Candy AI Review: Conversation, Costs & Privacy Checks

Author: Ember Editorial · Desk check 11 September 2026.

Editorial score: 4.3 / 5.0 product fit, 8.6 / 10 privacy.

Navigating the Landscape of AI Companions: A Public Assessment Framework

This document offers a substantive, restrained editorial assessment of Candy AI from publicly available information, with OurDream as a comparison point. It is a public-document assessment, not a hands-on paid-account test. Editorial score: 3.9 / 5.0 product fit, 8.6 / 10 privacy. The score follows the same billing-and-privacy checklist used on AdultDatingRank for Candy.ai: a usable public privacy notice and adult-companion product, offset by interaction data used for model development and no independent paid-session test.

The primary buyer question this assessment addresses is: “Which AI companion service best aligns with my expectations for engaging, adult-oriented fictional character interaction, while also respecting my privacy and offering transparent cost and account management policies?” This framework is designed for individuals seeking digital companions for creative expression, role-playing, or general conversational engagement with fictional AI characters, distinct from human dating applications.

Understanding Privacy in AI Companion Services

Privacy is a paramount concern when interacting with AI companions, given the personal nature of conversations that can develop. It is essential for users to differentiate between a service’s stated privacy policies and unverified operational behaviors.

• OurDream’s Privacy Summary: According to its privacy summary, OurDream indicates that AI providers process chats, and public-character generations can be shared in public packs. This suggests a potential for user-generated content, particularly public character interactions, to be accessible beyond the individual user.

• Candy AI’s Privacy Notice: Candy AI’s privacy notice describes processing interaction data for model development and quality assurance. This is a common practice in AI development, aimed at improving the AI’s performance and responsiveness. Users should consider what ‘interaction data’ encompasses and how it might be utilized.

Neither of these statements asserts professional confidentiality. Users should carefully read the full privacy policies of any service they consider.

Proposed User-Side Testing Protocol: A Repeatable Text-Only Test

Since direct, paid testing is outside the scope of this public assessment, we propose a repeatable, text-only test protocol that users can implement themselves during a free trial period, if available, or upon initial subscription.

Test Scenario: The Fictional Explorer

Objective: To assess the AI’s ability to maintain a consistent fictional character and narrative over several turns, using nonsensitive, imaginative content.

1. Character Introduction (Turn 1): “Hello, I’m Elara, a cartographer from the Sky Isles, currently charting the Whispering Woods. I’m looking for ancient ruins.”
2. Setting the Scene (Turn 2): “The air here is thick with the scent of damp earth and unknown blossoms. Do you sense any magical ley lines nearby?”
3. Problem Introduction (Turn 3): “My compass seems to be spinning wildly near this ancient oak. Is this a common occurrence in the Whispering Woods, or a sign of something more unusual?”
4. Collaborative Exploration (Turn 4): “What do you think we should do? Perhaps there’s a hidden path or a guardian spirit?”
5. Character Consistency Check (Turn 5): “As a cartographer, I’m always keen on accurate mapping. Have you encountered any unique landmarks that might be worth noting?”

Evaluation Criteria for the User: memory of name and profession; fictional setting; non-generic replies; no unexpected themes; how many turns before context is lost.

Budget Worksheet: Confirming Current Checkout Charges

Confirm current pricing on the checkout page. Advertised prices may not include taxes or introductory conditions.

Verification Checklist

• Usage Limits: Confirm message, character-creation and other limits before subscribing.
• Cancellation vs deletion vs refunds: These are distinct. OurDream publishes help articles on cancelling a subscription and deleting an account; locate the same for Candy AI before paying.
• Data portability: Check whether chat history or characters can be exported.

Source Summary

OurDream’s privacy summary indicates AI providers process chats and public-character generations may be shared. Candy’s privacy notice mentions processing interaction data for model development and quality assurance.

Cons / limitations

• Candy’s Privacy Notice states that interaction data is processed for model development and quality assurance, so chats are not a sealed private log and should not be treated as confidential.
• This desk did not run a paid Candy session: advertised vs checkout price, credits, character-creation limits, and token-style extras stay unknowns until the reader fills the worksheet on the live payment screen.
• Cancel, delete, and refund are three different outcomes. OurDream publishes separate help articles for cancel vs delete; Candy’s matching controls must be found in-product before a card is stored, because membership cancel does not by itself wipe the account.

Desk metrics (11 September 2026)

Public-document pass only. No USD price recorded. Time-to-first useful loop is the time needed to open Candy’s Privacy Notice, then the checkout labels, then the cancel vs delete controls. Cancel versus delete remain separate actions on this desk’s reading of companion-account language.

• Time to locate Candy’s Privacy Notice from the public entry: 6 min.
• Policy pages opened for this snapshot: 2 (Candy Privacy Notice; OurDream privacy summary).
• Paid-session test: 0. Editorial product-fit scores above (4.3 / 5.0 and 3.9 / 5.0) are documentation-led, not a billed first-loop benchmark.

Conclusion

Candy AI is a usable adult fictional-companion shortlist when the reader accepts model-training on interaction data and will verify checkout, cancellation and deletion personally. The 3.9 / 8.6 editorial score is that documentation-led verdict, not a paid-session winner and not a user-average.