How We Check

First, we use the app

Every app we review, we have used ourselves, on our own account, before giving it a score. The provider’s pages say what an app promises; using it shows what it does. The score rests on the second.

Five things we check in use:

• How far the free tier goes. We chat until the app stops us, count the messages, and note whether the counter resets each day or runs out once.
• Whether memory holds. We mention something specific, start a new session and ask about it.
• What voice and video actually give you. A live call and a text-to-speech clip are both sold as “voice”. On a free account the answer is often a lock icon, and the review says so.
• Where the filter fires. What gets refused, on which plan, and whether the refusal comes from the interface or from the character making up an excuse.
• What the checkout shows before you pay. Which plans are really on offer, the total charged and the renewal terms. We go as far as the payment screen and stop there.

Tests run on a free account. We do not keep paid subscriptions, so a feature that only a paid plan unlocks is described from the provider’s own pages, and the review marks which is which.

Reviews with screenshots say when they were taken. The Candy AI review has the full set, the Lovescape review shows the pricing screen, and screenshots for the other reviews are being added.

Then we check what the provider claims

One account on one day is a sample. Four more steps, the same for every app, set what we saw against what the provider publishes:

• Read the pricing page. We take the month-to-month figure, not the annual plan divided by twelve, which is the number most sites quote as the headline price. Where a provider hides the monthly rate behind a struck-through anchor, the review says the number is reconstructed rather than published. Where the site refuses automated reads altogether — Muah AI answers with HTTP 403 — we say the price could not be checked at source instead of repeating a figure found elsewhere.
• Set the free limit against what the provider says. Messages per day or in total, and whether the counter resets. Where a provider does not publish it — Candy AI does not — we count it ourselves and say the number is ours, not theirs.
• Read the terms. Who the company is, which country the contract sits in, how refunds work, whether cancelling is a button or an email, and what the statement descriptor says.
• Re-check monthly. Prices, limits and content policies in this category change without notice. Every figure carries the month it was verified.

Where our account and the provider’s pages disagree, the review shows both. Candy AI’s free message count visibly changes from one account to the next, so the review gives the five messages we got alongside the range other testers report.

How the score is built

The score out of five is editorial judgement applied to eight criteria. They are listed in order of importance: what a month really costs comes first, then whether the free tier lets you judge the app, then memory.

• Real monthly cost. The month-to-month price, plus what tokens, coins or credits cost at normal use.
• Free tier. Whether you can judge the app before paying: how many messages, and what is locked.
• Memory. Whether the character still knows you in the next session, on the plan you would buy.
• Voice, images, video. What they actually deliver, and whether they sit in the plan or are metered.
• Content rules. Where the filter fires, whether the policy is written down, and whether it has held steady.
• Refunds and cancelling. The refund window, how cancelling works, who the company is and where the contract sits.
• Chat privacy. How long logs are kept, whether they train the model, whether deleting the account deletes them.
• User ratings. Public ratings, and repeated complaints about billing and support.

There is no formula behind the number. How an app does on each criterion is our call — nothing settles mechanically whether 16K of memory you can read and edit beats 24K with facts pinned by hand — and a weakness near the top of the list counts for more than one near the bottom.

What the ranges mean:

• 4.5 and above. The strongest app we have tested. Its weak points are in the review, not left out of it.
• 4.0–4.4. Worth paying for, with one serious limitation to know about first.
• 3.5–3.9. Good for a specific use, with several limitations you will run into.
• Below 3.5. We would not pay for it. It is listed so you can see why.

Where the method stops

Our tests are short and run on free accounts, and nothing here is written from months of daily use. A problem that appears only after weeks — memory fading after 7–10 days of heavy use, for example — comes from repeated user reports, and the review says so.

We do not accept payment for placement, do not send apps a preview of what we are going to publish, and do not remove criticism at a provider’s request. The affiliate relationship is explained in full on the disclosure page.

If we have something wrong

Prices move and policies change, sometimes within days. If a figure on this site is out of date or plainly wrong, tell us and point at the page it came from — corrections are faster than arguments. Contact details are on the contact page.