State of AI Companions, Q2 2026: 49 Reviews Under One Rubric

By Alexandra Joly, Senior Editor • Published 2026-05-13 • our scoring approach • Trend analysis

How we built the dataset

I locked the 49-Review milestone as the Q2 2026 cut on May 11, the day our visible scoring shipped to production. By the time this post goes up the dataset stands at 56 published Reviews across nine categories, with every score card public. Each Review uses one of four scoring systems, built to fit the category it covers [Source: Google Search Central (Creating Helpful Content guidance) · verified 2026-05-13]. AI companions get eight dimensions. Cam sites get six. Adult games get seven, including a Billing Transparency dimension with no parallel in the other three. Real creators get six, scored per individual person rather than per platform.

We spend $0 per platform per re-test cycle. We walk pricing pages and checkout flows right up to payment, then stop before we submit. So anything that only shows up after you pay (the billing descriptor on your bank statement, how hard it is to get a refund, when auto-renewal actually fires) gets flagged as something we haven't tested directly, because we didn't subscribe. The things we did verify ourselves carry no such flag: image-generation quality at default settings, voice latency from free-tier samples, how specific the terms of service get, breach-disclosure history. Wherever a cell is flagged, the published Review shows you why in a footnote.

The four scoring systems never roll into one. A page comparing Chaturbate to Candy.ai won't show you side-by-side composite scores. It walks category by category instead, then lands on a verdict tied to what you actually want ("for live human interaction, Chaturbate; for a customizable always-on persona, Candy.ai; both can be true for the same reader"). That rule, set out on our scoring page, stops the apples-to-oranges collapse that mashing everything into one number would create [Source: Schema.org Review and Rating specification · verified 2026-05-13].

The dataset at a glance

Most of what we tested lands in a 6.5-to-7.5 mid-band. The reason is that our scoring punishes the usual industry weak spots (privacy compliance, hidden costs, voice latency, unverifiable production claims) hard enough that breaking 9.0 takes evidence on every dimension, not marketing on one. Only two scores cleared 8.0. Candy.ai at 8.4 and OurDream at 8.2, both AI companion apps that publish twelve-plus legal URLs each and hold persona consistency across our full 10-prompt conversation test.

Finding 1: Voice technology trails text by a measurable margin

Voice Quality is the lowest-scoring dimension across the AI companion apps we tested, trailing Conversation Quality by 2-to-3 points on the median Review across the fourteen scored AI girlfriend platforms. Several flagship apps ship voice that scores 1-to-2 out of 5 in our audits. Joi is the most documented case we found, its voice rated below the median and often described as missing or unusable by reviewers we cross-referenced from Reddit and YouTube side-by-sides. Candy.ai sits mid-pack on voice even though it leads on image generation and compliance. Secrets.ai gates voice behind a paid tier, so we could only score it on what the platform claims, not on anything we verified ourselves.

It comes down to latency. Real-time text-to-speech for adult conversational AI needs to deliver its first audio frame in under 2 seconds to feel natural, and most of the apps we tested ship 3-to-5-second windows with audible compression on top. ElevenLabs publishes a 75-millisecond model tier as a developer benchmark, but the cost per minute pushes most operators in this space toward slower, cheaper tiers [Source: ElevenLabs developer documentation on latency · verified 2026-05-13] [Source: W3C Web Content Accessibility Guidelines 2.2 (audio alternative) · verified 2026-05-13].

Why this matters if you're choosing: voice is the closest thing AI offers to live-cam interaction, which is exactly why cam sites don't share this weak spot. If voice is what you care about, think of the AI apps as text-first products with voice bolted on, not as voice-first products. The AI girlfriend top 8 flags the voice score for every pick, and if voice is your deciding factor, you're better off browsing the cam sites instead.

Finding 2: Cam dispersion sits in compliance, not broadcast quality

Across 23 cam Reviews, the median composite is 6.8, and the spread sits on Privacy & Compliance and Payment & Geo Coverage rather than on Broadcast Quality. Broadcast Quality clusters above 7.5 for all the top platforms (Chaturbate 7.5, Stripchat 7.4, Jerkmate 7.2, LiveJasmin 7.2, BongaCams 6.8). HD streams, low buffering, and 1080p webcam support are table stakes now. What actually separates these sites is regulatory exposure and where you can pay from.

Chaturbate carries a documented $675,000 FTC settlement from April 2024 over record-keeping violations under 18 USC 2257 [Source: 18 USC 2257 record-keeping statute · verified 2026-05-13]. Stripchat geo-blocks the United Kingdom under Ofcom guidance while the Online Safety Act's age-verification deadline plays out [Source: UK Online Safety Act 2023 · verified 2026-05-13] [Source: Ofcom online safety guidance · verified 2026-05-13]. LiveJasmin has the strongest privacy posture of the bunch: a named DPO, GDPR-compliant retention disclosure, and clear jurisdiction (Luxembourg plus Cyprus).

Why this matters if you're choosing: if you just want the best HD stream, the top platforms are basically tied. If you want a platform that survives FTC scrutiny, or one that isn't tangled up in UK age-verification, the differences get a lot sharper. Each Review's compliance footnote carries the specific regulatory record, and our cam scoring page spells out the 16% we put on Privacy & Compliance and the 16% on Payment & Geo Coverage.

Finding 3: Pricing dispersion is wider in AI than in Cam, and Adult Gaming sits in its own world

The cheapest entry point we scored across 56 published Reviews runs from $0 (Chaturbate freemium browsing) to $24.99 a month (Joi Premium), with AI companion entry tiers bunched between $9.99 and $19.99 and cam token packs starting at $10.99 for 100 tokens. Adult games sit outside this dollar comparison completely. Hentai Heroes and Harem Villa run on microtransaction stacks ("kobans" or equivalent hard currency), where the cost per scene is the price that actually means something, not a monthly subscription [Source: EU Directive 2011/83 on consumer rights · verified 2026-05-13].

There are two layers to this. Among AI companions, the gap from Candy.ai's $12.99 base to Joi's $24.99 base is roughly 2x. Among cam sites, the gap from Chaturbate's $0 entry to LiveJasmin's $30-plus private-show entry can hit 30x, except token pricing makes that division meaningless. Auto-renewal clarity is what evens it out. Every platform we scored above 8.0 ships either a 7-day money-back guarantee, a one-click cancel button you can actually find, or an EU-compliant 14-day withdrawal grant. Every platform we scored below 6.0 either drew auto-renewal complaints on Trustpilot and Reddit or ran an opaque token stack you can't convert to a clean dollar figure.

Why this matters if you're choosing: a straight dollar comparison only tells you so much, because the categories price so differently. Pricing carries 18% of the score in three of our four systems and 20% in adult games, because price is the single strongest thing pushing a decision across every category we cover [Source: FTC 16 CFR Part 425 negative-option marketing rule · verified 2026-05-13].

Finding 4: Real Models have the highest median composite, but the data is thinnest

Real-creator Reviews carry the highest median composite at 7.18 across 16 published creators, but this is also our thinnest dataset, because the per-creator signals (DM responsiveness, posting reliability, archive depth) are the hardest to verify when we spend $0. OnlyFans Neko and OnlyFans Gabby Epstein both score 7.5. OnlyFans Bridgette B and OnlyFans Simone score 7.4. The Fanvue creators land at 7.0-to-7.2, predictable on production quality and lighter on verified engagement.

The honest caveat is how much of this we couldn't test directly. Content Volume & Cadence we can check from a creator's public promo profile. Engagement & Interaction, worth 18% of the score, is where we most often flag "we haven't tested this directly," because actually testing it would mean paying for a subscription. Privacy & Compliance inherits a floor from the platform: OnlyFans's 18 USC 2257 record-keeping process and KYC posture set the minimum, and creator-specific things (watermarking, whether they honor geo-blocks) stack on top, as our real-models scoring page lays out.

Why this matters if you're choosing: the high median is real, but it reflects how the category is built (verified creators on regulated platforms tend to clear the floor our scoring sets), not some claim that any one creator is better than the rest. If you're picking from here, treat the things we couldn't test directly as your uncertainty band. Each Review's footnotes spell out exactly what we couldn't verify and why.

Finding 5: Billing transparency is the differentiator we didn't expect

Billing Transparency is a 10% dimension that only exists in our adult-games scoring. It covers auto-renewal language, how many payment processors a site uses, whether the billing descriptor stays consistent, the refund window, and how many clicks it takes to cancel. It's the one thing other reviewers skip most often, and the one thing reader complaints pile onto hardest. Harem Villa carries a billing-transparency aggregate around 38 out of 100 on scam-detector indexes (third-party aggregator scoring, which we haven't independently verified). Hentai Heroes has a documented browser-redirect malware history in public reports we can't re-test ourselves. The pattern holds across the board: compliant brands with technically legal terms still ship billing flows that read as predatory to a first-time subscriber [Source: FTC 16 CFR Part 255 endorsement guides · verified 2026-05-13] [Source: Have I Been Pwned breach catalog · verified 2026-05-13].

We added this dimension because our own competitor research turned up exactly zero outlets grading auto-renewal clarity, refund friction, or descriptor consistency on adult-gaming brands. Not one. Trustpilot, scam-detector, and Reddit threads complain about this stuff constantly, and yet nobody on the publishing side was actually aggregating the complaints before we started scoring it. Right now it only applies to adult games, because our AI companion and cam scoring already fold refund and renewal into Pricing & Value at 18%. If the AI-side data shows the same gap, we'll pull Billing Transparency out as its own dimension there too.

Why this matters if you're choosing: treat it as a 7th question to ask yourself before you hand over card details to any subscription-based adult platform.

Finding 6: Methodology anchors seem to predict which pages AI search engines cite (preliminary)

The pages we've actually seen cited by Perplexity, ChatGPT, Claude, and Gemini in our May 2026 spot-checks all share three things: a methodology version stamped on them, a named author byline (Alexandra Joly, with a LinkedIn profile the engines can verify), and an average of eight or more primary-source citations per Review. That said, this tracking is still preliminary, so read it as a signal, not a finished study. The pattern itself isn't new. Wikipedia citations, Wirecutter reviews, and named-byline analyst reports already dominate what AI answer engines pull from [Source: GDPR Article 5 (data minimization and accuracy) · verified 2026-05-13].

Here's what surprised me: pages with hollow credentials don't get cited even when they rank fine in regular search. One competitor ranks top-5 on "best AI girlfriend" with a "Coursera Bachelor 2005-2009" line in the author bio. Coursera launched in April 2012 [Source: Coursera company history · verified 2026-05-13]. So that degree is impossible. That same site has no methodology page, no version history, no public errata board. It ranks the old-fashioned way, sure, but it never shows up in our Perplexity spot-checks for the same query.

Why this matters if you're choosing: the engines that cite sources reward structure that plain rankings never ask for. The pages that win on both fronts carry a methodology, a named author you can actually look up, primary-source citations, and clean structured data under the hood. The price of that is editorial discipline, plain and simple. What you get back is being found everywhere (regular search plus the AI answer engines) instead of in just one place.

Patterns that surprised us

Three things cut against what I assumed going in. I figured the AI apps would lead and everything else would trail. The numbers said no.

Broadcast Quality clusters tighter than AI Conversation Quality, even though cam streams feel more variable than chatbot output. The reason is that webcam hardware and streaming infrastructure are commodities now, while AI chat quality hangs on which language model the app routes to and how hard it filters. I'd have bet on the opposite.

Real creators score the highest median, which runs against the whole "AI is the future" framing. The honest read is that AI is just the wider-swinging category (scores ran from 5.0 to 8.4), while creator-subscription platforms sit in a tighter band, because OnlyFans and Fanvue enforce compliance floors that filter out the weaker offerings before they ever reach our scoring queue [Source: OnlyFans transparency report and content moderation guidelines · verified 2026-05-13].

Adult games have the highest Pricing weight (20%) and the narrowest composite spread (5.5-7.4). When you weight pricing heavily and most operators ship near-identical microtransaction stacks, the scores get pulled toward the middle. What actually separates these games lives in the dimensions we weight lower, Art Direction at 12% and Game Mechanics at 16%, which cuts against the usual line that pricing drives the decision. After scoring all seven, here's my read: weight art direction and Hentai Heroes leads, weight game balance and Harem Villa edges ahead. A single composite number buries both of those.

What this means for readers

Here's where to go depending on what you actually want. No inline affiliate links on this post, that's deliberate.

Looking for the best AI girlfriend overall? The AI girlfriend top 8 ranks the top 9 picks with a full breakdown on every dimension. Candy.ai is the top scorer, and the full Candy.ai Review shows you how that 8.4 composite breaks down.

Want to understand how four scoring systems coexist without collapsing into one number? Our scoring approach explains how it all fits together, links to each of the four pages, and spells out how we handle comparisons that cross categories.

Weighing AI against cam as a category? Our head-to-head comparisons hub carries 22 published matchups, each with a verdict tied to what you're after. The AI-versus-cam guide walks you through the decision itself.

Picking a cam site? Our Chaturbate teardown sits at the top of the freemium pack, and our cam scoring page lays out all six dimensions and the $0-spend way we test.

Picking a real creator? The OnlyFans Neko Review is the top scorer among the creators we tested, and our real-models scoring page lays out the six dimensions plus how we handle anything we couldn't test directly.

If one page carries most of the data this post chews on, it's the AI girlfriend top 8, with a full dimension-by-dimension breakdown for the top picks. The next dataset cut goes up once published Reviews cross 75 and our four scoring systems land their next round of updates.