AI Companion Privacy Checklist: 12 Things to Check First

For adults. AI companion apps are generally 18+ services. This checklist is written for adult users and for anyone helping an adult choose an app.

People tell companion apps things they would not post anywhere else. That makes the privacy policy the most important document an AI companion app publishes, and the one fewest people read.

This checklist turns that policy into twelve concrete questions. Each one comes with what to look for, a real example from an official policy we read on 18 October 2026, and what a weak answer looks like. We have not tested how any company handles data in practice; this is about what the companies commit to in writing.

Why this matters: what researchers found

In February 2024, the Mozilla Foundation’s *Privacy Not Included team reviewed 11 romantic AI chatbots, including Replika, Chai and Eva, and gave every one of them its privacy warning label. According to Mozilla’s announcement:

• Ten of the 11 failed Mozilla’s Minimum Security Standards, such as requiring strong passwords.
• Researchers found at least 24,354 data trackers within a minute of using one app, Romantic AI.
• Most companies did not offer an opt-out from having chat contents used to train their AI. Only one, Genesia AI, did.
• None of the companies responded to Mozilla’s requests for more information.

That review is now more than two years old, and the apps and policies have changed. Mozilla has also moved its product reviews: the original *Privacy Not Included guide page now says the project has moved to Mozilla’s Nothing Personal magazine, with the archive available through the Internet Wayback Machine. The findings are still a useful map of where problems tend to sit.

Regulators are watching too. In September 2025 the US Federal Trade Commission ordered seven chatbot companies to explain, among other things, how they use or share personal information from chatbot conversations and how they enforce age restrictions.

The 12 checks

1. Is chat content listed as collected data?

Look for words like “messages”, “prompts”, “content” or “conversations” in the data-collected section. They almost always are, but you want it in writing.

Example: Candy AI’s Privacy Notice lists “chat prompts and AI outputs”. Replika’s policy lists text, photos, videos and voice messages you send.

Weak answer: the policy never mentions conversations at all.

2. Are chats sent to third-party AI providers?

Many apps run on models from outside companies. That means your messages may leave the app’s own servers.

Example: Candy AI’s notice says third-party LLM providers or hosts may receive chat message content. Replika’s policy says third-party AI providers must not use your data to train their own models.

Weak answer: “trusted partners” with no categories named.

3. Are chats used for training, and can you opt out?

Look for “train”, “improve our models” or “machine learning”. Then look for an opt-out.

Example: Candy AI’s notice says de-identified interactions may be used to train its models and moderation systems, with possible human review of de-identified data. Replika says small anonymised portions are used internally for safety and performance. We did not find a training opt-out described in either policy.

Weak answer: training is mentioned, but there is no opt-out and no mention of de-identification.

4. Does a human ever read conversations?

Moderation and dataset preparation can involve people, not just software. A good policy says so.

Example: Candy AI’s notice mentions human review of de-identified interactions during dataset preparation.

Weak answer: the policy is silent, which tells you nothing either way.

5. Is personal data sold or shared for advertising?

Check for “sell”, “share” and “advertising partners”, and check any US state privacy section.

Example: Nomi’s policy states it does not and will not sell or rent personal information. Kindroid’s says it will not sell or share personal data. Replika says advertisers will never access conversations, while noting some website data sharing may count as a “sale” under certain laws.

Weak answer: no statement at all on selling.

6. How long is data kept?

Retention is often the most revealing section. Look for specific periods.

Weak answer: “as long as necessary” with no period at all.

7. Can you delete your account yourself?

Self-service deletion in settings is better than an email request, because it does not depend on someone answering.

Example: Replika and Nomi both describe deleting your account in account settings. Candy AI’s notice gives an email route (support@candymail.ai) for deletion requests.

Weak answer: no deletion route described.

8. Does deletion really delete?

Read the exceptions. Some data may be kept for legal reasons, and some may sit in archives or backups.

Example: Nomi’s policy says information in “our archives” cannot be changed or deleted. Candy AI’s notice says erasure can be refused where data is needed for legal obligations, legal claims or contract performance.

Weak answer: deletion is promised with no exceptions mentioned, which is rarely the whole story.

9. Can you export your data?

Export lets you see what the company holds.

Example: Kindroid’s privacy documents describe a data export available every 30 days from the profile preferences menu.

Weak answer: no export route, only a formal access request.

10. Where is your data processed?

Look for international transfers. This matters for which privacy laws apply.

Example: Candy AI is operated from Malta, and its notice says data may be transferred outside the EU, including to the United States.

Weak answer: no mention of transfers or of where the company is based.

11. How is the account secured?

Mozilla flagged weak passwords as a common failure. Check whether the app allows strong passwords, sign-in through an existing provider, or two-factor authentication.

Example: Nomi’s policy says you sign in with a Google or Apple account. Candy AI’s notice mentions encrypted passwords.

The FTC also recommends turning off app permissions that are not needed, and deleting apps that request permissions they do not need to work.

12. What are the age rules, and how are they enforced?

All the policies we read set 18 as the minimum. Enforcement is usually a self-declared age gate.

Example: Candy AI’s Underage Policy says users must affirm they are over 18, and that it may add further verification where local rules require it.

Weak answer: no stated minimum age.

The checklist at a glance

Habits that protect you whatever the policy says

• Use a nickname, not your full name.
• Do not share your address, workplace, financial details or anyone else’s private information in chats.
• Use a separate email address for companion apps.
• Review app permissions after installing.
• Delete the account, not just the app, when you stop using it.

Best for / Avoid if

This checklist is best for adults comparing companion apps who want to know where their conversations go before they start.

It is not a substitute for a security audit. A policy is a promise, not proof. For the full pre-payment process, see how to evaluate an AI companion app before you pay, and for product-level detail read our Candy AI review and Candy AI alternatives. Our explainer on AI companion memory covers how remembered details relate to stored data.

Bottom line

Every companion app collects your conversations. The differences are in who else receives them, whether they train models, how long they are kept and how easily you can delete them. Run the twelve checks on the current policy before you share anything personal, and if an app cannot answer most of them in writing, choose one that can.