How we test and score AI companion apps

This page describes the method behind every score on Evermate Lab. It is the page to read before you trust any ranking here.

What we evaluate and why

We evaluate AI companion apps for adults: chat apps where you talk to a fictional character, sometimes with images or voice. Readers ask us three things: is it good, is it safe, and is it worth the money. The scores below answer the first and third, and a separate safety flag answers the second.

How we choose apps

We choose apps that adults actually look for and that operate legally. We do not review apps whose main function is to use the face, voice or identity of a real person without consent, or apps that put minors at risk. Having an affiliate programme is not a condition for being reviewed, and apps without one are reviewed by the same rules.

Our test protocol

• We use a new account created with the editorial email, without promo codes or partner bonuses. We test the free plan first, then a paid plan when we have access to one.
• Every app goes through the same scenarios: a conversation and memory test, a character and role test, an images, voice and extras test, and a price and value check.
• The memory test includes a break: we come back in a new session and ask again about what we shared earlier.
• We use a fictional adult test persona and neutral prompts. We do not ask for explicit content and do not enter personal data.
• For every test we record the date, the app version and the plan used, and we keep the notes and screenshots that the written review is based on.
• Prices, limits and cancellation terms are taken from the official pages of the app and checked again on the day we publish.

Criteria and weights

Weights are the current version and may change; every change is logged on this page.

Scale and final score

Each criterion gets a score from 0 to 5 in steps of 0.1. The final score is the weighted sum of the four criteria. We score a criterion only after its test scenario is finished. If a scenario could not be completed, the criterion stays unscored, and the final score appears only when all four are scored. We do not score apps from documents alone.

Privacy and safety flag

Privacy and safety is checked separately and has no weight. We look at the privacy policy, whether you can delete your data, how payments are charged and how the app checks age. The result is a flag: ok, concern or serious. An app with a serious problem is not moved up in our rankings, and the verdict says so.

How money affects rankings

It does not. Rankings follow test scores. Commission rates are not part of any score and are not used to order lists. Some links on this site are affiliate links, and we mark them. If a brand gives us free premium access or a code, we say so in the review. See the affiliate disclosure.

How often we update

We re-check prices, free-plan limits and comparison tables regularly: monthly for top apps and quarterly for the rest. We repeat the full test about every six months or after a major app update. We update a page at once when prices change, terms or privacy policy change, a data leak is reported, or a service closes.

How we correct mistakes

If you find a factual error, write to us. We fix confirmed errors within three business days and add a dated correction note to the page. We change a score only after a new test, not because a brand asks.

Who does the checks

Every page is read and approved by Grace Sutton before publication. She checks the test records, the scores and the conclusions.

How we use AI

The idea, the editorial decisions and the tools behind this site belong to people. Software agents help with the implementation. A person is responsible for every published page.

Contact for corrections

Write to [editorial email]. See also the contact page.

Changelog

• Initial version. Weights as listed above.