If mood analysis, pattern detection, and semantic search all happen locally — on your iPhone, your Mac, your Android phone — then the server never receives plaintext. Your entries stay encrypted in transit and at rest. The AI works on readable text only where you are.
This is how Apple’s on-device processing works for Photos: the face recognition, scene analysis, and memory generation happen on your device rather than on Apple’s servers. Apple has consistently pointed to this model as privacy-protective, and it largely is.
For journaling apps, on-device AI processing is possible. The constraints are that:
- Your device must have enough processing power to run the AI models.
- Features only work on the device where entries are stored — real-time cross-device AI insights require syncing in some form.
- The AI models run on-device rather than in a datacenter, which limits their scale and sophistication.
Some journaling apps have moved toward on-device processing. The challenge is that this architecture is harder to build, more expensive to develop, and less powerful than server-side AI. It is also harder to market. “AI insights, processed on your device” is accurate but less punchy than the broader category of “AI-powered journaling.”
Reading the Marketing Language
Journaling apps frequently claim both “encrypted” and “AI-powered.” These claims are not always dishonest — but they require reading closely.
“Encrypted in transit.” This is TLS — the same encryption that protects your credit card when you buy something online. It means your data is encrypted while moving between your device and the server. It says nothing about what the server does with your data once it arrives. Nearly every cloud service uses TLS. This is not a privacy differentiator.
“Encrypted at rest.” Your data is stored in encrypted form on the server. As described above, this protects against external breaches but not against the service itself, which holds the keys. All AI processing remains possible.
“End-to-end encrypted.” This is the strong claim. If true and accurately implemented, it means the service cannot read your entries. But the qualifier “if true” matters. Some apps use this phrase loosely to describe encryption that covers transit and storage but not client-side key control.
“Zero-knowledge.” Similar to end-to-end encryption: the service has no access to the encryption keys. AI processing server-side is impossible under a genuine zero-knowledge architecture. If a zero-knowledge app offers server-side AI features, ask how.
The question to ask any app that claims both encryption and AI: Where does the AI processing actually happen? If the answer is on a server, the service can read your entries. If the answer is on your device, ask which features are truly local and which require any server communication.
How Major Journaling Apps Handle This
Each major journaling app handles this differently, and their approaches reflect genuine trade-offs.
Day One uses end-to-end encryption through a passphrase-based sync mechanism. Users who enable this cannot share their journals with the service, and AI features are accordingly limited. Day One’s AI analysis features, where they exist, are built around models that process content locally or with explicit user consent to share.
Apple Journal syncs through iCloud with on-device processing for many features. Apple’s privacy documentation describes Journal as keeping your entries private to you, though the specific implementation of encryption for iCloud sync has nuances depending on whether Advanced Data Protection is enabled.
Notion and Obsidian represent different philosophies. Notion is a cloud-first tool with AI features powered by server-side processing — your content is accessible to Notion and its AI infrastructure. Obsidian is local-first, with an optional sync service (Obsidian Sync) that uses end-to-end encryption. Obsidian’s AI features, through plugins, can run locally using local AI models. The difference is clear: Notion’s AI is powerful and cloud-dependent; Obsidian’s AI is more limited and optionally private.
Journey offers end-to-end encryption as an opt-in feature. When enabled, AI features are limited to what can run locally or what you explicitly choose to process server-side.
The consistent pattern: apps with strong AI features tend to use server-side encryption, and apps with strong encryption tend to have limited or on-device AI features.
The Questions Worth Asking Your App
If you journal about things that matter to you — relationships, health, finances, personal struggles — these questions are worth answering before your entries accumulate over years:
- Is my data encrypted with keys only I hold? Or does the service hold the keys?
- If the service cannot decrypt my entries, how does the AI analysis work? Ask support. If the answer is unclear, assume the service can read your entries.
- What happens to my entries if I cancel? Does deletion actually delete, or is there a retention period?
- Has the service made specific disclosures about how AI processes my content? Look for a security or privacy page with technical details, not just reassuring language.
- Is the app open source? For apps claiming zero-knowledge encryption, open-source code means the claims can be independently verified.
Making the Choice
This is a genuine trade-off, not a situation where there is a clearly wrong answer.
If AI insights matter to you — if the weekly pattern analysis helps you, if the semantic search saves time, if the mood tracking is useful — you can choose an app that offers these features and understand that you are making a privacy trade-off. Your entries are readable by a third-party service, and that is the cost.
If the privacy of your journal entries is a priority — if you write about things you would not want disclosed, if you are in a jurisdiction where certain personal information carries legal risk, if you simply want to know that only you can read what you write — you can choose an app with genuine end-to-end encryption and accept that some AI features will be limited or absent.
What you should not do is assume both are true simultaneously. When a marketing page shows AI features and an encryption badge without explaining how both are possible, the most charitable interpretation is that the AI runs on-device. The less charitable but often accurate interpretation is that the encryption is real only in the technical sense that data is not stored in plaintext — while still being fully accessible to the platform.
The journals people keep are among the most intimate documents they create. How those documents are protected is worth understanding before they exist as a years-long record on someone else’s servers.
Storing Journal Entries Privately
For people who want private storage without the AI trade-off, one approach is to separate the writing tool from the storage layer.
A local writing app — Obsidian, iA Writer, or even plain text files — keeps your entries on your device. For backup and cross-device access, those files can be synced to cloud storage that encrypts your content server-side without analysing it.
daftei stores files you upload — including plain text, PDFs, and document files — without analysing their contents, without using them to train AI models, and without connecting them to advertising. Files are encrypted in transit using TLS 1.3 and at rest using AES-256. The app does not read what you write.
This approach means you manage the writing tool and the storage separately. For users who want both real privacy and thoughtful journaling, that separation may be worth the slight added friction.
The underlying truth is that where you write and where you store what you write are separate decisions. Making them separately — with clear eyes about what each service can and cannot see — is how you get what you actually want from both.