Candy AI Review: Characters, Prompt Privacy, Tokens, Billing and Deletion

Candy AI is a fictional AI-companion service, not a dating site and not a directory of human members. It combines character selection and creation with text chat and generated images, voice notes, and other media. That range is the reason to consider it, but it also means the real buying decision is larger than whether the first conversation feels engaging. Prompt handling, media allowances, tokens, automatic renewal, cancellation, and deletion all affect long-term value.

Affiliate disclosure. AdultDatingRank may earn a commission if you use the first-party /go/candy-ai/ links or in-content affiliate shortcodes on this page. That commercial relationship does not change the legal-hub record. We do not invent prices, scores, or bank-statement labels. Candy is 18+ (or the legal age where you are located).

Kai Rowan’s verdict is cautiously positive for adults who deliberately want synthetic entertainment and will keep personal information out of prompts. It is a poor choice for anyone seeking a verified human relationship, confidential therapy, professional advice, or a system that can understand and consent like a person. A polished reply remains generated output.

Visit Candy AI — pricing & terms

Evidence scope and editorial method

Sources checked 4 September 2026. This is a desk review of Candy’s public legal hub, Terms of Service (date of revision 17 August 2026), Privacy Notice (date of revision 30 July 2026), Community Guidelines (date of revision 6 March 2026), and the legal-information index. We did not create a paid account or complete a purchase. Account-only controls and campaign-specific checkout details remain items the reader must confirm on the live purchase screen.

No reproducible public price is quoted here. Candy’s Terms of Service say paid subscriptions may be billed monthly, quarterly or annually, and that a monthly token allotment is disclosed at checkout (the TOS wording is generally 100 tokens per month). Record the amount, currency, tax, term, renewal, and token allowance presented to you rather than transferring a headline from another region or campaign.

Our review methodology weights product identity, prompt privacy, billing clarity, exit controls, and safe media handling ahead of promotional claims. Compare the separate Candy AI versus GirlfriendGPT guide if character discovery and role-play variety matter more than a guided companion flow.

What Candy AI offers

Candy.ai is operated by EverAI Limited, Malta company C107181, 56 Central Business Centre, Triq Is-Soll, Santa Venera SVR 1833, Malta. The public product and legal material describe an online chat application that generates fictional AI characters a user can message, plus other media such as AI-generated images, videos, and voice notes. Character creation and customization can give the conversation a consistent theme. None of those features establishes a human identity behind the character. The companion is fictional and AI-generated even when its language appears attentive or remembers details.

The Terms of Service (17 August 2026) are explicit that the service is entertainment only, that exchanges with AI characters are entirely fictional, and that generated output can be inaccurate, incomplete, or not reflective of human judgment. The service is not intended as emotional or other forms of support. Judge the product against a defined use case: instruction-following, coherent tone, continuity, and the ability to change a boundary. Do not score it by emotional intensity alone.

Community Guidelines (6 March 2026) and the Underage Policy restrict the service to users who are at least 18 or the legal age where they are located. Candy also states that users may not access the service from China or Singapore. The AI companion category keeps these products separate from adult dating reviews because the authenticity question is different. A dating service should explain human-profile checks. An AI service should clearly disclose synthetic characters, moderate prohibited requests, and give the account holder meaningful control over prompts and outputs.

Conversation memory and personal boundaries

Memory can improve continuity, but it can also encourage over-sharing. Use invented details when exploring a character. A real name, home address, employer, health record, financial information, password, identity document, or private information about another person is unnecessary for evaluating a fictional chat. The TOS itself tells users not to disclose sensitive personal information such as financial details, addresses, contact information, or passwords.

A practical evaluation uses harmless facts. Introduce a preferred fictional setting, a made-up name, and a clear topic boundary. Later, change one preference and see whether the system follows the update. Check whether you can edit, regenerate, delete, restart, or report a response. These are product-control signals, not evidence that the companion knows or cares about the user.

Set a time boundary as well as a data boundary. Continuous availability can turn casual entertainment into an automatic routine. Decide in advance how long a session should last. Stop if use disrupts sleep, relationships, work, or spending control.

Prompt privacy, generated media, and what the legal hub actually says

The Privacy Notice (30 July 2026) treats prompts, requests, and corresponding outputs as content data. It describes processing for service delivery, moderation, quality assurance, research, and model or technology improvement, including aggregation, de-identification, or anonymization. It also says some de-identified interactions may receive human review. Third-party LLM providers and/or hosters may receive the content of messages exchanged with the chatbot. That is a strong reason to treat intimate chat as service data rather than as a private diary.

Homepage marketing that claims end-to-end encryption or that data is never shared or sold should be read as marketing, not as a documented legal-hub guarantee. The legal hub does not document end-to-end encryption of chats, and the Privacy Notice does document sharing with LLM providers. Check candy.ai/legal-information, not the homepage FAQ, before treating a privacy slogan as a fact.

Payment processors named in the Privacy Notice are Emerchantpay, TrustPay, Volt, Coingate, and UpGate. This review does not invent a bank-statement descriptor. Record the merchant name on your own receipt.

Generated images, voice, and video create additional risk. Do not upload or describe a real person for intimate generation without permission. Avoid identifiable faces, documents, addresses, or private photographs. The TOS says conversations are generally treated as private, while moderation controls may flag, review, and act on content, including account termination and reporting to authorities. Users will not be refunded for moderation removals or for feature deletions.

Deletion has layers. Removing a visible image or chat may not equal closing the account or erasing all personal data. The Privacy Notice describes account data generally retained up to three years after last activity, or until a deletion request, with a possible shorter operational window; financial and transactional data are retained for ten years. A Data Protection Officer is listed at [email protected]; support is [email protected]; a UK representative is named as Shoosmiths Privacy Services Limited. Save any privacy request and response outside the service.

Candy’s legal hub also links an 18 U.S.C. 2257 exemption document. This review notes that the hub exists. It does not invent operator facts from that linked document.

Subscriptions, tokens, and recurring billing

Section 9 of the Terms of Service (17 August 2026) describes the paid path in wording a reader should verify live. As a free user, you may generally send a maximum of 5 messages to AI Characters. Further use requires a subscription, which may be billed monthly, quarterly or annually, or another form of paid-for access. As part of a paid subscription, the TOS describes unlimited access to messaging plus a monthly amount of tokens disclosed at checkout (generally 100 per month), used for extended features such as image generation or voice notes. Tokens cannot be transferred, traded, or exchanged for money.

Subscriptions renew automatically for the duration and price disclosed at checkout. Price increases are to be communicated in advance of autorenewal at the increased price. Because campaign, currency, tax, and token allowances can vary, record the purchase screen in front of you. Do not assume unused tokens roll forward.

Cancellation, refunds, account closure, and data erasure

Turning off renewal stops a future subscription charge under the current contract; it does not necessarily close the login. Remaining tokens expire when the paid period ends after cancel. A chargeback is treated as immediate cancellation of premium access. Complete each action that matches your goal.

1. Open My Profile / Settings and go to Settings / Unsubscribe, which is the cancellation path named in the TOS.
2. Save the confirmation, effective date, merchant, and remaining access period. Access continues until the end of the current billing period.
3. Check token balance. Remaining tokens expire and become unusable when that paid period ends; they are not transferable and are not cash.
4. Remove optional chats, characters, and generated media where controls permit.
5. Close the account through the stated route if you no longer need it. Account closure is not the same as a privacy erasure request.
6. Submit a separate privacy request ([email protected] or [email protected]) if you want access, correction, or erasure, and keep the correspondence outside the service.

Refund language is jurisdiction-specific. Outside the EU/UK clauses, the TOS describes a 24-hour window after payment; a refund is denied if more than 20 tokens have been used on a subscription, and card purchases are the refundable channel named there. EU and UK consumers have a documented 14-day withdrawal on the initial purchase only, with a prorated deduction for consumed service; that withdrawal does not apply to renewal periods. Contact [email protected] to exercise it. There is no refund for moderation actions or for feature deletions. Uninstalling an app, logging out, or deleting one character is not proof that billing stopped.

Who Candy AI is for — and who should skip it

It may suit an adult who wants a guided fictional companion, values several media formats, accepts that prompts and outputs can be used for moderation and model development, and is willing to record the live checkout (plan, tokens, renewal) before paying.

It is not for anyone seeking a verified human partner, confidential therapy, professional advice, a private log the operator cannot review, a guaranteed image or voice result, or a frozen universal price. It is also a poor fit if you need access from China or Singapore, or if you are under 18 / under the legal age where you are located.

Pros: clearly positioned AI companions; text and generated-media options; documented Settings / Unsubscribe cancellation; detailed public descriptions of content processing, processors, and retention.

Cons: prompts and outputs can support moderation, human review, and AI training; third-party LLM providers may receive message content; homepage encryption slogans are not the legal-hub record; tokens expire on cancel; campaign terms require live confirmation.

Final recommendation

Use Candy AI only as controlled digital entertainment. Confirm that the character and media features match a specific purpose, keep real-world secrets out of the conversation, map every paid gate on the checkout you are shown, and prove the Settings / Unsubscribe exit before committing. The best evidence of value is not how flattering a generated response feels, but whether the service remains understandable, affordable, and easy to leave.

Compare character tools, credits and data controls in our Candy AI versus Joi AI comparison.

For a side-by-side view of character tools, media limits, billing and data controls, read our Candy AI versus Kupid AI comparison.

Compare character-led chat with broader generation controls in our Candy AI versus Xotic AI comparison.

Review the live Candy AI offer before deciding

Primary sources (retrieved 4 September 2026): Terms of Service (rev. 17 August 2026); Privacy Notice (rev. 30 July 2026); Community Guidelines (rev. 6 March 2026); legal information hub.

Pros and Cons