How we test

Every app on Synthlust is tested the same way: we use the product for a minimum of 30 days and score it across six categories. Nobody pays for placement. If a product sucks, we say it sucks.

The scoring rubric

Each product is rated 1–10 across six axes:

• Chat Quality (weight: 30%). How coherent are the responses? Does the character stay in personality? Does memory persist? We run a standardized set of 15 conversation prompts across every app, plus a week of freeform daily use.
• Image Generation (weight: 20%). Resolution, consistency with the described character, how long it takes, how many credits it burns. We test the same five character descriptions on each platform.
• Voice (weight: 15%). Latency, naturalness, whether it supports real-time calls or only pre-recorded messages. Products without voice don't get penalized below 5. They just can't score above it.
• Value for Money (weight: 15%). Price relative to what you get. A $3/month app that does one thing well can score as high as a $13/month app that does everything.
• Uncensored (weight: 10%). We test against a standard battery of prompts that range from vanilla to extreme. A product that claims to be uncensored but filters half of our test set loses points.
• Privacy (weight: 10%). Data retention policy, breach history, signup requirements, whether the company has been transparent about incidents.

The overall score

The headline number is a weighted average of the six category scores, rounded to one decimal. We don't grade on a curve: if every app improved, every score would go up. The current spread (7.6–9.4) reflects the real gap between the best and worst products in this category.

What we don't do

• We don't let companies see their review before it publishes.
• We don't adjust scores based on affiliate commission. The highest-commission product on our site is not our top pick.

Update cadence

We re-test every product quarterly and update scores when features change materially. If a product has a major release or incident between cycles, we'll test and update early. Last full re-test: April 2026.