Character.AI Chats Feel Private—Its Policy Still Treats Them as Platform Data

Character.AI chats are not private if “private” means a confidential record accessible only to you. Messages are content submitted to the platform, where they may be processed to operate, personalize, improve and protect the service.

That does not make every conversation public. Character.AI distinguishes ordinary submitted content from actions that make a Character or Voice public, but public visibility is only one boundary: company processing, service-provider access, model use, retention and legal disclosure are separate questions.

A claim-by-claim privacy matrix

The Character.AI privacy policy effective July 1, 2026 defines chat communications as submitted user content and establishes these boundaries:

• Collection: Character.AI collects chat communications, posted images or videos, biography descriptions and shared Characters. Voice recordings are collected when a user chooses certain voice features. Account, purchase, preference and support information may also be collected.
• Technical tracking: Character.AI and third-party providers collect information including IP addresses, device and browser details, access times, page views and service identifiers such as User ID, Chat ID and Session ID.
• Visibility to other users: actions such as creating a public Character or public Voice can make that content visible to others. Ordinary conversations are not classified as public by default, but neither are they defined as records accessible only to the account holder.
• Company and provider access: information may be disclosed to affiliates and to employees, consultants, service providers or vendors that need it for work such as storage, technical support, customer service, analytics, fraud prevention and content moderation. Authorization for these disclosures does not mean every listed recipient routinely reads every conversation.
• Model and service improvement: submitted information may be used to analyze, maintain and improve the service, train AI or machine-learning models, and develop features and algorithms. Regional controls can narrow some uses without converting a chat into user-only storage.
• Retention and deletion: retention depends on processing purposes, user choices and legal requirements. There is no universal deadline by which every copy of a deleted chat or account must disappear.
• Legal and corporate disclosure: information may be disclosed in response to warrants, subpoenas, court orders and other applicable processes, or during a merger, financing, asset sale or acquisition.

Why emotional privacy is different

An AI companion can feel patient, personal and nonjudgmental. That feeling describes the user’s relationship with the conversational character; it does not describe who operates the infrastructure, which processors handle the data or what legal rules govern the stored record.

A study first submitted on January 13, 2026, based on interviews with 15 users of companion platforms including Character.AI and Replika, found that emotional safety encouraged disclosure while participants remained aware of institutional risks; the AI-companion privacy study also identified uncertainty and feelings of powerlessness around platform-level control. Its small qualitative sample identifies a pattern rather than measuring how all users behave.

A sympathetic response from a Character therefore provides no evidence about technical access or retention. Discussing medical, legal or therapeutic subjects also does not create the professional confidentiality that may apply when communicating with a licensed clinician or lawyer.

What the model-training control changes

Character.AI’s model-training page states that certain submitted data is retained to train and improve AI models, that steps are taken to reduce personal information before data enters a generative-model training dataset, and that users in the European Economic Area and United Kingdom can disable “Improve the Model for Everyone” so new content is excluded from generative-model training while ratings, reports and data used for search, recommendations or safety classifiers may still be processed.

The setting is therefore narrower than a universal “do not use my data” switch. It does not promise immediate deletion of existing chats or prevent processing required to provide, secure or moderate the service.

What account deletion can and cannot do

Users can delete an account through the account profile page and can correct, update or delete certain profile information there. Depending on location and legal exceptions, a user may also request access, a copy, correction, deletion, restriction or an objection through the support form or privacy email. Identity verification may be required.

Account deletion is not an unconditional promise to erase every record immediately. Information may remain where needed for legal compliance, service security, dispute resolution or enforcement of agreements. A formal privacy request can identify the categories at issue—such as chats, uploaded media, voice data, profile fields and created Characters—but the available right and its exceptions depend on the user’s jurisdiction.

There is also a specific exception for shared creations. If a Character has been made available to other users, its characteristics may be preserved and the Character may remain active after its creator deletes their data and account. Creators should therefore keep personal information out of Characters they intend to share.

Children still need a disclosure boundary

Age restrictions can reduce exposure without making permitted activity confidential. From November 25, 2025, users under 18 could no longer engage in open-ended Character chats, although creation and media features remained available, according to the Australian eSafety Commissioner’s Character.AI safety guide.

UNICEF’s June 2026 business recommendations call on AI-chatbot companies to test risks to children’s privacy, safety, autonomy and mental well-being, apply stricter safeguards to higher-risk companion products, minimize age-assurance data and tightly limit retention of intimate conversational data. These are recommendations to companies, not a finding that any particular Character.AI chat is confidential.

Information that should stay out of a chat

Sensitive personal information is unnecessary for using the service. A practical boundary is to avoid submitting anything whose storage, review, disclosure or retention by an online platform would create harm:

• passwords, authentication codes, recovery answers or security keys;
• payment-card, bank-account, tax or government-identification numbers;
• exact home, school or workplace addresses and real-time location details;
• medical, therapy, legal, disciplinary or employment records;
• intimate media, voice recordings or identifying details about another person without permission;
• confidential information belonging to a school, employer or client;
• details that identify a child’s routines, contacts or pickup arrangements.

When context is necessary, replace real names and unique details with neutral placeholders and remove combinations of facts that could identify someone. A Character.AI conversation may feel personal while still remaining data entrusted to a platform under its published processing rules.

Also read:

• Character.AI Confirms Mass Deletion of Fandom Characters, Says They're Not Coming Back
• Is Microsoft Office 365 Security Policy Adequate to Protect Your Data?
• These Tools Are Best for Big Data Professionals