Field notes / 01
Who Can Read AI Chats? Map Five Access Routes
Separate device access, shared links, character creators, staff and model providers. Use an access worksheet before sharing a sensitive detail.
Read guide + get worksheet →The AI companion field guideUnited States · Adult readers 18+
AI CompanionPrivacy:Understand, Compare and Control Your Data
Know who can read your chats, how long your data stays, and what deletion actually removes.
Six practical guides and private worksheets. Policy excerpts, not a security ranking. No account needed.
Original AI-generated characters · Illustrations only
Choose your next step
A practical answer.
A record you can keep.
Six focused guides, each with a downloadable worksheet. Start with the question that brought you here.
Field notes / 02
Chat History, Memory and Training: Three Different Controls
Use a six-category data ledger to distinguish personalization, logs, archives and media. Learn what a visible deletion can—and cannot—demonstrate.
Read guide + get worksheet →
Field notes / 03
Delete AI Chat Data: Keep, Remove, Confirm
An exit plan covering export checks, separate billing, retention triggers and incomplete confirmations, with a request log you can download.
Read guide + get worksheet →
Field notes / 04
Anonymous AI Chat: Separate Identity From Visibility
Compare what a nickname, private browsing and no-sign-up access actually address. Review device traces, payment links and account recovery trade-offs.
Read guide + get worksheet →
Field notes / 05
Local AI Privacy: Trace Text, Voice and Image Routes
Map the endpoint and each optional component before calling a setup local. Includes a configuration worksheet and a limited offline check.
Read guide + get worksheet →
Field notes / 06
AI Chat Exposure: Choose the Right Recovery Route
Separate oversharing, account compromise, public links and provider incidents. Keep a minimal incident log and use the matching official support route.
Read guide + get worksheet →01 / Choose with context
AI Companion
Privacy Comparison
Different questions. Different answers. Start with four policy findings—not an overall verdict on any app.
| Service | What the policy describes | Important limit | Evidence |
|---|---|---|---|
| Replika | Third-party models can process chats to generate replies. Their own training is contractually prohibited. | Internal improvement is separate. The up-to-60-day period for messages starts at contract termination; other categories and exceptions differ. | DocumentedPolicy §§ 2, 3, 6 ↗ |
| Nomi | Account deletion is described as taking about 28 days after confirmation. | Training and communications archives, plus legal exceptions, are addressed separately. | DocumentedPolicy §§ 4, 6 ↗ |
| Candy.ai | Training-data preparation may include human review of de-identified or anonymized interactions. | The policy’s 30-day debugging-log period is not a general chat-deletion promise. | DocumentedPolicy § 2 ↗ |
| CHAI | Processing may continue during the account’s lifetime and for up to five years after deletion. | Longer periods may apply for specified legal reasons. Account deletion is subject to retention obligations. | DocumentedRetention & deletion policy ↗ |
Swipe the table horizontally to see every column. “Documented” means stated in a source—not independently confirmed in the underlying systems.
Compare the same question across services. These four excerpts concern different policy fields and cannot support a privacy ranking. Before choosing, complete the access map and data ledger for each candidate. Price alone does not establish privacy.
02 / Check your companion
AI Companion
Privacy Profiles
A passport.
Not a score.
- A specific answer to a specific question.
- The source behind the answer.
- The exceptions you should not miss.
These excerpts are not complete privacy reviews. Other fields still need a scoped assessment.
Kindroid and Character.AI: further research is planned in the supplied report. No findings are asserted here.
PM / POLICY EXCERPTDocumented
Nomi
Account deletion · Policy §§ 4 and 6
- Request method
- Account settings.
- Published timeframe
- Approximately 28 days after deletion confirmation.
- Exceptions
- Training and communications archives; legal requirements. The policy says retained training-archive information is no longer attributable to you after deletion.
- In-app test
- Not performed
A policy statement is not proof that every server-side copy has been erased.
Read the official source PM / POLICY EXCERPTDocumented
Replika
AI processing · Policy §§ 2 and 3
- Reply generation
- Conversation data can reach third-party model providers to generate replies.
- Third-party training
- Providers are contractually prohibited from using the data to train their own models.
- Internal improvement
- The policy separately describes anonymized fragments for proprietary safety and chatbot improvement.
- In-app test
- Not performed
“Used to generate a reply” and “used for model training” are different fields.
Read the official source PM / POLICY EXCERPTDocumented
Candy.ai
Training and human review · Policy § 2
- Stated purpose
- Training and development of AI models and moderation technology.
- Human review
- Dataset preparation may involve review of de-identified or anonymized interactions.
- Scope limit
- This excerpt does not assess every reason for staff access or every retention category.
- In-app test
- Not performed
Read the exact processing purpose—not just a generic “private chat” label.
Read the official source PM / POLICY EXCERPTDocumented
CHAI
Retention and account deletion
- Published period
- Account lifetime and up to five years following deletion.
- Exceptions
- Longer periods may be needed for specified legal claims, disputes or obligations.
- Deletion limit
- The account-deletion provision refers back to those retention obligations.
- In-app test
- Not performed
An available deletion control does not establish immediate removal of every data category.
Read the official source Different faces. The same questions.
Eight original fictional adult characters illustrate the guide. These are not app listings, live users or brand-specific avatars.
FICTIONAL AIFICTIONAL AI
FICTIONAL AI
FICTIONAL AI
FICTIONAL AI
FICTIONAL AI
FICTIONAL AI
FICTIONAL AI03 / Understand access
Are AI Girlfriend
Chats Private?
Start with a more useful question: private from whom? Check each audience separately in the app's documentation.
Other people
Can someone see a shared link, notification preview, public post or conversation on your device?
Check visibility and sharing.
Character creators
Does the provider explain what a character’s creator can access? Keep this separate from employee access.
Check the creator’s actual role.
The operator
What do the rules say about support, moderation, contractors and other human access to content?
Check the conditions of access.
What should an “encrypted” claim explain?
Ask what is protected, against whom, and at which stage. Do not replace a specific explanation of chat access with a single encryption badge.
Prepare questions for the providerWhat should I check before reporting a conversation?
Look for an explanation of which messages are included, who receives the report and how long that material is retained. A brand-specific answer needs a source for that workflow.
04 / Follow the data
How AI Companions
Store and Use Your Data
Keep these three questions on separate lines in your privacy passport.
Chat history
What conversations are stored? What happens to that category when you delete a visible chat?
Saved memory
What information is kept for personalization? Is there a separate control for reviewing or removing it?
Model training
Which uses are described for internal and third-party models? What does any opt-out actually cover?
A conversation, a remembered detail, a training example: ask about each one.
Anika · AI-generated illustration, not a service-specific avatarA useful retention answer needs
a data category + a starting event.
05 / Leave with a plan
How to Delete AI Companion
Chats and Personal Data
Start with what you want to keep. Then define what you want removed. This is a planning checklist—not a tested instruction for every app.
Check your export
Identify what you need to save. Check whether the export includes chats, media and memory.
Define the scope
Separate chat history, saved memory and account data. Check subscription cancellation separately.
Use the official route
Find the documented deletion control or verified request channel. Keep the request reference.
Read the confirmation
Record what was confirmed, which exceptions remain and when the stated period ends.
Keep what matters. Be specific about what you want removed.
Ren · AI-generated illustrationFictional AI characterNo sign-up.
≠
No logs.
Amara · AI-generated illustration
06 / Everyday privacy
Anonymous AI Companions:
What Privacy Claims Really Mean
- Which account or device details can be linked to a conversation?
- What appears on a shared screen, lock screen or receipt?
- Which companies receive registration, payment or age-verification data?
“No sign-up” and “no logs” are different claims to investigate.
Use the review checklist07 / Look beyond the interface
Local AI Companion Privacy:
Keeping Conversations Under Your Control
A local interface does not by itself establish local processing. SillyTavern, for example, supports connections to local and cloud models.
For your setup, check the selected model endpoint and each enabled voice or image component. The diagram is a list of checkpoints, not a tested app configuration.
Read SillyTavern's API documentationMap your configuration
Your interfaceWhere do you type?
The selected modelWhere is the request processed?
Optional componentsWhere do voice and image requests go?
08 / Know the scope of your rights
AI Companion Data Rights
in the United States
Your location and the business involved matter. A California-specific right is not automatically a right for every US user.
California
The CCPA provides specified rights for California residents in relation to covered businesses, with exceptions. Start with the official explanation.
California Attorney General: CCPA ↗Check your applicable rules
This template does not determine legal eligibility or a statutory deadline. Ask the provider which process applies to your request.
Prepare a general information request09 / When something has already happened
AI Companion Data Exposure:
What to Do Next
Start with the situation you need to understand. You do not need to paste personal conversations into this site.
I shared personal information
Clarify which data categories are involved and what the provider says about removal.
Draft scope questions →Someone accessed my account
Find the app's official account-recovery and support process. A data request is not an account-recovery tool.
Find the provider's source →A conversation was shared
Check the product's shared-link controls and who may have received the conversation.
Review visibility questions →10 / Evidence, not a confidence trick
AI Companion
Privacy Research
An answer and the strength of its evidence are different things. These labels describe evidence—not safety scores.
Documented
Stated in an official source. Not independently verified in the service’s infrastructure.
Observed in test
Observed in a defined test. The platform, date and limits still matter. No such tests are claimed here.
Not disclosed
No clear answer found in the sources checked. This does not mean “definitely unsafe.”
Conflicting statements
The checked sources disagree. Show the disagreement rather than silently choosing an answer.
This edition: four documentary excerpts checked against official policies on . No in-app deletion tests, infrastructure audits, verified prices or overall privacy ratings. “Not assessed” means a field has not been reviewed; it is not the same as “Not disclosed.”
What a complete evidence record should contain
Record the claim, data category, exceptions, source URL, policy date, check date, platform and evidence status. When something changes, explain the change and its limits before updating the comparison.
source_url · policy_date · checked_at · platform · data_category · claim · exceptions · evidence_statusHow this guide handles evidence
This guide does not assign privacy scores or claim infrastructure tests. Policy findings are linked to official sources; practical scenarios are illustrative. Commercial links are separate from those findings. A changed policy sentence does not prove a technical change in the service.
All character portraits in this edition are AI-generated illustrations. Names and ages describe fictional adult characters; they do not represent live users, service endorsements or official app avatars.
11 / Turn a question into a next step
AI Companion
Data Request Builder
Prepare a plain-language draft. Review it, add account details through a verified channel, and send it yourself.
Your request, in your handsRuns in this page
Missing a clear answer? Ask a clear question.
Clara · AI-generated illustration · The builder prepares text, not an automatic request.