AI Companion Privacy Check: Chat, Image and Voice Data Questions
Companion chats can contain intimate preferences, fictional scenarios, uploaded prompts and generated media. The useful question is not whether a page says “private,” but what the policy says happens to each kind of data.
Map the data you plan to share
List what you may enter: email, payment details, chat text, prompts, uploaded images, generated images, voice clips and device information. Then search the privacy notice for each category. A broad sentence about improving the service is not the same as a clear explanation of retention, model improvement, service providers and deletion.
Look for training and improvement language
Check whether chats or uploads may be used to improve models or services, whether an opt-out exists and whether the rule differs between free and paid accounts. Do not assume a private conversation means the content is excluded from automated review, safety systems or product improvement. If the wording is unclear, avoid uploading identifiable material.
Understand generated media ownership
Generated images, audio and video may have separate rules from chat text. Check who may use the output, whether public galleries are opt-in or automatic, and whether deleted media can remain in backups. Never upload another person's private image without consent, and avoid material that could identify you if the retention answer is unresolved.
Check sharing and service providers
Payment processors, hosting providers, moderation services and customer-support tools may receive limited data. A policy should explain categories and purposes even if it does not name every vendor. Pay attention to vague permission to share data for undefined business purposes or with future partners without a clear boundary.
Choose a low-exposure first session
Use a separate email, avoid real names and locations, do not upload personal photos, and keep the first chat fictional until you understand the controls. Privacy is easier to protect before sensitive data exists. The safest recovery plan is not needing to recover exposed material later.
Build a content exposure ladder
Divide possible content into low, medium and high exposure. Fictional prompts with no personal details sit at the low end. A recurring nickname, location, relationship story or recognizable room can become identifying when combined. Personal photographs, voice samples and intimate real-world details belong at the high end. Match each level to what the privacy notice says about storage, service improvement, human review, sharing and deletion. Start your first session at the lowest level. Move upward only if the controls are clear and the benefit is worth it. This ladder prevents a vague privacy promise from becoming permission to share everything at once. If deletion and reuse rules are unknown, keep the content fictional.