How We Test AI Companions, Cam Sites & Adult Games

bestgirlfriend.ai is an editorial comparator covering AI girlfriend apps, AI boyfriend apps, live cam platforms, adult games, and real-creator subscriptions. We are independent, ad-free, and reader-supported. This page is the parent of 5 dedicated scoring pages and explains how the 5 systems fit together, who runs the tests, what we deliberately don't claim, and how affiliate revenue gets kept out of the editorial scoring chain.

The three best consumer methodology pages in adjacent spaces (Wirecutter's "How we work", Consumer Reports' "How we test", and the published test methodologies at RTINGS) share three properties: a public dated methodology, named human testers, and an explicit firewall between commercial relationships and editorial scoring. We built ours on the same three properties, adapted for AI companions, live cams, adult games, and real creators.

How does bestgirlfriend.ai test AI girlfriend apps?

The protocol is identical across every AI companion in our catalog: same 10 prompts, same image set, same voice phrase. That's the only way two scores stay comparable. Vary the test and you can rank anything #1 by accident. The eight categories and their weights live on how we score AI companions, with the sub-tests written out so any third party with a free account and a stopwatch could re-run the test on us.

I'd say about half the time, the score I had in mind after the free-tier prompts is the score the paid tier confirms. The other half, paying reveals either a much better product (Joi's long context window only really fires once you've paid) or a much worse one (some catalogs gate the actually-good prompts behind a tier most users won't unlock). Both outcomes get logged.

How does bestgirlfriend.ai score AI companion apps?

The weights aren't arbitrary. They came out of about six months of reading complaint patterns specific to this space: Reddit threads, app store reviews, the occasional tip-line email. When readers churn from an AI girlfriend app, they complain about pricing surprises, persona inconsistency, and privacy worries far more often than they complain about voice realism or video gen quality [Source: Wirecutter, How Wirecutter Works (testing methodology precedent) · verified 2026-05-26]. The weights reflect what actually hurts and helps the people using these products, not what the marketing departments push at us [Source: Google Search Central, E-E-A-T quality rater guidelines · verified 2026-05-26].

Privacy and Compliance gets 14 percent because that's the dimension where a bad call costs the reader the most. A leaked chat log or a billing descriptor reading "ADULT SERVICES LLC" on a shared family card is not a 6-month-later concern, it's a same-week concern.

How does bestgirlfriend.ai test live cam sites?

Cam scoring is built differently because the products are built differently. There's no Conversation Quality category because there's no AI persona; there's a real person on camera doing what humans do. There's no Image Generation category because the models are real human beings, not pixels. The two heaviest weights (Model Variety and Pricing and Tipping Flow at 18 percent each) reflect the two things that, when broken, cause cam viewers to leave a site within minutes. Full criteria and the testing toolchain at how we test cam sites.

The $0-spend rule matters more than I thought it would when we set it. About once a quarter, a cam brand asks why we won't recharge their tokens to "experience the full product." The answer is that the moment we recharge, we've created a stake in the recharge converting, and the only honest version of cam scoring is one where we have no stake in the recharge converting.

How does bestgirlfriend.ai score adult games?

Billing Transparency only exists in the adult games scoring because the adult games space made it necessary. Scam-detection services routinely flag legitimate brands here as "high risk", usually because of the way the billing descriptor changes across the 4 or 5 payment processors these games rotate through. No competitor in this space grades it openly. We do, because false-positive billing fraud flags are the most common reason a reader emails us saying "your recommendation showed up on my card as something I didn't recognize." Full criteria at how we test adult games.

A reader sees a charge labeled "SEGPAY SUPPORT BV" on a card statement, panics, and disputes it. The chargeback often hits before the user remembers signing up for the game. Half the time the brand isn't shady. The descriptor just doesn't tell you what you bought. We grade for that.

How does bestgirlfriend.ai score real creators?

The conflict-of-interest discipline matters here more than on any other scoring system on the site. A reviewer who subscribes to a creator to score them creates a stake in that creator-reviewer relationship that no disclosure resolves cleanly. So we score from public-facing data only: free preview profiles, the cadence of promotional posts on the creator's X account, aggregated Reddit subscriber commentary (we require at least 5 independent reports for any verified claim), and industry press where it exists.

Some categories are always partially un-verified by us as a result. Engagement and Interaction (DM response time, custom request quality) is the obvious one. We flag the gap transparently on every creator scorecard rather than fake first-hand data we don't have. Full criteria at how we score real creators.

How does bestgirlfriend.ai score affiliate networks?

The B2B affiliate network scoring is the inverse of the other 4 systems. Where the others commit to $0 spend, this one commits to active operator experience. The lead reviewer holds live publisher accounts on every network scored, and the dashboard, payout, tracking-architecture, and account-manager-responsiveness scores get graded from six months of direct hands-on. Full criteria at how we score affiliate networks.

Disclosure that matters: CrakRevenue is our primary monetization partner. CrakRevenue's 9.2/10 score stands regardless. If their payouts dropped to zero tomorrow, the score would still reflect their offer breadth, dashboard polish, and payout history. The editorial-commercial firewall is the entire point.

Why 5 different scoring systems instead of one?

The table below shows the three biggest scoring systems side by side. The categories, weights, and review counts are not interchangeable, and that's the point. A reader using an AI girlfriend app does not care about peak-hour model variety on a cam site. A reader using a cam platform does not care about persona consistency across ten conversation turns. Calibration is the entire point of building 5 systems instead of 1.

Last reviewed: 2026-05-26

How are bestgirlfriend.ai scores calculated?

Composite scores carry a tier label so readers don't have to reverse-engineer what 7.6 versus 8.4 actually means. The thresholds are the same across the 5 scoring systems: 9.0 and above is Best in class, 8.0 to 8.9 is Excellent, 7.0 to 7.9 is Strong, 6.0 to 6.9 is Good or Mixed depending on which scoring system, 5.0 to 5.9 is Average or Weak, 4.0 to 4.9 is Below average or Skip, and anything at or below 3.9 is Avoid. A platform scoring below 5.0 is not recommended on bestgirlfriend.ai regardless of commission. That's rule one of the Score-Lock Framework, which I'll explain below.

Do affiliate commissions affect bestgirlfriend.ai scores?

This is where I get to call out what most reviewers in this space do, and what we do differently. Most of the AI girlfriend, cam, and adult game reviews you've read online were written by people whose payout depends on how flattering the review is. That's not a moral failing on their part. It's just how their incentive structure works. It's also how you end up with 5 different sites all ranking the same brand #1, the same brand that happens to have the most aggressive commission.

We score each platform before we ever look at what it pays. The score gets version-locked on publish, and any later movement requires a documented re-test with the change logged in the page's update history. "Commission rate change" is an explicitly forbidden reason in those logs. The full disclosure setup (the Score-Lock Framework, our click-tracking codes, the rel="sponsored" tag applied to every commercial link) is documented on our affiliate disclosure page. The methodology side of the contract is editorial; the disclosure side is commercial. Both sides cross-reference each other on every review we publish.

I'd rather lose a commission than soften a take. The commission goes away if we lose reader trust anyway.

How often are bestgirlfriend.ai scores updated?

The cadence below is the floor, not the ceiling. Detected changes (a new pricing page, a regulatory docket dropping, an infrastructure migration announcement, a UI overhaul tweet from the platform's official account) trigger an out-of-cycle re-test on the affected category. Each review's hero shows both the Last full retest date and the per-category last-tested date, so readers can see at a glance which numbers are fresh and which are due.

Last reviewed: 2026-05-26

What is the bestgirlfriend.ai Score-Lock Framework?

The Score-Lock Framework governs every commercial decision on bestgirlfriend.ai. It exists so the firewall between commerce and editorial is testable, not just promised. Every rule has a trigger, a published cadence, and an audit trail.

Rule	What it means in plain English	Trigger / audit
Score-floor	No platform with a composite below 5.0 is recommended on bestgirlfriend.ai	Composite below 5.0 = exclusion from listicles and recommendations
Score-lock-at-publish	Scores fix at publish; only documented re-tests on cadence can move them	Each update-log entry must cite a re-test or a trigger event; "commission rate change" is a forbidden reason
CTA freshness automation	Affiliate links auto-validate weekly against the offer source-of-truth	Stale offers get stripped from CTAs; the review stays online with a note

The Score-Lock Framework is documented in identical wording on the affiliate disclosure page. The two pages are the two halves of the same trust contract.

Who tests the platforms?

Alexandra Joly holds a Licence Information et communication from Université Sorbonne Nouvelle (2013-2016), worked as a freelance journalist on tech and society topics from 2022, and joined bestgirlfriend.ai as Senior Editor in January 2026. The byline on every review is verifiable on her LinkedIn and the editorial workflow is documented on the about page. Editorial review by the in-house team resolves any score disagreement greater than one point before publish. A second named editor will be added in a future revision once the team grows.

One more thing on the who, since readers ask. We publish in volume, and the honest way to do that is with help: AI tools draft the prose from my test notes and translate the pages into the languages we run in. They test nothing and set no scores. Every number traces back to a test a human ran, and I read every published line before it ships.

If you find a factual error, a stale data point, or a score that does not reconcile with the cited evidence, write to [email protected]. Corrections get logged at the top of the affected page for 60 days and recorded in the page's update log. The audit trail of corrections is the trust signal.

Can I see bestgirlfriend.ai's raw test data?

The asymmetry between public and private artifacts is intentional. Public artifacts (regulatory dockets, Wayback captures, app store snapshots) get linked directly because anyone can re-fetch them. Private artifacts (full chat transcripts, full screenshot sets, our internal scoring spreadsheets) stay under editorial control because publishing them would expose unrelated PII, raw model outputs that aren't ours to republish, and internal calibration notes that aren't editorial output. The middle ground is the sanitized summary on request.

How do you compare an AI girlfriend to a cam site or a creator?

A page comparing Candy.ai to Stripchat, for example, never displays "Candy.ai 8.4 / Stripchat 7.6" as if those numbers lived on the same scale. They don't. Instead, the bridge page walks the reader through what AI gives that cam doesn't (unlimited persona consistency, full customization, no peak-hour scarcity) and what cam gives that AI doesn't (real human spontaneity, real-time tipping economy, real bodies). The verdict says "depends on your intent" and names the two intents explicitly.

I've been asked maybe a dozen times why we don't just collapse everything into a single 0-10 score per platform regardless of category. The answer is that the moment we do, we'd be inventing precision that doesn't exist. The category-by-category narrative is harder to write but it's the only honest version.

What does 'we haven't tested this directly' mean on a score?

This is one of the 5 non-negotiables below. The post-purchase reality of cam recharge (the billing descriptor on the credit card statement, the refund-claim friction, the auto-renewal triggered after seven or more days) is the most common group of un-verified data points on cam reviews, because we never recharge. For those sub-criteria, we rely on aggregated user reports: at least 30 recent reviews across review aggregators or Reddit threads, cross-checked across two independent sources, with a footnote on the score naming the secondary sources. Honest uncertainty beats false certainty every time.

How does the versus consistency rule work?

The versus consistency rule is one of 5 non-negotiables shared across every scoring system. The 5 are tabulated below. They apply to every published review on the site regardless of which scoring system the page falls under.

Last reviewed: 2026-05-26

Where do I read the full criteria for each category?

This page is the executive summary; the dense work lives on the 5 subpages. Each subpage repeats the parent landing's cross-system explanation in its first section, then dives into the categories, the sub-tests, the un-verified flag policy, the version history, and a FAQ specific to that product type. The 5 subpages also cross-link sideways. Readers landing on how we test cam sites can jump to how we score AI companions in one click, how we test adult games in another, how we score real creators in another, and how we score affiliate networks in another, and link up to this parent page for the cross-system narrative.

How do I report an error in your methodology?

A correction request that arrives on Monday gets acknowledged the same day. If we agree with the correction, the page is updated within seven business days and the change gets logged in the page's update history. If we disagree, we explain why in the email reply and, if the disagreement is substantive, we publish the exchange (with the requester's permission) on the page itself. The audit trail is what separates this site from a marketing surface that lifts scores on request and erases the changes silently.

Frequently asked questions

Last reviewed: 2026-05-26

Sources

1. The New York Times Wirecutter, "How Wirecutter Works" (current revision). nytimes.com/wirecutter/about
2. RTINGS.com, TV test methodology change-log. rtings.com/tv/tests/changelogs
3. Federal Trade Commission, Advertising and Marketing guidance (16 CFR Part 255). ftc.gov/business-guidance/advertising-marketing
4. Google Search Central, "Creating helpful, reliable, people-first content" (E-E-A-T criteria, current revision). developers.google.com/search/docs/fundamentals/creating-helpful-content

Cite this page

If you reference this methodology page in academic, regulatory, or journalistic work, please cite as:

Joly, Alexandra (2026, May 26). How bestgirlfriend.ai Tests AI Companions, Cam Sites, Adult Games, Real Creators, and Affiliate Networks. bestgirlfriend.ai. https://bestgirlfriend.ai/methodology

Related pages

• How we score AI companions: 8 categories, free-tier testing protocols, version history.
• How we test cam sites: 6 categories, $0-spend hands-on checkout protocol, regulatory record handling.
• How we test adult games: 7 categories including the Billing Transparency one specific to this product type.
• How we score real creators: 6 categories for individual creator subscriptions, $0-spend public-data audit, no-subscription protocol.
• How we score affiliate networks: 6 categories for B2B adult-niche affiliate networks, first-hand operator-account audit.
• About Alexandra Joly: Senior Editor profile, credentials, LinkedIn, contact.
• Editorial process: how reviews get written, fact-checked, and updated.
• Affiliate Disclosure: how we earn money and the Score-Lock Framework that keeps it out of editorial.
• Contact us: for corrections, methodology challenges, and journalist inquiries.

---

Alexandra Joly, Senior Editor · Last reviewed 2026-05-29