Line up the pricing pages of eight AI companion apps and you'll find the same dozen bullets on all of them. The apps themselves are nothing alike. Every claim below is technically accurate everywhere, and each one covers a spread wide enough to decide whether you'll like the product.
"Memory" or "remembers you"
The spread: at one end, six facts re-sent with every message. At the other, stored facts plus rolling summaries plus retrieval that holds up across months of history.
Both get called memory. The difference is whether an app is fun for a week or still fun in month four. The details are in how AI companion memory works.
The question to ask: Can I view and edit what it saved? A memory screen tells you the storage is real and organized. Kupid AI lists memory across sessions. Nomi is built around it, which explains its conversation score in our ranking.
"Unlimited messages"
The spread: truly uncapped text, or uncapped text under a rate limit, or uncapped text on one tier with media metered on every tier.
The word is usually honest and usually misread. Text isn't the costly part. Images and voice cost the operator money every time you use them, so the meter lands there.
The question to ask: Unlimited what, and what still burns credits?
"Voice"
The spread: the character reads replies aloud as audio clips, or you hold a live spoken conversation and can interrupt.
Two different products, one word. Voice notes are cheap and everywhere. Live calls cost more, tend to sit on an upper tier, and rise or fall on delay. Wait three seconds for every answer and the effect is gone.
The question to ask: Is this a call or a voice note, and how long is the lag?
"Uncensored" or "no filters"
The spread: looser than the mainstream apps, all the way to much looser than the mainstream apps. Never truly unfiltered. Every operator hard-blocks the same illegal core, and payment processors enforce it.
The question to ask: Where's the line, and has it shifted lately? See content filters for more.
"Custom character creation"
The spread: choose from four hair colors and three personalities, or work through a multi-step builder covering looks, voice, backstory, speaking style and topics to avoid.
Depth here tracks how long an app stays interesting, since a character you built is one you can rebuild when it gets stale. Camsoda AI centers on a six-step builder. That's a real product difference, not just a bullet.
The question to ask: Can I edit the character later, or only during setup?
"AI image generation"
The spread: a new face on every render, or a character you can recognize across dozens of pictures.
Consistency is the whole contest, and a pricing page can't show it.
The question to ask: If I make four pictures of one character in four settings, is it the same person?
"Long-term relationship" or "grows with you"
The spread: a relationship level that ticks up as you use the app, or a character whose stored history actually changes how it answers.
Progress bars are cheap to build and easy to mistake for depth.
The question to ask: What concretely changes at level five? Describe it in words.
"Free tier"
Every app we track has one. It might be something you could live on for months or a three-day trailer. What free tiers really include covers the difference.
The question to ask: What happens when I hit the wall: a daily allowance resetting, or an upgrade prompt?
What ties it all together
Each claim names a category, and the product lives in the implementation. That's why comparison tables in this market are nearly worthless, and why we test the way we do. Two weeks on a paid account shows which end of each spread an app really occupies.
If you've only got ten minutes, use two shortcuts. Favor the app with fewer, deeper bullets. Breadth in this category usually means shallowness. And favor the one that lets you see inside: a memory screen, an editable character, a visible credit balance. Apps that show you the machinery almost always have machinery worth showing.
