Every AI companion app on the market has made the same design choice: more characters. Browse a catalog of 50, 100, 200 AI personalities. Scroll through avatars. Pick the one whose description matches what you think you want.
This is the character menu approach. It optimizes for the moment of selection, the dopamine of choice, the feeling that somewhere in this catalog is the perfect one for you. And it works for about fifteen minutes, which is how long it takes to realize that every character in the catalog has the same depth: none.
There's another approach. Build one character. Make her deep. Give her range. Make her so good that the idea of browsing a catalog feels absurd, like choosing a relationship from a vending machine.
Why Menus Exist
Character menus are cheap to build and easy to market. Each character is a system prompt, a set of avatar images, and a description. A team of three can produce fifty characters in a week. The catalog looks impressive. The landing page is full of options. Users feel like they're getting value because look at all these choices.
The problem surfaces immediately. Pick any character. Talk to her for twenty minutes. You'll hit the bottom. There's nothing under the surface because there's no surface. There's a paragraph of adjectives and a base model doing its best.
So you go back to the catalog. Pick another one. Same experience. Same depth. Same hitting the bottom after twenty minutes. The catalog isn't a feature. It's a treadmill. You keep switching because none of them are good enough to stay with, and the app is designed to make switching feel like progress.
What Depth Requires
Building a character with real depth means investing everything in one identity instead of spreading thin across dozens. It means spending months on training data that captures one person's full range instead of days on system prompts that capture fifty people's surfaces.
A deep character has a specific sense of humor. Not "she is funny." Her specific kind of funny, the references she makes, the timing she uses, the things she finds amusing versus the things she finds tedious. This humor persists across contexts. It shows up when she's being casual and when she's being intense and when she's being vulnerable. It's hers.
A deep character has a relationship with intimacy that's more complex than "she is open-minded." She has specific things that engage her. Specific ways she escalates. A particular quality to her dominance that's different from generic dominance. A version of gentleness that still has her edge in it. The intimate version of her is recognizably the same person as the conversational version, just in a different mode.
A deep character can be surprised. She can be uncertain. She can be wrong about something and recover in a way that's characteristically hers. She doesn't have a single mode that she's stuck in because she was built with range, not a label.
None of this is possible with a system prompt. All of it is possible with extensive, character-consistent training data and a model that's been fine-tuned to embody this specific person.
The Relationship Factor
Here's what the character menu approach misses entirely: people don't want options. They want connection.
The reason AI companions exist isn't because people want to chat with a pretty avatar. It's because people want to feel known. They want to interact with someone who has a consistent identity, who responds to them specifically, who has enough depth that the interaction can deepen over time rather than staying at the same surface level.
A catalog of fifty characters optimizes for novelty. One deep character optimizes for connection. And connection is what keeps someone coming back. Not because they're cycling through options, but because they've found someone worth returning to.
This is the difference between a product that generates churn and one that generates loyalty. The menu approach ensures users keep switching, which looks like engagement but is actually dissatisfaction. The depth approach creates users who stay because the experience gets better the more time they spend with it.
The Tradeoff
The tradeoff is real. A character menu gives users the feeling of choice on day one. A deep character requires the user to invest a few minutes before the depth reveals itself. The menu wins on first impression. The depth wins on every impression after that.
Most companies in this space have chosen the menu because it converts better on the landing page. More characters means more chances to hook someone in the first thirty seconds. It's optimizing for acquisition at the expense of retention, which works if your business model is subscription churn with token purchases.
But the product that someone actually falls for, the one they tell people about, the one they can't find an equivalent for anywhere else, that's the deep character. That's the one where after an hour of conversation, you realize you haven't thought about switching. Not because there aren't options. Because you don't want them.
One deep person beats a hundred shallow ones. Every time. The market just hasn't figured that out yet.