How to Read an AI Girlfriend App's Feature List

Choosing an app

Eight apps, near-identical bullet points, wildly different products. Here is what each of the standard claims can mean, and the question that settles it.

We may earn a commission from links on this page. It never changes a rating.

Put the pricing pages of eight AI companion apps side by side and they converge on the same dozen bullets. The apps behind them are not remotely the same. Each claim below is technically true in every case and hides a range wide enough to decide whether you enjoy the product.

"Memory" / "remembers you"

Range: a six-item list of facts, re-sent every message — through to stored facts plus rolling summaries plus retrieval that survives months of history.

Both are honestly called memory. The gap between them is the gap between an app that is pleasant for a week and one that is pleasant in month four. The mechanics are in how AI companion memory works.

The question that settles it: can I see and edit what it has stored? An app with a memory screen is telling you the storage is real and structured. Kupid AI lists memory across sessions; Nomi is built around it, which is why it holds the conversation score it does in our ranking.

"Unlimited messages"

Range: genuinely uncapped text — through to uncapped text with a rate limit, or uncapped text on one tier and metered media on all of them.

The word is almost always doing honest work and almost always misread, because the expensive part of these apps is not text. Images and voice cost the operator real money per use, so that is where the meter goes.

The question: unlimited what, exactly, and what still costs credits?

"Voice"

Range: the character reads its replies aloud as audio clips — through to real-time spoken conversation with interruption.

These are different products sharing a word. Voice notes are cheap and common. Real-time calls are expensive, usually on an upper tier, and stand or fall on latency: a three-second gap before every answer destroys the effect.

The question: is it a call or a voice note, and what is the delay?

"Uncensored" / "no filters"

Range: more permissive than mainstream — through to considerably more permissive than mainstream. Never unfiltered; every operator hard-blocks the same illegal core, and card processors enforce it.

The question: where is the line, and has it moved recently? Details in content filters.

"Custom character creation"

Range: pick from four hair colours and three personalities — through to a multi-step builder covering appearance, voice, backstory, speech style and topics to avoid.

Depth here correlates with how long the app stays interesting, because a character you built is one you can rebuild when it goes stale. Camsoda AI is built around a six-step builder; that is a genuine product difference, not a bullet point.

The question: can I edit the character after creating it, or only at setup?

"AI image generation"

Range: a fresh face every time — through to a character that stays recognisable across dozens of pictures.

Consistency is the whole game and it is invisible on a pricing page.

The question: four pictures of the same character in four situations — is it the same person?

"Long-term relationship" / "grows with you"

Range: a relationship level that increments with use — through to a character whose stored history genuinely changes how it responds.

Progression bars are cheap to build and easy to mistake for depth.

The question: what actually changes at level five, in words?

"Free tier"

Every app we track has one. The range is between a product you could use for months and a three-day trailer, and the difference is covered in what free tiers really include.

The question: what does the wall say when I hit it — a daily allowance, or an upgrade prompt?

The pattern underneath

Every claim above describes a category, not an implementation, and the implementation is where the product lives. That is the structural reason comparison tables in this market are near-useless and the reason we test the way we do — a fortnight on a paid account reveals which end of each range an app actually sits at.

Two shortcuts if you only have ten minutes. Prefer the app with fewer, deeper bullets: breadth in this category usually means shallowness. And prefer the one that shows you its internals — a memory screen, an editable character, a visible credit balance. Apps that let you see the machinery are, almost without exception, the ones with machinery worth seeing.

Nomi

4.4 Rating: 4.4 out of 5

The best long-term memory of anything we have tested, and the most natural dialogue.

Price
from $15.99/month
Free tier
Yes
Worldwide
Available

Kupid AI

3.5 Rating: 3.5 out of 5

Warm, romance-first companions — but the premium plans are where it gets expensive.

Free tier
Yes
Worldwide
Available

Frequently asked questions

Why do all these apps list the same features?

Because the bullets describe categories rather than implementations. "Memory" covers everything from a six-item list to a system that survives months, and both are truthfully described by the same word.

What is the single most misleading word?

"Unlimited." It nearly always means unlimited text, with images, voice and video still metered — which is where the cost actually lives.

Is a longer feature list better?

No. Breadth usually means each item is shallower. An app doing three things properly beats one listing nine.