Character AI Private Bot vs Public: Separate Visibility From Data Access

Private and public answer only the visibility question

Character.AI’s help article establishes a public-versus-self-only choice for Characters and posts. It does not describe the setting as an encryption, retention, deletion or conversation-access control. That narrow scope is the useful finding: the label supports a visibility conclusion about the created item, while the rest of the data lifecycle needs separate evidence. Do not expand “visible only to you” into claims about search indexes, direct links, conversation dashboards, model processing or backups unless a current first-party source addresses that field. A careful user can still act on the documented part—choose the intended visibility—while treating every unaddressed privacy field as an open question.

Separate character discovery from conversation access

Discovery of a Character and access to a conversation are different questions, but the short FAQ supplies evidence only for the first at a high level. It says a created Character or post can be public or just visible to its creator. It does not explain what another user, the Character creator, Character.AI or a processor can access inside a conversation. It also does not explain direct-link behaviour, indexing, message retention or deletion. The correct comparison therefore has two columns: “documented visibility choice” and “conversation-data questions requiring another source.” Leaving the second column unknown is more accurate than filling it with either privacy assurances or alarming assumptions.

Build a Four-Layer Exposure Map

The only reliable way to evaluate Character AI private bot vs public risk is to construct an explicit four-layer exposure map. The map forces concrete inputs (visibility setting, subscription tier, sharing behavior) into observable evidence, supported interpretations, unsupported inferences, and next actions with clear stop conditions.

Four-Layer Exposure Map for a hypothetical private Character

Layer 1: Community visibility

Current evidence: Creators can choose public or self-only visibility for Characters and posts.

What remains unknown: Search, recommendation, indexing and direct-link behaviour are not described in the cited article.

Safe assumption: Rely only on the displayed visibility choice, not on an undocumented distribution guarantee.

Next action: Confirm the live setting on the exact Character and do not distribute its URL.

Stop condition: Stop if the interface does not show the intended self-only state.

Layer 2: Character creator

Current evidence: The cited article does not answer whether a Character creator can access a user’s conversation.

What remains unknown: Dashboards, logs, reports and every other creator-to-conversation route.

Safe assumption: Creator access is unresolved; neither “creators can read it” nor “creators can never read it” is supported here.

Next action: Check current first-party privacy documentation and omit identifying details meanwhile.

Stop condition: Stop before entering sensitive content if the required access boundary remains undocumented.

Layer 3: Service operator/processors

Current evidence: The FAQ points readers to the privacy policy; it says only that moderators might review community-visible material.

What remains unknown from this source: Operator and processor access, encryption, retention, deletion and private-conversation review.

Safe assumption: No conclusion about those fields is supported by the visibility FAQ alone.

Next action: Read the current privacy policy for each required field and minimize submitted data.

Stop condition: Do not use the service for the sensitive use case if the necessary field is absent or unacceptable.

Layer 4: Deliberate user sharing

Current evidence: The cited article does not document conversation-sharing controls.

What remains unknown: Export, screenshot, forwarding and shared-link behaviour.

Safe assumption: Content copied outside the service is no longer protected by an in-product visibility choice.

Next action: Do not deliberately repost identifying or intimate material.

Stop condition: Stop sharing if the destination has a broader audience or different retention policy.

This completed decision tool converts an ambiguous “private” label into four evidence checks. It deliberately records unknowns instead of treating the absence of a statement as proof that access exists or does not exist. The resulting decision can be conservative without inventing platform behaviour.

Treat moderation and operator processing as distinct

Moderation and operator processing are separate questions. The visibility FAQ supports one moderation statement: if a Character or post is made for the community, moderators might review it for potential negative impact. It does not say every self-only Character or conversation is reviewed, nor does it prove the opposite. The separate NSFW FAQ says pornographic content is against the terms and will not be supported, but that policy statement does not reveal a specific review workflow. Storage, model processing, training settings, subprocessors and deletion each need their own current evidence. Keeping those rows separate prevents a narrow community-review sentence from becoming an unsupported claim about a private chat.

Reduce what an intimate prompt can reveal

An effective risk-reduction strategy does not require guessing how the backend works. Replace names, addresses, employers and biographical details with fictional placeholders. Do not paste credentials, financial information, identity documents, private images or another person’s secrets. Keep the intended use inside Character.AI’s published content boundary; its current FAQ says pornographic content is prohibited. Calls also deserve a separate check: the official Calls & Voice FAQ says call conversations are stored as text chat. None of these precautions proves privacy. They simply reduce the consequences if a policy field is broader than expected and give the user a clear stop rule when documentation is insufficient.

Choose an adult service without calling it a vault

When Character.AI’s documented boundaries feel insufficient, users sometimes evaluate adult-oriented alternatives. AISoul, the publisher of this page, operates an adults-only browser companion limited to one active fictional companion that the user initiates. Registration requires only email/password or supported sign-in; no date-of-birth or identity-document field is visible. The free allowance resets daily by Beijing calendar day and supplies 50 messages plus 5 finite-gallery photos and 2 clear clips over the account lifetime. An active Pass removes published quantity caps on chat and eligible gallery-photo or short-clip retrieval while unlocking 18+ media. All media comes from a finite pre-generated gallery that supplies the closest eligible described match or fallback; the service is not unlimited unique files, prompt-to-video, live calls, or fresh generation. Chats are not public to other users, yet AISoul makes no end-to-end-encryption claim and external providers may process content. One-time, non-renewing Passes cost 7-Day $4.99, 30-Day $8.99, 90-Day $19.99, or Annual $49.99, with final price varying by region, tax, and bank approval. AISoul is not therapy, crisis care, dating, or a human relationship. For users seeking an alternative whose visibility and data boundaries are explicitly scoped to a single companion, see the dedicated private AI chat overview.

Additional context on broader platform differences appears in Character AI alternative and the longer examination is AI girlfriend chat private and safe?. Each resource separates documented capabilities from unverified assumptions.

Character AI privacy questions the setting cannot answer alone

The visibility toggle alone cannot resolve five common privacy questions. The answers below rely strictly on the verified first-party pages.

Is a private Character AI bot visible to other users?

The official article says a creator can set a Character or post to be public, meaning visible to anyone, or just visible to the creator. That supports a self-only visibility conclusion for the created item. The article does not describe search, recommendations, indexing or link sharing, so this answer should not add promises about those routes.

Can a Character creator read every private conversation?

The official short FAQ does not establish that creators can or cannot read every private conversation. It states only that visibility can be restricted to the creator and that community-visible material may be reviewed by moderators. No dashboard access, export rights, or real-time monitoring details are supplied. Users must therefore treat creator access as an open variable and minimize sensitive content accordingly.

Does private mode provide end-to-end encryption?

The cited privacy-setting FAQ does not establish end-to-end encryption. It discusses visibility of created Characters and posts, not encryption architecture. That silence is not proof that the service lacks every encryption control; it means private mode cannot be used as evidence of E2EE. Require a current security or privacy statement before relying on that property.

Can moderators review a private Character or conversation?

The FAQ states only that moderators might review a Character or post made for the community. It does not confirm or deny review of a self-only Character or a conversation. The NSFW FAQ establishes a content rule, not the trigger or reviewer for a particular chat. Treat private-conversation moderation scope as unresolved until a current source answers it.

What should I avoid putting in an intimate AI chat?

Avoid real names, addresses, phone numbers, financial details, or any information that could identify a living person. Do not describe illegal acts, non-consensual scenarios, or content that violates Character.AI’s published prohibition on pornographic material. Refrain from pasting external chat logs or embedding credentials. These restrictions reduce harm if any layer of the exposure map later proves more permeable than expected.

Two related questions remain open. The Lite FAQ documents more memory features and private Lorebooks in c.ai+, but it does not say those features change data retention. The sources reviewed here also do not establish a conversation-export route. In both cases, the correct answer is “not established by these sources,” followed by a check of the live account controls and current privacy policy—not an assumption that the feature exists or is absent.

Official privacy wording, unresolved access and publisher limits

Character.AI’s privacy-setting FAQ (checked 2026-09-07) states: creators can set Characters/posts public or visible only to themselves and that community-visible material may be reviewed by moderators. The NSFW FAQ confirms pornographic content is against terms and will not be supported. The c.ai Lite FAQ (updated 2026-09-04) and Character Voice and Calls FAQ supply subscription and feature boundaries but do not address encryption, full deletion guarantees, or creator access to every conversation route. These pages are the sole official sources used: https://support.character.ai/hc/en-us/articles/15063955587611-What-about-NSFW, https://support.character.ai/hc/en-us/articles/14992633549723-Are-Characters-and-conversations-private-, https://support.character.ai/hc/en-us/articles/55388952900379-C-ai-Lite-FAQ, https://support.character.ai/hc/en-us/articles/23957274129691-Character-Calls-Voice-FAQ.

AISoul publishes this research and operates the adults-only service described earlier. All AISoul facts were checked on the same date against https://www.aisoul.work/pricing.html, https://www.aisoul.work/about.html, https://www.aisoul.work/privacy.html and https://www.aisoul.work/terms.html. The comparison avoids any claim that AISoul media constitutes generation or unlimited unique files.

What was not tested: actual moderator dashboards, deletion latency, backup recovery, link-forwarding behavior, search-engine indexing of private links, real-time log access by creators, or any live chat on either platform. No prices were invented; all subscription figures match the official pricing page. No output quality, latency, memory performance, or moderation behavior beyond the supplied documentation was described. The Four-Layer Exposure Map remains a decision tool, not a benchmark or user test. Exact update date: 2026-09-07.