AI Companion App Privacy Settings & Safety Checklist

AI companion conversations tend to be more personal than a typical chatbot exchange, so the privacy questions matter more than usual. The good news is that most of what you need to know is checkable before you type a single message. Work through the checklist below and you'll have a realistic picture of how an app treats your data.

1. Data storage and deletion

Start with the most important question: can you get your data out, and can you get it removed?

• Conversation deletion: can you delete individual chats from within the app?
• Account deletion: is there a clear "delete account" option, or only a way to "deactivate"? Deactivating is not the same as deleting.
• Retention after deletion: the privacy policy should say whether data is removed immediately or kept for a period, including in backups.
• Export: some apps let you download your data — useful for knowing what they hold.

How to verify

Search the privacy policy for the words "retention," "delete" and "backup." A trustworthy policy states a specific retention period and describes the deletion process. Vague language like "we may retain data as needed" without a timeframe is a signal to be cautious.

2. Training and data sharing

Many providers improve their models using real conversations. That's not automatically a problem, but you should know whether it's happening and whether you can opt out.

• Training use: does the policy say chats may be used to train or fine-tune models?
• Opt-out: is there a setting to exclude your conversations, and is it on or off by default?
• Third parties: is data shared with advertisers, analytics providers or other partners?
• Human review: some apps have staff review flagged chats for safety — worth knowing if that's the case.

How to verify

Look in both the privacy policy and the in-app settings. Search the policy for "train," "third party" and "share." Then open the app's privacy or data settings and check for a training or personalization toggle, and note its default state.

3. Age verification and content controls

Age verification and content filters are two separate controls, and it's easy to confuse them. An age gate governs who can access the app or its mature content; a content filter governs what the app will generate. Both vary widely across the category and neither is standardized.

Apps with mature content should make their 18+ status obvious and back it with some form of age gate. A stronger age check is generally a sign the provider takes compliance seriously.

4. App permissions

A text-based companion app needs surprisingly little access to your device. Each permission it requests should map to a feature you actually use.

• Notifications: reasonable for message alerts.
• Microphone: only needed if you use voice input or replies.
• Photo library / camera: only for features that involve uploading or generating images.
• Contacts / location: rarely necessary for a companion app — question these if requested.

How to verify

On iOS and Android you can review and revoke permissions in your device settings at any time. The app store listing's data-safety or privacy section also summarizes what the app collects — compare that against what the app asks for in practice.

5. Where the privacy settings actually live

Most of the controls above exist, but they are rarely on the first settings screen. Across the apps we have examined, they cluster in four predictable places — so if you cannot find a control, check these before concluding the app does not offer it.

Two practical notes. Account deletion is deliberately harder to find than conversation deletion, and is sometimes only available on the web version rather than in the app. And a filter toggle that lives behind a paid tier is a payment check, not an age check — worth knowing before you treat it as a safety feature.

6. Extra checks if you use mature modes (18+)

If you intend to use an app's adult or mature mode, the privacy stakes rise: the conversation log becomes more sensitive, and so does anything the app generates or stores alongside it. Four additional checks are worth the few minutes.

• Does the age gate do anything? Enter an under-18 date of birth during signup. If the app accepts any input without consequence, treat its age assurance as decorative — and remember a store rating is not a control either.
• Is mature content off by default? Default-on is the red flag, not the mere existence of the setting.
• Where do generated images go? If the app generates images in-chat, check whether they are stored on its servers, whether deleting the chat deletes them, and whether they carry provenance metadata.
• What happens on payment? Mature tiers are usually paid. Check which processor handles the transaction and what appears on your statement, since the app's own privacy policy may not cover the payment provider.

7. What the 2026 rules require — and what they do not

Two regulatory changes now shape what companion apps must tell you, and both are frequently overstated. Knowing the actual scope keeps you from assuming a protection that does not exist.

• EU AI Act, Article 50 — a disclosure rule, applicable since 2 August 2026. Systems that interact directly with people must make clear you are talking to an AI unless that is obvious, and generative output must carry machine-readable marking so it is detectable as synthetic. Enforcement sits with national market surveillance authorities, and penalties for breach reach €15 million or 3% of worldwide annual turnover. What it does not do: require age verification, break reminders, crisis protocols or content restrictions. We covered the detail in the EU AI Act disclosure rules explainer.
• California SB 243 — a safety rule for companion chatbots, operative since 1 January 2026. This one is prescriptive where the EU rule is not: AI disclosure, break reminders for users known to be minors, protocols for self-harm content, and measures to keep sexually explicit material away from minors. An app can satisfy Article 50 and still be nowhere near SB 243 compliance.

Neither rule makes an app-store age rating meaningful. A July 2026 audit of 4,346 AI character and companion listings found about 60% carried a rating that classifies them as accessible to minors, and the two big stores frequently disagree about the same app — the reason we treat store ratings as a weak, self-declared signal rather than a scoring input. The numbers are in our write-up of the store-rating data.

8. The 2026 safety guidelines the app stores actually enforce

Ask what the “2026 AI companion safety guidelines” are and you get two different answers, because two different rule-makers are involved. Legislators set the layer described in section 7. The app stores set a second, separate layer — and that one is the layer you can verify from inside the app in about two minutes, because each requirement corresponds to a screen or a button that either exists or does not.

Two things follow. A missing in-chat report control is now a policy gap rather than a design preference, which makes it a fair signal about how the developer treats safety in general. And distribution channel matters more than it used to: an app that ships only as a web app or a direct download sits outside both stores’ review, so nothing above has been checked by anyone other than the developer. Where legislation applies as well, enforcement is separate again — Colorado’s chatbot law, for instance, routes complaints to the state attorney general rather than to a store, as we set out in our breakdown of HB 26-1263.

9. Data protection tips: the five-minute pass

If you have a few minutes before signing up, these are the checks with the best ratio of effort to information. Each returns a yes or a no, so you can run them on any app without knowing anything about it in advance.

1. Search the privacy policy for three words — retention, train, share. A policy that answers all three with a specific period or a named purpose is a different class of document from one that answers none.
2. Find account deletion before you create the account. If you cannot locate it in the policy or in a trial account’s settings, assume it takes an email request — and that the app is aware this is inconvenient.
3. Sign up with a dedicated email address. It costs nothing, keeps companion-app marketing out of your main inbox, and makes any later breach notice easy to attribute.
4. Decline every optional permission on first launch. Grant them back only when a feature you use actually needs one. This is the only item here that improves your position rather than merely informing it.
5. Check the training toggle’s default, not its existence. Off by default is the meaningful state; a toggle you have to hunt for and switch off yourself is a weaker commitment.
6. Test the report control on a real reply. Under the store policies in section 8 it belongs inside the chat, and finding it also shows you what the developer says happens next.
7. Note who processes payment before you subscribe. The app’s privacy policy does not govern the processor, and the statement descriptor is what anyone else looking at the bill would see.

None of this requires trusting a review, including ours. It is the same pass we run before an app enters a comparison, and its results are what we mean when we call an app’s data handling documented or unclear.

Turning the checklist into a decision

You don't need every answer to be perfect; you need them to be clear and reasonable for what the app does. An app that documents retention, offers real deletion, explains training use, gates age appropriately and requests only relevant permissions has earned a baseline of trust. One that's vague on all four deserves more caution — regardless of how good the conversation feels. Privacy is one of the five criteria we weigh in How to Choose an AI Companion App, and where we can confirm these details for a specific app, we note them in our reviews and rankings.

This checklist is general guidance, not legal advice, and the specifics differ by app and by region — always defer to the app's own current privacy policy and your local regulations.

Frequently asked questions