How we score AI companion apps
Five criteria with fixed weights, two calibration anchors, and a stated research basis on every profile. This is the scorecard behind every number on the site.
By AI Companion Radar Editors
· 4 min read
Every app on AI Companion Radar is scored out of ten on the same five criteria with the same weights. The number is recalculated whenever we re-test, and every profile shows the date of its last review. This page is the scorecard behind that number, so you can see exactly what a 7.4 means and where you might weigh things differently.
The five criteria
Criterion | Weight | What we look at |
|---|---|---|
Conversation quality | 30% | Whether replies stay in character, follow instructions, avoid repetition and hold a scene over a long session. Model choice and reply length limits count here. |
Memory | 20% | Recall of facts, events and preferences across sessions; whether you can see and edit what the app remembers; whether memory is gated behind a higher tier or extra credits. |
Media generation | 20% | Image and video quality, character consistency across generations, reliability of completion, and the real cost per image or clip. Apps with no media generation are scored on the other four criteria and the result is scaled to ten. |
Pricing honesty | 15% | Is the price visible before sign-up? Does the renewal price match the intro price? Is there a lasting free tier? How much of the product sits behind tokens, coins or credits on top of the subscription? |
Content policy clarity | 15% | Whether the published terms say what adult content is permitted and what is banned, whether the marketing matches the terms, and whether age assurance is explained. A clear “not allowed” policy scores well here; “unfiltered” marketing over terms that ban explicit content scores badly. |
How a score is built
Each criterion gets an integer from 1 to 10. The overall score is the weighted average, rounded to one decimal. Two anchors keep the scale honest across reviewers: Candy AI sits at 8.0 (conversation 8, memory 7, media 9, pricing 7, policy 9) and OurDream AI at 7.8 (8, 8, 7, 7, 9). An app that beats both on every criterion should score above them; an app that only beats them on marketing should not. How the two anchor apps differ in practice is set out in OurDream AI vs Candy AI.
Desk research versus hands-on testing
Every profile states its research basis. Desk-researched profiles are built from the app’s own site, pricing pages, terms, store listings and a reading of public user reports. Hands-on tested profiles were run by us on a paid account for at least a week. Where an app hides its pricing behind a login, we say so and quote what reviewers report, with the date. Model disclosure is part of the same reading, and the apps that name their models are listed separately.
What does not affect the score
Whether the app pays anyone. It does not pay us; we run no affiliate links and sell no placements.
User counts, awards and “number one” badges. We do not repeat them.
Traffic or popularity. A widely used app with poor memory still gets a poor memory score. The reverse holds too: several little-known apps score near the top of the directory.
Rankings by use case
The overall ranking orders apps by their score. Use-case rankings such as best for video or best memory use the same criterion scores but only include apps where the feature exists on a published plan, and readers should treat them as a shortlist rather than a verdict.
Corrections and re-tests
Profiles are re-checked at least once a quarter, and sooner when an app changes its pricing or policy. If you spot an error, email hello@aicompanionradar.com with the app name and a source. Corrections with a source are usually live within a week and do not change scores on their own; a re-test does. Refund terms are not scored on their own, but the refund policies of 55 apps are compared in one table.
Apps in this article
MyBabes
Multi-feature
8.3
The broadest creative toolkit in the category: eight named video engines with published per-second prices, two image model lines, a story studio, AI-refereed roleplay games and a creator paywall. Browser only, no voice calls, and the token maths take a minute.
Images
Video
Voice
Memory
From €9.99/mo (3-month plan)
Editor’s pick
Candy AI
Multi-feature
8.0
The most complete media stack in the category, chat, voice, images, 120-second Live Action video and Story Mode, but the subscription is only the entry fee. Tokens for anything visual push real spend well past the headline price.
Images
Video
Voice
Memory
From About €3.99/mo (annual promo); €13.99 monthly; tokens extra
Editor’s pick
OurDream AI
Multi-feature
7.8
A US-registered companion with a published coin price list, decent DeepSeek-based chat and a real free tier. Video is the weak link: short, expensive and often unreliable. Good value on the annual plan for chat and images.
Images
Video
Voice
Memory
From $19.99/mo or $119.88/yr ($9.99/mo) with 1,000 DreamCoins/mo
More from Radar
AI companion apps and your privacy: what 55 apps do with your chats
44 of 55 apps store your chats, 13 may train on them, 34 do not say. The privacy terms of every app in the directory, in one table.
Kindroid pricing: plans, add-ons, credits and what a month costs
Kindroid costs $13.99 a month on the web, $15.99 in the apps, $98.97 with both add-ons. Every price, three worked monthly bills, the post-October trial, cancel steps and refund wording.
SpicyChat vs MyBabes: roleplay chat or generation studio?
SpicyChat wins on text roleplay, named chat models, free chat and its $5.95 entry price. MyBabes wins on images, video, stories and creator income. They suit different users.
How to cancel Kupid AI: weekly billing, refunds and deletion
Cancel Kupid AI in your account on kupid.ai or by emailing hello@kupid.ai. Weekly plans renew every seven days, sales are final under the terms, and plan Coins expire after six months.
What images and video really cost on 55 AI companion apps
Ten different in-app currencies, most rates unpublished. The lowest plan, the unit, and the price of an image and a clip on every app in the directory.
CCBill charge on your card: what it is and which AI app billed you
A CCBill charge means a site that uses CCBill billed you. Look it up on support.ccbill.com, cancel there, ask for a refund, and see which AI apps name their processor.
Dondi AI pricing: Premium, token packs and the 24-hour refund rule
Dondi AI Premium is $14.99 a month ($5.99 on the yearly plan) with 200 tokens; images from 5 tokens, clips 50. Refunds only within 24 hours despite a 7-day promise. Worked bills and fine print.
Gay AI boyfriend apps: which ones have a real male-for-male catalogue
Janitor AI, SpicyChat and Chub AI have real gay-tagged catalogues; MyBabes has a dedicated “Create him, gay” builder; OurDream AI asks your orientation; Kindroid lets you build him.
Is Eva AI safe and worth it? Privacy, prices and the catch
Eva AI is run by Novi Limited in Cyprus and collects your messages to develop its AI with no opt-out; prices run $11.99 to $15.99 a month plus unpriced Neurons, and store purchases are final.