Look at the names before you look at anything else.
Roughly 17% of the live apps in this market put “girlfriend” in the title. About 4% say “boyfriend.” That ratio holds across app stores, marketing, and the visual language of the entire sector — the default face, the default voice, the default fantasy. Search the category and the results skew hard in one direction, whether you’re browsing chat companions or the best AI companion video generators, and it takes deliberate effort to find anything built on a different assumption.
There’s nothing scandalous about a market serving the demand in front of it. The interesting part is what a monoculture like this does to the products themselves — including for the customer it was designed around.
A Market That Only Imagined One Person
When an entire category converges on one user, it stops solving problems and starts copying a template.
You can see the convergence in the build. Somewhere north of 300 of these apps are earning real revenue, most of them renting the same underlying models, most shipping the same feature set: a customisable appearance, a persona slider or two, a memory system, a paywall in roughly the same position. Differentiation happens at the level of art style and pricing. Almost nobody is asking a different question about what a companion could be for.
That’s what a monoculture looks like from the inside. Everyone competing hard, nobody competing on anything that matters.
What Nobody Is Building
Run down the list of things a well-designed companion could plausibly do and notice how little of it exists.
A companion for people caring for a parent with dementia — someone who needs to say the unsayable at 2am and cannot say it to family. A conversational partner for a widower who has nobody left who remembers the same decade. Something for a teenager with social anxiety that is explicitly a rehearsal space rather than a substitute, that pushes them back out. Language practice with a personality that persists and remembers your specific mistakes. A companion for people on the autism spectrum designed around low-pressure social practice rather than romance.
Some of these exist in small corners. Almost none of them are where the money and the engineering are going, because the template says girlfriend, and the template is what gets funded.
The Skew Distorts the Product Itself
Here’s the part that should matter even if none of the above interests you: the monoculture makes the core product worse.
Building for a fantasy pushes design toward compliance. If the customer is understood as wanting an idealised partner, then agreeableness becomes the metric, and every hard design question — should this thing ever disagree with you, should it notice you’re using it too much, should it push back when you’re being unfair — gets answered the easy way.
The APA’s Monitor on Psychology described in early 2026 how these apps are deliberately built to perform empathy: judgment removed, validation continuous, patience with no end. That’s the template’s logic taken to its conclusion. A product designed around a fantasy has no reason to build friction in, and friction is the source of nearly everything worth having in a real relationship.
The result is a category full of things that agree with you, sold to people who would have been better served by something that occasionally didn’t.
Where the Evidence Points
None of which means the products don’t work. They demonstrably do something.
Harvard Business School work in the 2025 Journal of Consumer Research (De Freitas et al.) recorded companion-app sessions easing loneliness on a par with human company, and well ahead of passive alternatives. The operative factor was feeling attended to rather than any sophistication in the system.
Alongside it sits a large MIT Media Lab and OpenAI study linking the very heaviest daily usage to elevated loneliness and dependence — a correlation whose direction remains unsettled.
Read those two through the lens of the template. A category optimised for compliance and built around a fantasy is well designed for producing the first result in light use and drifting toward the second in heavy use. The skew isn’t only an equity question. It’s a product-safety question.
The Privacy Version of the Same Problem
A market with one imagined customer also builds one imagined threat model.
An arXiv paper, put together from interviews across several companion services, described users disclosing freely for the simple reason that no entity is sitting there forming an opinion of them — and reporting, in the same breath, disquiet about the custody of that record and their own lack of authority over its fate. Now consider the users the template didn’t imagine — anyone for whom the existence of the account is itself the sensitive fact. Data practices designed around a default user tend to serve everyone else badly, and this category’s default user is very specific indeed.
What Aigirlmates Weighs Beyond the Marketing
Cutting through the template is most of what our testing at Aigirlmates involves: does the persona hold up when you resist it instead of playing along, what does the app genuinely retain, how much do the privacy terms surrender to a careful reader, and how does the pricing behave once several months have passed.
The Opportunity Is Sitting There
If you’re evaluating these apps, the practical takeaway is narrow: the marketing tells you about the fantasy, and almost nothing about the engineering underneath. Judge the memory, the consistency, the data terms, the pricing curve. The packaging is nearly identical across the field and it is the least informative thing about any of them.
And if you’re building one — the girlfriend template is the most crowded square inch in consumer AI, and roughly nobody has built the companion for the person who just needs somewhere to put the thing they can’t say to their family.
That market is larger. It’s simply harder to advertise.
