It's worth being unsentimental about this. A paper notebook in a drawer has a threat model of roughly one: someone finds the drawer. An AI journal has your most private sentences travelling over a network, being processed by a language model that likely belongs to a third party, and sitting in a database owned by a company that may not exist in five years.
That's not a reason to avoid AI journaling. It's a reason to know what you're agreeing to. Here's what to actually look for.
1. Does it train on your writing?
This is the first question, and the answer should be easy to find. If an app trains its models on your entries, your writing becomes part of a system you can't inspect or extract yourself from later.
Most reputable journaling apps say plainly that they don't. Look for that sentence explicitly, and check whether it also covers the underlying model provider — an app can honestly say it doesn't train on your data while sending it to an API that does, unless it's on a no-training tier. Good apps state both.
2. How long do they keep it, and can a human read it?
Retention and staff access are separate questions from training, and often less clearly documented.
Reasonable answers: entries are retained while your account exists and deleted on request; access is restricted and logged; support staff can't casually browse your writing. Watch for policies that reserve broad rights to review content "for quality" without saying who or when. And check that deletion actually deletes rather than hides — you want the word delete, not deactivate.
3. Is it encrypted, and how?
Almost everything is encrypted in transit and at rest now; that's table stakes and doesn't tell you much. The meaningful distinction is end-to-end encryption, where the provider can't read your content at all.
Here's the honest tension, and beware anyone who papers over it: an AI companion cannot respond to writing it cannot read. True end-to-end encryption and AI reflection are largely incompatible unless the model runs locally on your device. So an app offering both may mean the AI parts are excluded from E2E, or that E2E applies only to entries you don't run AI on. Day One is known for strong encryption on the journaling side for exactly this reason; the AI features sit differently. It's fine either way — just know which you're getting.
There is no AI journal where nobody could read your entry. There are AI journals where nobody has a reason to, and says so in writing.
4. Can what you write become a public web page?
This one gets overlooked and it's the most consequential, because it's the only failure mode that's permanent and public. Some journaling apps have sharing, profiles, or publishing features. The question is what happens by default.
If an app publishes or indexes anything without an explicit choice on your part, that's a real problem — search engines cache aggressively, and getting an indexed personal page fully out of Google takes weeks and isn't always complete. Check whether sharing is opt-in, whether shared pages are indexable by search engines, and whether you can revoke a link.
We'll be direct about our own answer, since it changed. In Inkling, entries and conversations are private to your account. Turning an entry into an article doesn't publish it — articles are reachable only by a link you choose to share, and search-engine indexing is off by default. If you want your writing to be findable in Google, you turn that on yourself, per article or as a default, and the setting says exactly what it does. We think publishing someone's journal to a search engine should require them to have decided that, deliberately.
The questions worth asking any app
Do you train on my entries — and does your model provider?
What exactly is deleted when I delete an entry, and how fast?
Who at your company can read my writing, and under what conditions?
Is anything I write public or search-indexable unless I say otherwise?
Can I export everything and leave?
That last one matters more than people expect. An app you can walk away from with your writing intact is one you can afford to trust provisionally.
A reasonable way to think about it
Decide what tier of honesty you want to commit to a service. Plenty of people happily write about their week, their work, and their ordinary sadness in an AI journal, and keep the two or three things they'd never risk anywhere on paper. That's not paranoia — it's proportionate.
And whatever you choose, read the privacy page before you write the hard entry, not after. It takes four minutes, and it's the only part of this that you control completely.