How We Test
Most rankings in this category are sorted by commission rate. Ours is sorted by a published formula that anyone can check.
The six scoring categories
| Category | Weight | What it measures |
|---|---|---|
| Conversation | 25% | Model quality, roleplay coherence, response latency. |
| Memory | 20% | Long-term recall architecture and user control over it. |
| Media | 20% | Image and video generation quality, cost and throughput. |
| Voice | 10% | TTS quality, live calls, custom voice design. |
| Value | 15% | Cost per unit of real usage, not headline price. |
| Trust | 10% | Corporate transparency, refund terms, billing conduct. |
Verified vs reported
We label every claim on this site. Verified means we confirmed it ourselves, by inspecting the platform's own shipped client code, reading its public API responses, or pulling its company registration from the relevant national registry. Reported means we have the operator's published claim and nothing stronger.
This distinction matters more than it sounds. Marketing copy in this category is frequently contradicted by the operator's own legal terms. We have found platforms advertising "lifetime" benefits whose contracts quietly cap them at six months. Where the marketing and the contract disagree, we score the contract.
What we will not do
- We do not invent testing stories. If we have not verified something, the page says so rather than dressing it up as hands-on experience.
- We do not hide the commercial relationship. We work with many of the platforms we list, and every page that carries a partner link says so.
- We do not alter the facts for a partner. Pricing, capabilities and company details are recorded as we find them. A partner can ask us to correct an error; it cannot ask us to remove an accurate one.
- We keep caveats short and fair. Every platform has trade-offs. We note them plainly without turning a review into a warning.
Why trust is a scored category
Ten percent of every score is corporate conduct: whether the operator can be identified, whether its refund terms are lawful where the buyer lives, and whether there is a documented pattern of billing complaints. A platform can have excellent technology and still be a bad purchase. A ranking that ignores that is not much use to someone spending their own money.
Corrections
Scores change when the facts change. Platforms currently covered: Secrets AI, Candy AI, Swipey AI, DreamGF, Nomi AI, CrushOn AI. Last revision 2026-08-22. If you can show that something here is wrong, we will correct it and note the change.
See the rankings →