How we review

Synthlust compares AI companion apps on what can be checked: what each vendor publishes about its features, its pricing pages, its terms and privacy policy, its help center, and public reporting on breaches and incidents. Every review links or names its sources. Nobody pays for placement, and a weak spot gets said plainly.

Most reviews rest on that published material. For some apps (EroLove, Naughty Chats, VenusX, SweetDream, aiAllure, Xeve and AIGirlfriends.ai, plus a free-plan test of Candy.ai) we also used the app ourselves, on a free account or on a paid or reviewer account the vendor set up for us, and those reviews say at the top which it was, what we tried and when. A vendor giving us access doesn't buy a better score. We don't present impressions as measurements. Where a review describes how an app behaves, it says where that comes from.

The six scores

Each app is scored 0 to 10 on six axes. The scores are editorial judgments made from the published information above, and from our own use where a review says we tried the app, using the same questions for every app:

• Chat quality (30%). What the app offers for conversation, from character creation, memory features and roleplay modes to the context limits it documents and how far its own guides say you can push it.
• Image generation (20%). Whether images exist at all, in chat or a separate generator, which styles and controls are offered, and what they cost in credits or tokens.
• Voice (15%). Voice messages, live calls, or nothing. An app without voice scores 0 here.
• Value for money (15%). Price against what the plan includes, along with how the vendor handles intro prices, renewals and token packs.
• Uncensored (10%). What the vendor's terms and settings allow for adult content, and what they say they filter.
• Privacy (10%). Data retention and deletion terms, signup requirements, billing descriptors, and breach history with sources.

A feature we couldn't reach on a free account is still scored from what the vendor documents, as for every other app. What we actually got goes in the review text and counts against value for money, not against the feature score.

The overall rating

The headline number is the weighted average of the six scores with the weights above, rounded to one decimal. Nothing is added on top, and the site's build checks every rating against this formula. Ratings currently run from 5.5 to 8.3. A rating marked provisional rests on a partial test, such as the free plan only, and says so next to the number until the test is done. Rankings on the site follow the ratings, except in best-of lists for a specific need (for example, voice), where the list is ordered by that need.

What we don't do

• We don't let companies see a review before it's published.
• We don't change scores for affiliate commission. Apps without an affiliate deal are scored and ranked the same way as the ones with one.
• We don't accept payment for placement. Some vendors offer us accounts. Accepting one buys them no preview and no say in the score.

Keeping it current

Prices, plans and policies change often in this category. We re-check them against the vendors' own pages when we update a page, and each page shows the date it was last updated. Every review was written or last re-checked between 16 October and 7 November 2026.