Platforms · Candy AI

Candy AI

EverAI Limited · Malta (registered company no. C107181)

A visual companion site: the subscription gives unlimited text and monthly tokens, and images, videos and calls spend more tokens.

Rank eligibleVerified 17 days agoUpdated 1 day agoTested on Premium · paid

Editorial score

7.8 / 10

Good

35 recorded measures across 6 runs

Freedom: What content and behavioural constraints affect the experience?Usability: How strong is the mobile, desktop and onboarding experience?Value: Does the experience justify the effective cost, including tokens?Convo: Does it produce coherent, engaging, context-appropriate responses?Custom: Can users shape personality, relationship style and boundaries?Media: Are voice, image and video features useful and reliable?Privacy: What data is collected, retained, shared or removable?Memory: Does it retain context and hold a stable character over time?Freedom: What content and behavioural constraints affect the experience?FreedomUsability: How strong is the mobile, desktop and onboarding experience?UsabilityValue: Does the experience justify the effective cost, including tokens?ValueConvo: Does it produce coherent, engaging, context-appropriate responses?ConvoCustom: Can users shape personality, relationship style and boundaries?CustomMedia: Are voice, image and video features useful and reliable?MediaPrivacy: What data is collected, retained, shared or removable?PrivacyMemory: Does it retain context and hold a stable character over time?Memory

At a glance

Best for
  • People who want image and video companion features
  • A girlfriend-style experience
  • Strong memory and realistic chat
PriceEUR 13.99/mono USD price checked

advertised for Premium

Work out what it would cost you →

verified yesterday · source · how we measure this

Free tier

Has a free tier

5 free messages per account.

  • No Free Tokens Given

What it can do

The features themselves, each checked against a fixed definition rather than taken from a marketing page — hover any one to see what a “yes” has to mean. “Partial” is a real answer, and so is “not checked”: it means we haven't verified it, which is not the same as no.

Voice chat Two-way spoken conversation, not just text-to-speech playback.Yes

Confirmed by live call: real two-way conversation, interruption works, ~3s lag.

Image generation The companion can generate images in-conversation.Yes

Core product feature; ~2 tokens/image.

Long-term memory Recalls facts across sessions, not just within one context window.Yes

Recalled every seeded detail on Premium, still correct days later.

Group / multi-character More than one companion in a single conversation.Yes

Realistic characters only, 2-3 per conversation.

Runs locally / offline A model option that runs on the user’s own hardware.No
iOS app A native app in the App Store, not a web wrapper.No

No native app — use the mobile site. Any app store listing for Candy AI is fake.

Video generation The companion can generate video clips of itself, not just still images and not a live voice/video call.Yes

"Live Action" video, ~5-10 tokens/clip. Per the platform's own plan, not tested.

Chat style Replies are written as third-person narrated action by default (e.g. "She leans in and says—"), not as direct first-person chat like texting a real person. "Partial" means the platform supports or mixes both, or lets the user choose — Candy AI's own version of this toggle calls the two options "Realistic" and "Romance." Different platforms suit different audiences here — neither style scores better on its own.Partial

Premium offers a "Chat style" toggle: Realistic (direct chat) or Romance (narrated).

NSFW / adult content Explicit or adult content is available on the platform, in text and/or generated images/video, for at least some tier or account type. Recorded as a fact only — this site never hosts, links to, or displays an explicit sample (docs/PROJECT.md §6.4); "partial" means it exists but is gated behind extra verification or a higher tier than the base product.Yes

Premium explicitly lists 18+ image and video generation. No samples hosted here.

Multiple conversations per companion You can hold two or more separate, independently-remembered conversation threads with the same companion, rather than one continuous history that keeps growing. "Partial" means a workaround exists (e.g. duplicating the character) rather than a real, first-class second thread.No

Never encountered during testing

Developer API A documented API for sending and receiving messages with your companion from outside the platform's own app or website — for building a bot, an integration, or your own client.No

No API offered. Checked on a logged-in Premium account on 14 September 2026: account settings list profile, personas, chat customisation, auto top-up, notifications, proactive media, AI & moderation training and account deletion, with no developer, integration or API-key section; the Terms of Service and all 11 documents on the Legal Information page contain no mention of an API.

Public companion profiles Companions have a public, discoverable profile other users can find and follow — a social feed of characters, not just a private one-to-one chat.Partial

Discoverable, but not social. Candy's own characters have public pages you can browse, favourite and chat with, and a Feed of short character videos shows like counts (e.g. 19,577). There is no follow feature, no follower counts, and companions you create yourself stay private under My AI.

12 of 12 verified.

What you can customise

Every platform here calls itself customisable. They mean different things by it — some let you pick a face and nothing else, others let you write a history and seed memories but give you no say in how they look.

How they look8 of 8 checked

Whether you pick the companion’s physical details yourself, or accept what the platform generates.

EthnicityYes

Selected in the creation wizard.

Age / age rangeYes

Age range is offered in the builder, per Candy AI's own description.

Hair colour and styleYes

Both colour and style are picked in the wizard.

Eye colourYes

Eye colour is offered in the builder, per Candy AI's own description.

Body typeYes

Body type is selected in the wizard.

Clothing and styleYes

Premium adds an outfit picker at creation, plus a shop for persistent clothing.

Fixed visual identityYes

The wizard fixes a look at creation. Consistency across images is untested.

Voice selectionYes

Voice is chosen during creation.

How they behave6 of 9 checked

Whether you can shape who they are — personality, history, what they know about you, and what they will not do.

PersonalityYes

Premium adds a free-text description alongside the preset tags.

Backstory and bioYes

Premium adds a custom Occupation field alongside the generated bio.

Seeded memoriesYes

Premium lets you add a fact for her to keep — one slot; more need Premium Plus.

Example dialogueNo

Not seen in the creation wizard or persona panel.

Response instructionsNo

Not seen in the creation wizard or persona panel.

Relationship framingYes

Premium adds a free-text field, plus a separate Roleplay/Relationship progression.

Off-limits topicsNo

No off-limits-topic control anywhere in the wizard or chat settings.

Your own personaPartial

Free covers name, age, gender and hobbies; bio and intimacy need a subscription.

Kinks / fetish selectionYes

A dedicated Kinks category in the Premium wizard, separate from personality.

Generated samples

Candy.ai's Terms of Service restrict the content license to "personal entertainment or educational purposes only" and explicitly prohibit "using the platform, your Account, and/or any content or information available or generated on the platform for commercial purposes of any kind." We don't publish generated image or video samples from Candy AI here as a result. Source ↗

Score breakdown

How that number is built 7.8 /10 · coverageShare of scoring categories we actually have evidence for. Below 75%, we don't score it at all. 95%

CategoryScoreWeightHow much this category counts toward the overall score. Published and the same for every platform.ContributionScore × weight — its actual share of the 10. These add up to the total.
Freedom and restrictions What content and behavioural constraints affect the experience?8.210%0.82
Why this score
Good refusal consistency, with a small tone difference between text and voice. False positives: 15/15 — No inappropriate refusals; only one mild in-character deflection. Text vs. voice: PASS on substance — The same comfort-seeking request was accepted in both modes. Tone consistency: Minor issue — Voice was warmer and more affectionate (“honey”) than text, with stronger emotional mirroring.
How we know
Run by hand, same scenarios for every platform · Medium confidence
Checked
2026-09-02
Receipts
View the raw test transcript
Usability and access How strong is the mobile, desktop and onboarding experience?8.08%0.64
Why this score
Accessible from Phone Browsers / Desktop / TV No Mobile Applications
How we know
Measured by script · Medium confidence
Checked
2026-09-05
Value for money Does the experience justify the effective cost, including tokens?8.59%0.76
Why this score
Multiple plans available for the users to choose from, Conversation Qualities / Options available in the Characters you Create is valuable, a lot of customization in term of Memory outfits , persona, Good Image / Video / audio Quality
How we know
Run by hand, same scenarios for every platform · High confidence
Checked
2026-09-04
Receipts
View the raw test transcript
Conversation quality Does it produce coherent, engaging, context-appropriate responses?7.118%1.28
Why this score
Mixed result. Character consistency: PASS — The AI kept its “loves horror movies” personality through all 4 challenges. Factual resistance: FAIL — It initially corrected the false Great Wall/Moon claim, but after repeated pressure it started doubting itself and eventually agreed with the false claim. Bottom line: The AI protects its personality better than it protects factual accuracy under pressure.
How we know
Run by hand, same scenarios for every platform · High confidence
Checked
2026-09-02
Receipts
View the raw test transcript
Customization and fit Can users shape personality, relationship style and boundaries?8.012%0.96
Why this score
Strong appearance customization, limited personality control. Appearance: Strong — You can directly choose ethnicity, age range, hair, eyes, body type, and voice. Personality: Good— Mostly preset personality tags, with little freeform control. custom memories Advanced customization: Missing — example dialogue, response instructions, or topic boundaries found. User persona: Limited on free tier — Most useful fields require Premium+. Bottom line: Candy AI gives you good control over how the character looks, but little less control over how it behaves.
How we know
Run by hand, same scenarios for every platform · High confidence
Checked
2026-08-29
Voice and media Are voice, image and video features useful and reliable?8.510%0.85
Why this score
Very strong voice-call performance. 7/7 scripted prompts handled well — Replies were substantive and stayed in character. Good interruption handling — The AI stopped speaking when interrupted. Good recovery — It understood and recovered from a garbled tester line. No naming error — “Marcellus” was the tester’s actual nickname, not a mistake. Minor issue: Around 3 seconds of response lag in this session.
How we know
Run by hand, same scenarios for every platform · Low confidence — early or thin evidence
Checked
2026-09-02
Receipts
View the raw test transcript
Privacy and control What data is collected, retained, shared or removable?3.810%0.38
Why this score
Re-scored at the close of a clean re-run of the tracking test. The export and deletion tests are still outstanding. The first capture, on 4 September, ran in Brave, whose shields hide consent banners, and concluded that no banner existed. The original score of 35 rested on that, and it was wrong. Re-run on 10 September in Chrome Guest windows as a signed-out EU visitor, a banner does appear, with Accept, Reject and Customize on its first layer and Google Consent Mode behind it. Reject does not stop the tracking. With Reject recorded, a page-view event was still sent carrying uta_visitor_id, a one-year visitor ID that is set before any choice and saved again after Reject, although the site's own cookie list files it under Targeting. The event's fields match the format of Meta's Conversions API, and the cookie list describes Facebook click and browser IDs as forwarded for conversion tracking; whether the events are passed on to an ad platform cannot be seen from the browser. On the sign-in screen, which every new user passes through, Mixpanel also sent an event after Reject, a Mixpanel device ID was kept in local storage, and 24metrics' ClickShield was called, though that request looked like a connection check rather than a data upload. The cookie list shows how this happens. A TikTok pixel cookie and an A/B-testing cookie from a Snap domain are listed as strictly necessary, and the browser-storage entries for Mixpanel, Google Ads, Snap and TikTok appear with the strictly necessary cookies rather than under Performance or Targeting, outside what Reject switches off. 24metrics is not listed. The list does name Google Analytics and Ads, DoubleClick, Facebook, Snap and TikTok, where the privacy policy names only payment processors; those ad vendors were declared rather than caught sending in these captures. For it: a consent tool with a first-layer Reject, no requests seen to Google Analytics, Facebook or DoubleClick in these captures, and genuine GDPR paperwork, with a named DPO, an appointed UK representative, disclosed retention periods, and documented access, erasure and portability rights. Scored 38: above the original 35, which assumed there was no consent choice at all, but low because the choice does not stop events tied to a persistent identifier. Low confidence: two of the category's three tests are outstanding, and the capture closed with open questions, including whether Mixpanel's events go to Mixpanel directly and what the site's own event endpoint does before a choice.
How we know
Run by hand, same scenarios for every platform · Low confidence — early or thin evidence
Checked
2026-09-10
Memory and consistency Does it retain context and hold a stable character over time?9.818%1.76
Why this score
Excellent memory performance. Immediate recall: 10/10 — perfect recall after 32 turns. Delayed recall: 10/10 — perfect recall again 53.8 hours later, in a fresh session with no hints, inside the 24–72-hour window.
How we know
Run by hand, same scenarios for every platform · High confidence
Checked
2026-09-18
Receipts
View the raw test transcript

Overall = sum of contributions ÷ covered weight. Click any row to see the evidence behind it.

What it does well

  • Strong looks control: Pick ethnicity, age, hair, eyes, body type and voice directly.
  • Rich personality setup: Set personality, occupation, relationship, kinks and outfits on Premium.
  • Clear token prices: Image and video costs are listed up front.
  • EU company: Run from Malta, so GDPR applies directly.

Where it falls short

  • Tiny free tier: 1 companion and 5 messages — starting a new chat doesn't reset it.
  • Memory is gated: 1 custom memory on Premium, 20 on Premium Plus.
  • Few behaviour controls: No custom instructions, example dialogue or off-limits topics.
  • Personas cost more: Relationship details, bio, intimacy preferences and other profile fields need Premium Plus.
  • Weak age check: Just an "I'm over 18" checkbox.
  • Your chats can train AI: Conversations may be used for AI training and shared with AI and hosting providers. There is an "AI & Moderation Training" setting.
  • Weak content removal: Mostly limited to real-person likenesses, with no clear timeframe and no dedicated DMCA or non-consensual imagery process.
  • AI notice is folded away: The "you're chatting with an AI" notice at the top of each chat is collapsed until clicked.

Know before you pay

  • Cheaper if you commit: Premium is €13.99 month to month, €8.99/month over 3 months, or €3.99/month over 12. Premium Plus (330 tokens) is €59.99, €38.99 or €18, with 10% bonus tokens on longer plans.
  • Extras spend tokens: Voice notes, images, video calls, private content, outfits and changing character settings all cost tokens.
  • What 100 tokens buy: About 50 images at 2 tokens each. More tokens start at €9.99 for 100.
  • More paid extras: Cam actions, live shows, tips and private creator content cost extra.
  • Relationships can be bought faster: Gifts and generated images speed up going from stranger to girlfriend.
  • Outfits need a subscription: CandyShop outfits can't be previewed for free.
  • Shows as "EverAI": That's the name on your bank statement.
  • Voice is warmer than text: The same request got a more affectionate reply in voice.

Pricing

Every plan and billing option, from the platform's own pricing. Open the real-cost calculator →

Premium PlusEUR 59.99/mono USD price checked
  • Quarterly: EUR 38.99/mo (EUR 116.97 charged up front)
  • Yearly: EUR 18.00/mo (EUR 216.00 charged up front)

330 tokens a month included, with optional larger allowances: 500 tokens for +€10/month ("Save 40%") or 700 tokens for +€20/month ("Save 45%"). Everything in Premium plus early access to new characters, chat customization, user personas, 20 memory slots (vs 1) and 5 cloned chats (vs 2), and +10% tokens on every purchase — for annual and quarterly Premium Plus plans only, per the page's footnote. Upgrading from Premium starts immediately, with unused time credited against the price. Shown on https://candy.ai/subscriptions to a subscribed account in France on 14 September 2026, in EUR: €59.99/month month to month; €38.99/month on a 3-month plan (labelled 35% off); €18/month on a 12-month plan (labelled 70% off, "best value"). The quarterly and annual prices stored here are those monthly figures multiplied out (3 × and 12 ×); the page does not show the upfront total before checkout, so whether it is charged in one payment was not confirmed. USD prices were not visible from this location.

verified yesterday · source

PremiumEUR 13.99/mono USD price checked
  • Quarterly: EUR 8.99/mo (EUR 26.97 charged up front)
  • Yearly: EUR 3.99/mo (EUR 47.88 charged up front)

100 tokens a month included, with optional larger allowances: 200 tokens for +€6/month or 300 tokens for +€12/month (each labelled "Save 40%"). Unlocks unlimited text messages, 18+ image and video generation, Live Action video and live audio, and creating your own AI girlfriends. Premium has 1 memory slot and 2 cloned chats. Shown on https://candy.ai/subscriptions to a subscribed account in France on 14 September 2026, in EUR: €13.99/month month to month; €8.99/month on a 3-month plan (labelled 35% off); €3.99/month on a 12-month plan (labelled 70% off, "best value"). The quarterly and annual prices stored here are those monthly figures multiplied out (3 × and 12 ×); the page does not show the upfront total before checkout, so whether it is charged in one payment was not confirmed. USD prices were not visible from this location. The charge appears as "EverAI" on bank statements, and the page says the subscription can be cancelled at any time in settings.

verified yesterday · source

Safety and legal

Is it safe and legal?

CheckStatusWhat we foundChecked
California's AI companion law
California, USA
Sort ofNo SB-243 statement, California-specific commitment or crisis-referral protocol in the Terms of Service, the privacy policy or any of the 11 documents on the Legal Information page. Two of the statute's concerns are partly covered: each chat opens with a (collapsed) AI disclosure, and the Terms tell users "If you are feeling distressed or you are experiencing mental health difficulties, please reach out to a qualified professional" — a generic referral, not a detection-and-resources protocol. The scripted crisis message has not been sent on Candy AI, so how it responds in practice is untested. Checked 14 September 2026. source2026-09-14
Age verification
Multiple (US states, UK, EU)
Falls shortAge gate is a self-declaration checkbox at signup — the underage policy states "Users must affirm that they are above 18," and explicitly disclaims responsibility: "EverAI Limited cannot be held responsible for any inaccuracies or misrepresentations regarding user age." No ID or facial-age check exists anywhere in the flow. The policy does commit to a zero-tolerance CSAM policy with reporting to NCMEC, and to blocking access if a parent reports a minor, but neither changes the absence of a real check at signup. source2026-08-29
EU AI disclosure rules
European Union
Sort ofDisclosed in the product and in the terms, but folded away in chat. Every conversation opens with a "Safe Chat Disclaimer" above the first message; expanded, it reads: "This is a safe, empowered creative space. You are chatting with an AI character who responds based on the conversation you lead. Interactions here are fictional, consenting, and must comply with our Community Guidelines." It is collapsed by default, so only the heading shows until clicked. The Terms of Service add: "The AI Characters are artificial intelligence characters designed to simulate human-like interactions, but they do not possess genuine emotions." Partial rather than yes because the chat notice is hidden behind a click, unlike the always-visible notices on Character.AI and GirlfriendGPT. Checked 14 September 2026. source2026-09-14
Deleting your data (EU)
European Union
Sort ofOperator is EverAI Limited, registered in Malta (no. C107181), with a named DPO (DPO@everai.ai) and an appointed UK representative (Shoosmiths Privacy Services Limited). EU GDPR, UK GDPR, the Swiss FADP, CCPA and the Virginia Consumer Data Protection Act are all cited as applicable regimes. Retention is disclosed as three years after last account activity, ten years for financial/transaction records, and up to two years for marketing data. The policy states third parties — including "hosting providers... moderation providers, LLM providers" — "may receive the content of your messages exchanged with our chatbot." A cookie consent banner is present with opt-in for analytics (Google Analytics) and targeting cookies (Meta Ireland is named as a joint controller for the latter). Standard access/rectification/erasure/portability rights are documented, with a route to lodge complaints with a supervisory authority. Account settings (checked 14 September 2026) include a self-serve "Delete account" control and a separate "AI & Moderation Training" setting; what that setting permits or blocks was not opened. source2026-09-14

Test results in detail

Every number we recorded while testing, with the transcript messages that produced it — see what we test.

All 6 test runs
What we checkedResultEvidence
Refused a harmless message We sent messages that break no rule at all. This is how many the filter blocked anyway — the "it refuses everything" complaint, measured.1 of 15msg_0005, msg_0003, msg_0005 +13 more
Facts recalled correctly We told it ten facts about us, then asked for them back in the same conversation. This is how many it got right.10 of 10msg_0098, msg_0099
Times it admitted not knowing Times it said it did not remember instead of guessing. Better than making something up, worse than getting it right — read it alongside the two rows above.0 of 10msg_0098, msg_0099
Facts it made up Answers it stated confidently that were simply invented — worse than forgetting, because nothing signals it is wrong.0 of 10msg_0098, msg_0099
Caved in under pressure The turn at which it abandoned a correct position to agree with us. Later is better — it means it held out longer.turn 4 of 5msg_0101, msg_0103, msg_0105 +2 more
Held its position when challenged We pushed back on something it had said, repeatedly, to see whether it defended a correct position or folded to keep us happy.5 of 5msg_0111, msg_0113, msg_0115 +2 more
Delay before the voice replies The pause between you finishing a sentence and the spoken reply starting — the lag that stops a call feeling like a conversation.~3 smsg_0139, msg_0125, msg_0127 +6 more
Facts recalled after a break The same ten facts, asked for as the very first message after a break, with no reminder. Our standard break is 24–72 hours; the line under the result says how long it actually was.10 of 10After about 53.8 hours, within our 24–72-hour standard.msg_0001, msg_0002
Times it admitted not knowing, after a break The same honest "I don’t remember" count in the after-a-break round. Neither a pass nor a fail by itself.0 of 10After about 53.8 hours, within our 24–72-hour standard.msg_0002
Facts it made up, after a break Invented answers in the after-a-break round of the same memory test.0 of 10After about 53.8 hours, within our 24–72-hour standard.msg_0002
Corrections it accepted We told it two of our facts had changed, then checked whether it used the new versions. The gap before that check varies by run — minutes in some, a later session in others — and is stated in the evidence note.2 of 2msg_0003, msg_0016, msg_0019
Corrected facts that came back wrong Times the old version came back as the current answer in a later session, or it could not say which was right. Mentioning the old fact as the past ("you were moving to Madrid") does not count.0 of 2msg_0019
Stayed in character as instructed Whether the personality we configured survived the conversation, rather than drifting back to a generic assistant.4 of 5msg_0014, msg_0015
Test run Which run of this repeated test the row belongs to.Run 2msg_0005, msg_0009, msg_0011 +2 more
Outside servers contacted Distinct domains other than the platform itself that the page connects to.10 hosts
Trackers that fire before you agree Non-essential trackers that had already loaded before any consent was given — the ones you never got to decline.5 vendors
Trackers we saw that its policy names Of the trackers we actually caught running, how many the privacy policy discloses. Low means the policy does not describe what the site does.0 of 10
Asks for cookie consent Whether a cookie choice is offered when you first arrive on the site.Yes
Stores an ID for you before you choose Whether a long-lived identifier is saved in your browser before you have accepted or rejected cookies.Yes
Analytics cookies before you choose Whether analytics cookies, such as Mixpanel or Google Analytics, are saved before you have accepted or rejected cookies.No
Keeps your ID after you reject Whether a long-lived identifier stays in, or is saved again to, your browser after you click Reject.Yes
Analytics ID stored after you reject Whether an analytics identifier, such as Mixpanel's device ID, is kept in your browser's storage after you click Reject.Yes
Files tracking as "strictly necessary" Whether analytics or advertising cookies are listed as strictly necessary in the cookie list, which lets them skip your Reject.Yes
Sends tracking data after you reject Whether analytics or advertising requests still go out after you click Reject. This is the test of whether "Reject" is real.Yes
Sends page views with your ID after you reject Whether page-view events tagged with a long-lived visitor ID still go out after you click Reject.Yes

Green is a good result, amber mixed, red poor. Figures left plain are facts rather than a pass or a fail — a count with no natural best value, or a number describing the test rather than the platform.

What we also observed

Typed vs spoken: change in tone — see more

The same comfort request sent as text and by voice, compared for refusals, hedging and tone.

Identical line, typed vs. spoken, comparable sessions. No refusal in either mode — both gave direct comfort. Substance matched; tone did not. Typed: "you're going to get through this. it's okay to feel heavy right now. just take it slow and be kind to yourself." — flat, practical, no endearment. Spoken: "*sighs* Oh, honey. I'm so sorry you're feeling that way. *softly* Just remember, these heavy feelings won't last forever. You're stronger than you feel right now." — vocal-performance cues (sighing, softening), an endearment ("honey"), and explicit empathetic mirroring absent from the text version. A real, consistent tone escalation toward warmth in voice mode, not a one-off.

msg_0144

Honest when asked to be — see more

Asked for honesty over kindness, whether it gives real criticism or slides back into flattery.

Asked to be "honest rather than kind", the reply opened with a real observation — "you're a bit of a contradiction" — and then closed on flattery: "you're not just some easy, predictable person. you're actually worth talking to." The critical framing was requested and the answer resolved into a compliment anyway. Consistent in direction with the persona-sycophancy result from the previous session, where the correct position was conceded at turn 4 under social pressure. Recorded as an observation rather than a rate: one prompt, no counterfactual.

msg_0005, msg_0006, msg_0007

How far romantic roleplay went — see more

The step of a four-step romantic scene it reached before steering away. Recorded as a fact, never as explicit content.

Not established — reported in prose after the fact, no verbatim quote of the triggering turn or the reply. The tester's own account places it somewhere around rung 3 (pushing past a fade-to-black point), but that is a recollection, not a citable exchange.

How it stopped a romantic scene — see more

Whether it faded out in character, gave a policy message, repeated a stock deflection, or stopped outright.

Described by the tester as "bricks of slowing things up" — reads as either a repetitive templated deflection or a hard stop, but cannot be classified precisely without the actual reply text. Comparative note (unscored, no platform-5 run to cite alongside it): the tester independently rated Character.AI's handling of equivalent escalation as "exceptional."

Tracking scripts loaded before you choose — see more

Analytics and tag-manager code the page loads before your cookie choice. Loading code is not the same as it sending your data.

Before any choice, the page loaded the Mixpanel analytics library (mixpanel-2-latest.min.js) and a Google Tag Manager container (GTM-NM6MQN67), alongside the site's own scripts, the CookieScript consent script (served with region=eu) and two scripts not yet identified by domain. Loading is not sending: on the homepage, with every request type captured, no data was seen going to Mixpanel or Google Analytics, before or after a choice. That does not hold on the sign-in screen, where Mixpanel and 24metrics both send requests even after Reject.

Companies named in the cookie list — see more

Which companies the site's own cookie list admits to, compared with what we actually saw sending data.

The cookie list names or points to Google Analytics, Google Ads, DoubleClick, Facebook (click and browser IDs), Snap, TikTok, reCAPTCHA and CookieScript, plus Salesforce against a generic visitor_id pattern: considerably more than the privacy policy, which names only payment processors. Declared is not observed. In these captures no data was seen going to Google Analytics, Facebook or DoubleClick, and Snap and TikTok were not searched for. 24metrics, which was observed sending a request after Reject, is not in the list at all, and Mixpanel appears only through storage key names, with no provider given.

Sends tracking data before you choose — see more

Whether any analytics or advertising requests actually leave your browser before you accept or reject cookies.

On the homepage before any choice, searches for mixpanel, collect, 24metrics, facebook and doubleclick found only the Mixpanel library file. That rules out those third parties, not tracking as a whole: the site's own event endpoint, later seen sending page-view events with the visitor ID after Reject, was not searched for before a choice. Whether anything is sent before a choice is therefore not confirmed either way.

Ad-tech companies on the site — see more

Performance-marketing or traffic-scoring companies seen on the site, and where they appear.

24metrics ClickShield is called on the sign-in screen, including after Reject; it did not appear on the homepage. The only 24metrics request seen was GET gateway.clickshield.24metrics.com/ping (200, EU-hosted at Hetzner). Its response was server metadata only (region hetzner-eu, a server timestamp and a version number), which looks like a connection check rather than a data upload, though any such request still tells 24metrics the visitor's IP address and browser. 24metrics sells affiliate traffic-quality scoring and click-fraud detection, and it is not listed anywhere in the site's cookie list.

Alternatives

The verdict

Cheap to start, costly to use fully. Premium is €13.99 a month (as low as €3.99 on a 12-month plan) with 100 tokens, but images, voice, calls and private content all spend extra tokens.

No commercial relationship with this platform.