AI Character Chats Feel Private—Use This Disclosure Checklist First

An AI character chat can feel like a private, one-to-one exchange, but the interface does not tell you how the provider stores, uses or reviews the conversation. Before sharing personal information, check model training, other processing, retention, deletion, human access and the rules for the user’s age group.

Do not treat one privacy toggle as a complete answer. Character.AI provides a useful example: its privacy summary says it collects information users provide and data generated through interactions, uses information to operate and improve the service, and offers rights that may vary by region.

Build a disclosure worksheet before opening the chat

Create one worksheet for each service and account you are evaluating. Record the provider’s answer, the page or setting where you found it and the date checked; if the documentation is silent, write “not stated” instead of assuming the most protective interpretation.

1. Model training: Can chat content improve generative models? Who can opt out, and does the choice cover earlier conversations or only new content?
2. Other processing: Can the same data still support search, recommendations, personalization, moderation, fraud prevention or safety systems?
3. Retention: How long are chats, account records, reports, deleted material and backups kept?
4. Deletion: Can you remove a message, a conversation or the entire account? What happens to backups and material already incorporated into model development?
5. Access and sharing: Under what conditions may employees, contractors, moderators or service providers handle conversation data?
6. Age and safety: What is the minimum age, how is age assessed, and which chat features, filters and reporting tools are available to minors?

Capture the controls visible to the actual account, not settings another user reports seeing. Availability, labels and defaults may differ by region, platform or age category.

Separate the training choice from every other data use

A training opt-out answers a narrow question: whether covered content can be used to improve generative models. It does not necessarily stop storage or processing for other purposes stated by the provider.

Character.AI’s model-training page says users in the EEA or UK can disable “Improve the Model for Everyone”; new content then will not train its generative models, while data may still improve search, recommendations and safety classifiers. Feedback submitted through ratings or reports is handled separately from that choice.

On the worksheet, record “new content” as the documented scope and list every use that can continue. If the provider does not explain what happens to earlier chats, do not infer retroactive removal. This distinction prevents opting out of model development from being mistaken for confidentiality.

Ask what deletion actually removes

“Delete” may mean hiding a message, clearing a conversation, closing an account or exercising a legal erasure right. Check which action is available and whether the documentation addresses active systems, backups, safety records and material previously used for model improvement.

Record the request route, any identity-verification step, the stated response period and the region attached to the account. Conditional language such as “may have the right to delete” is not a promise that every request removes every copy.

Save any information you are entitled to keep before closing an account, along with the deletion confirmation. If the provider does not explain an exception or retention period, mark it unresolved. The issue is broader than whether a conversation is publicly visible: nonpublic content may still be processed under the service’s terms and policies.

Check the current age experience, not an old description

Age gates and content controls address different risks from privacy. Verify whether minors can chat at all, how the service determines age, what happens after an account changes age category and how users or guardians can report content.

Character.AI’s current Reading Mode instructions say verified under-18 users can browse and view saved past chats but cannot use chat features. That is a material change from older descriptions of a restricted teen-chat experience, so the worksheet should use the newest applicable product instructions rather than treating an older safety page as the current feature set.

Keep each safeguard on a separate line: age assurance, feature access, input controls, output filtering, character availability and reporting are not interchangeable. None establishes that conversation data is private, and none should be treated as a guarantee that unsuitable material cannot appear.

Set a disclosure limit that survives imperfect controls

After completing the worksheet, decide what remains off-limits even when every available setting is enabled. Avoid information that could harm you or someone else if it were retained, reviewed, inferred, exposed or associated with the account.

• Passwords, authentication codes, recovery answers and payment credentials
• Full names combined with addresses, phone numbers, schools or workplaces
• Government identifiers and medical, legal or financial documents
• Another person’s private messages, images or confidential information
• Live location, travel plans or routines that reveal when someone is alone

For a sensitive subject, replace identifying facts with broader descriptions. A hypothetical user discussing a workplace conflict could write “a manager at a medium-sized company” instead of naming the employer, colleagues and office. Redaction cannot remove every privacy risk, but it reduces the directly identifying information entrusted to the service.

Proceed only when the documented controls, unresolved questions and remaining disclosure fit your risk tolerance. Recheck the worksheet after a policy notice, a settings change or a move into a different region or age category.

Also read:

• Character.AI Chats Feed Model Development—Share Less Than You Think
• Character.AI Chats Feel Private—Its Policy Still Treats Them as Platform Data
• Lumo 2.0 Adds Images and Memory While Keeping Chats Out of Training