What this desk actually is
Tested Companions is an independent English-language review desk for adult AI companions and one live 1v1 video-chat product. We publish on testedcompanions.com, with longer Candy and LuckyCrush pieces on candyai.guide and luckycrushreview.com. We are not Candy AI, LuckyCrush, EverAI, Hello World SAS, Xotic, Joi, Lovescape, or FantasyGF. We do not host login forms for those products. We do not impersonate their domains.
The objects under review are entertainment products for adults. Characters are fictional. Live video involves other adults. We do not test anything involving minors, and we will not write “how to jailbreak a filter” copy. “Uncensored” in our tables means adult fiction or live adult is in-product. It never means “no law, no policy.”
Scores sit on a 1–10 scale. They are editorial. A 7.6 is not a mean of 153 hidden rubrics. It is a dated judgment after the five checks below, written so a reader can disagree with the arithmetic in public.
Five boring checks
1. First-party paper
Subscriptions pages, help centers, terms, privacy notices, refund language, blocked-content lists. Dated. Quoted. When EverAI’s help center says Premium includes 100 tokens (fetched late August 2026), that sentence goes in the review with the fetch date. Marketing adjectives (“ultra-realistic”, “unlimited fun”) are ignored unless a checkout or help article backs them.
2. Checkout arithmetic
Monthly vs quarterly vs annual. What is billed up front. What auto-renews. Token, chip, XOT, and Neuron packs. Per-action rates from official price lists where they exist (Xotic publishes XOT costs) and from independent paid testers where the vendor will not print them (Candy’s 2 / 4 / 10 / 20 ladder in July 2026). We convert packs into a dollars-per-clip or dollars-per-minute number a human can use.
3. Company records
Malta C107181 for Candy / EverAI Limited. Hello World SAS, Paris, RCS 838 199 917, plus Tech World Entertainment B.V. in Doetinchem for LuckyCrush. Xotic Tech OÜ, Estonia 17225128. Warmtech Ltd, Cyprus HE 418620, for Lovescape. JOI-AI LIMITED, UK 13414098, for Joi. ADVANCED AI LABS S.R.L., Romania, for FantasyGF. A real company is the due-diligence floor, not a character reference.
4. Demand + complaints
US Google Ads search volume so we know which questions are real. “candy ai” ~246,000/month and “luckycrush” ~165,000/month in the 2026 window we cite on the hub. Trustpilot and forum piles are sampled, not averaged into fake precision. A 2.8/5 on ~30 LuckyCrush reviews is a smell. A 4.8/5 on 40 Xotic reviews is also a thin sample. We say the sample size in the review.
5. What the product is for (and is not)
Every stub names a buyer and a skip. Candy is chat-first with a pixel meter. Xotic is itemised video spend. Joi is a 5-second clip toy with a second currency. Lovescape is chip video. FantasyGF is a Candy-like meter with crypto checkout. LuckyCrush is live men–women 1v1, not an AI girlfriend. Mixing those categories in a single “best AI girlfriend 2026” list is how affiliate mills lie. We will not.
Source hierarchy
When sources conflict, this order wins:
- Live checkout and in-app confirm screens. Prices move overnight. The page you see today is the authority.
- Vendor help center and terms, with a fetch date.
- Company registries and processor names printed on those terms.
- Independent paid-account ledgers published in 2026, named when we rely on them.
- US Google Ads Keyword Planner volume (12-month averages unless we say otherwise) as a demand signal, not as quality.
- Trustpilot, Reddit, and app-store piles, sampled. We quote patterns. We do not pretend 30 reviews are a census.
Affiliate roundups that reprint each other’s Neuron tables without a date are not sources. If we cannot date a number, it does not go in a table.
US Google Ads volume as a demand signal
Search volume tells us which queries a human actually types. It does not tell us who is good. Candy’s brand query is a quarter-million US searches a month. LuckyCrush is ~165,000. “luckycrush review” is ~110. Most people who type the brand want the product, not the critique. That is why review domains have to say, in the first screen, that they are not the product.
Volume also flags SERP vacuums. Xotic’s official site is xotic.ai. Hyphenated lookalikes (xotic-ai.com and cousins) exist because the vendor’s organic SEO is weak. We review the product at xotic.ai. We do not squat a fake login on a hyphen domain. If you landed here on a hyphenated query, the official app is still xotic.ai; this page is the desk note.
We will update volume figures when we pull a new Planner export. Until then, treat the hub’s 1 September 2026 line as the dated snapshot.
Trustpilot is sampled, not averaged
Paid random cam and adult AI products polarize. Five-star “it works if you pay” next to one-star “bots and billing.” Averaging that into 3.7/5 and calling it science is how content farms pad a table. We do this instead:
- Record the score, the adjective Trustpilot prints, and the review count on the fetch date.
- Read the newest page of reviews, not only the highlight reel.
- Separate “the company took the money and 404’d” from “I did not like the minutes.” LuckyCrush fails the second complaint often and the first one rarely.
- Ignore invited-review streaks when the sample is tiny. Forty five-stars on a new AI companion is a marketing event until it is not.
How a score is built
The 10-point number is a weighted editorial call, written in public:
| Weight | What we look at | What can move it |
|---|---|---|
| ~30% | Price honesty: sticker vs media meter | A published rate card helps. A $3.99/mo headline that becomes $50 in clips hurts. |
| ~25% | What you actually get (chat, stills, clips, live people) | Itemised video and live 1v1 are not the same product. |
| ~20% | Company paper and billing clarity | Registry + processor named in terms. Descriptor you can dispute. |
| ~15% | Complaint pattern, sampled | Thin 4.8 and thin 2.8 both get a caveat, not a trophy. |
| ~10% | Who it is for, stated without hype | A product that names its skip earns the last point. |
Current desk scores (1 Sep 2026): Candy 7.6, Joi 7.4, Lovescape 7.3, FantasyGF 7.1, Xotic 7.0, LuckyCrush 6.2. LuckyCrush is a different category. It is scored against live cam value and trust, not against Candy’s token ladder.
What we will be wrong about
Vendors change prices without a press release. Promo codes rot. Token rates drift. A help-center sentence from 28 August can be stale on 8 September. When that happens, checkout wins and we patch the page. We will not pretend a snapshot is a contract.
We do not operate a 30-person lab. We do not generate a thousand clips per product every month. We triangulate vendor paper, paid-tester ledgers, and our own checkout captures. If a number came from someone else’s paid account, we say so. If we have not personally burned a Neuron pack this week, we do not write as if we did.
Cadence
This protocol is dated 1 September 2026. The hub comparison table uses the same date as dateModified. Brand stubs inherit it. When a vendor ships a new SKU (Candy Premium Plus at 330 tokens; Lovescape Wan 3.0 in late August 2026), we note the ship date rather than silently rewriting history.
Corrections: hello@testedcompanions.com. Send a URL, a screenshot date, and the number you think we got wrong. We do not provide product support for the brands. That is their help center.
FAQ
No. That number is a content-farm prop. We run five checks, date them, and publish the arithmetic.
Yes. Help centers and tester ledgers are supporting evidence. Homepages are advertising.
To separate real demand from invented roundups. Volume is not quality. It tells us which brand queries are worth a careful review.
No. See the disclaimer. Product URLs are candy.ai, luckycrush.com, xotic.ai, joi.com, lovescape.com, fantasygf.com.