Guides

AI companion privacy: what the app knows, keeps, and can be made to hand over

Intimate conversations are the most sensitive data most people will ever type. Here's what AI companion apps actually retain, what they train on, and how to reduce it.

Published 2026-07-04 · Updated 2026-09-11

An AI companion conversation is, for most people, the most sensitive text they will ever type into a computer. More revealing than their search history, more candid than their messages to friends, and — unlike either of those — deliberately drawn out of them by a system designed to encourage disclosure. It is worth knowing exactly where it goes.

This page is about the category, not about any one app. Every claim an app makes about privacy needs checking against its own policy, and the useful skill is knowing what to check.

The five questions that actually matter

Skip the marketing page. These five answers, all of which live in a privacy policy or a support FAQ, tell you nearly everything:

1. Is your conversation used to train the model?

This is the big one. Some apps state plainly that conversations are never used for training. Others reserve the right to use them "to improve our services", which is broad enough to include training and human review. Others offer it as a setting — and default it to on.

Why it matters: if your conversations train a model, fragments of what you wrote can in principle surface in somebody else's output, and humans may read samples for quality review. Look for the words "train", "improve our models", and "human review" specifically.

2. How long is it retained, and can you actually delete it?

"You can delete your data" and "your data is deleted" are different claims. Ask: does deleting your account delete the conversation history, or merely deactivate the login? Is there a retention period after deletion, and how long? Do backups get purged too, and when?

A good policy gives a number of days. A weak one says deletion happens "in accordance with our retention schedule", which is a sentence that means nothing.

3. What identifies you?

An email address is an identity. A phone number is a stronger one. A card payment ties your legal name and billing address to the account permanently, regardless of what username you chose. Most people's mental model is "I used a pseudonym so I'm anonymous", and the payment record quietly makes that false.

4. Who else gets the data?

Look for the third-party list: analytics providers, advertising SDKs, crash reporting, and — crucially — which company actually runs the language model. Many apps are a front end over somebody else's model API, which means your conversation is processed on a third party's infrastructure under a separate privacy policy. That's not inherently bad, but it doubles the number of companies holding your text.

5. What happens under legal compulsion, or a breach?

Every company with your data will hand it over under a valid legal order; a policy saying otherwise is not credible. The real question is how much they hold to hand over. And for breaches, the category's history is the relevant evidence: adult and companion apps have been breached before, and the consequences are worse than for ordinary services precisely because the content is intimate. Assume a breach is possible and plan your exposure accordingly.

What "private" looks like in practice, ranked

AI companion privacy by setup
ApproachWho can read itRealistic privacy
Open-source model on your own computerYouHighest — nothing leaves the machine
Paid app, training opt-out, email-only signup, prepaid cardThe app, its model providerGood
Paid app, default settings, card paymentThe app, model provider, payment processor, analyticsModerate
Free app funded by advertisingAll of the above plus ad networksLow
Free app with no published policyUnknownTreat as public

The ranking is uncomfortable but consistent: the more you pay, the better your privacy tends to be, because a free product has to make money from something other than you paying, and the something is usually advertising or data. The exception at the very top is self-hosting, which is both the most private option and free.

Seven things you can do today

  • Find the training setting and turn it off. It's usually in privacy or data settings, and it's usually on by default.
  • Sign up with a dedicated email address that isn't your main one and isn't your real name. This single step breaks the link between the account and the rest of your digital life.
  • Don't use a phone number if the app allows email instead.
  • Pay with a prepaid or virtual card if your bank offers one. You cannot be anonymous to the payment processor, but you can keep the app from holding your name.
  • Keep real identifiers out of the conversation. The app's memory system is designed to retain what you tell it — your employer, your street, your friends' names — and it will. Treat the chat as permanent and quotable.
  • Don't send photographs of yourself. They are biometric data, they cannot be un-shared, and no feature in any of these apps needs them.
  • Check what shows on your card statement before the first charge if that matters to you. Most apps publish the billing descriptor in their FAQ.

The one rule that covers all of it: write as though the transcript could one day be read by somebody you know. Not because any particular app is untrustworthy, but because that assumption costs you almost nothing and is the only protection that survives a breach, an acquisition, a policy change or a subpoena.

Why the memory feature makes this harder

The thing that makes a companion feel continuous is a stored, searchable record of what you've told it — see how AI companion memory works. Privacy and that feature are in direct tension: a companion that remembers nothing is far more private and far less compelling. There's no clever way around the trade-off. What you can do is choose deliberately rather than by default, and decide which specifics you're willing to put into the permanent record to get the experience you want.

If you want maximum privacy

Run the model locally. An open-weight language model on your own computer with a local interface sends nothing to anyone, has no policy to change, cannot be breached remotely, and costs nothing per month. It needs a reasonably capable graphics card, an afternoon of setup, and you forgo polished apps and easy image generation. For anyone whose privacy concern is real rather than abstract, it is the only answer that doesn't depend on trusting a company.

Common questions

Are AI companion conversations private?

They are private from other users, but not from the company running the app. Conversations are stored on its servers, may be accessible to staff, are processed by whichever company provides the underlying language model, and can be handed over under a valid legal order.

How private they are beyond that depends on specific answers: whether they are used for training, how long they are retained, and whether the app is funded by advertising. Only a local model running on your own computer is private in the strong sense.

Do AI girlfriend apps use my chats to train their AI?

Some do, some don't, and many make it a setting that is switched on by default. Look in the privacy policy for the words train, improve our models, or human review — that last phrase means staff may read samples of conversations.

If there is an opt-out, it is normally in privacy or data settings, and turning it off is the single highest-value privacy action available to you.

Can I use an AI companion anonymously?

Close to it, but not completely. You can sign up with a dedicated email address that isn't tied to your name, avoid giving a phone number, and pay with a prepaid or virtual card so the app never holds your legal name.

You cannot be anonymous to the payment processor, and you cannot be anonymous if you tell the companion identifying details — its memory system is specifically built to retain them.

What happens to my AI companion data if I delete my account?

That depends on the policy, and the distinction worth checking is whether deletion removes your conversation history or only deactivates the login. A trustworthy policy states a specific retention period after deletion and says whether backups are purged.

Vague wording about a retention schedule, with no number of days, should be read as the data being kept indefinitely.

Is it safe to send photos of myself to an AI companion?

No, and there is no feature that requires it. A photograph of your face is biometric data, it cannot be un-shared, and it converts a pseudonymous account into one that identifies you permanently.

This is the one piece of advice in this guide with no trade-off attached — declining costs you nothing.

Which AI companion setup is the most private?

An open-weight model running locally on your own computer. Nothing leaves the machine, there is no company policy that can change, no account to breach and no monthly cost.

It requires a reasonably powerful graphics card and some setup, and you give up polished apps and easy image generation. Among commercial options, a paid app with training opted out and a prepaid card is the realistic best.

Why there are ads on this page

This site is free and has nothing to sell you — no subscription, no account, no paywall. It is paid for entirely by advertising, which includes pop-under and full-screen ad formats from our ad partner as well as our own house banners for other sites we run. We do not control which third-party advertiser you see. Nothing on this page is a paid review, and no company has paid for a favourable mention.