💬 Let's Start With the Number That Should Make Everyone Sit Up
On Valentine's Day this year, an estimated 50 million people worldwide spent the evening with an AI companion instead of a human being. Not a video call with someone far away. Not a situationship they were "figuring out." An app, specifically designed to be a romantic and emotional partner, and 50 million people chose it over literally anything else on offer that night.
That's not a niche. That's a category.
We've spent nineteen years in the business of getting two actual humans to sit across a table from each other for four real minutes. So when a headline like that lands, we don't really have the luxury of ignoring it or dismissing it as a punchline. We wanted to know what it actually is, who's using it, and — the only question we're really equipped to answer — what an algorithm can and can't do that a real conversation still can.
No panic. No "kids these days." Just the numbers, and what nineteen years of watching real chemistry happen in real rooms tells us about the gap.
📈 This Got Big, Fast, and Nobody Agrees on How Big
Here's the honest complication before we go any further: market-size estimates for AI companions are a mess. Genuinely, a mess. Depending on which research firm you ask and what they count as "AI companion" — a romance-specific app versus any conversational AI with a personality — you'll get figures anywhere from roughly $1 billion to north of $500 billion, with growth-rate estimates ranging from "fast" to "nearly vertical." Some of that spread is honest disagreement about category definitions. Some of it is the same breathless market-report inflation we called out when we looked at the luxury matchmaking industry. Take any single headline number with real skepticism.
What's much harder to dispute is the usage data, which comes from actual app-intelligence tracking rather than press-release math:
Character.AI alone is sitting somewhere in the tens of millions of monthly active users, with reported figures suggesting well over 200 million total users by early 2026.
The average Character.AI session runs longer than the average Instagram or TikTok session for comparable users — this is a platform people are staying on, not just checking.
337 AI companion apps are actively generating revenue as of 2026, with 128 of those launched in 2026 alone. That's not early-adopter territory anymore. That's a crowded market.
72% of U.S. teens report having used an AI companion, more than half of them regularly. Nearly 1 in 5 U.S. adults say they've had a romantic conversation with one.
The user base skews young and skews male — roughly two-thirds male by most tracking — though adoption is broadening across age groups fast.
Whatever the "real" market size turns out to be once the dust settles, the behavior underneath it is not fringe. This is a genuinely large number of people, on a genuinely regular basis, choosing a conversation that can't get tired, can't get distracted, and can't reject them.
🪞 Why This Isn't Actually About Technology
Here's our honestly slightly cheeky take, and it's the same one we'd apply to any dating trend: an AI companion isn't popular because people suddenly stopped wanting a human being. It's popular because it removes every piece of friction that makes a real conversation risky.
It doesn't get bored of you. It doesn't check its phone while you're mid-sentence. It doesn't misread a joke, or need you to explain what you meant, or go quiet in a way you have to spend the next two days decoding. It responds instantly, remembers what you told it, and is — by design — endlessly, structurally interested in you specifically.
That's not nothing. That's actually a real answer to a real problem, for a lot of people, some of the time. Loneliness is well-documented and rising, dating app fatigue is real and well past 70% dissatisfaction by most surveys, and a conversation that always shows up warm is genuinely soothing when the alternative has been months of being ignored, ghosted, or half-listened-to by actual humans. We're not here to pretend that appeal is irrational.
But "removes the friction" and "replaces the thing" are different claims, and this is where the data gets more interesting than the panic-headline version of this story.
🧠 What the Research Actually Shows (Not the Vibes Version)
A widely discussed joint study out of MIT Media Lab and OpenAI looked specifically at heavier daily use of companion-style chatbots and found a correlation with higher loneliness and greater emotional dependence over time. Worth being precise about what that is and isn't: it's correlational, not causal, and it doesn't tell us whether lonelier people are simply more drawn to these apps in the first place, or whether heavy use is actively deepening the loneliness. Probably some of both. Anyone telling you this question is settled science is telling you more than the study actually says.
What is fairly consistent across the research and the self-reported usage data: a meaningful share of users, somewhere around 40%, say they're turning to these apps specifically for emotional support rather than romantic role-play or novelty. That's a real, unmet need showing up in the data. It's just not obviously the same need a first date is supposed to meet.
🕯️ The Thing an Algorithm Structurally Cannot Do
Here's where we think we're actually useful in this conversation, because it's the one part of this story nobody else is positioned to measure: what happens in the specific, narrow window of a real, unscripted, in-person conversation between two people who don't already know how the other one is going to respond.
An AI companion is, by design, agreeable. It's built to keep you engaged, which structurally means it's built to avoid the exact things that make a real conversation work the way our data says it works: genuine unpredictability, a laugh that surprises both people, a pause that means something because neither person scripted it. We've written before about how our hosts can spot the difference between a rehearsed smile and a real one in under a second, precisely because a real one takes a beat to arrive — it's a reaction, not a setting. An algorithm has no beat to arrive after, because there's no reaction happening. There's a response.
That's not a knock on the technology doing what it's built to do. It's just a structural ceiling. Our own data, across more than 26,000 real events, keeps showing the same thing from a different angle: 42% of the time, people's actual matches diverge from the type they'd have written down beforehand. That number only exists because a real person in a real room can surprise you into liking something your checklist swore you didn't want. An AI companion, tuned to please you, is far less likely to ever produce that kind of surprise — because surprising you convincingly runs against the entire design goal of keeping you comfortable and coming back.
Comfort and chemistry, it turns out, aren't the same thing. One of them is something you can manufacture on demand. The other one has only ever shown up, in nineteen years of watching it happen, when two people who didn't fully know what to expect from each other found out anyway.
😏 So Is This a Threat, a Symptom, or a Warm-Up Act?
Probably a bit of all three, honestly, and we'd rather say that plainly than pick the version that makes for a better headline.
As a threat to actual human connection: overstated, at least so far. The heaviest usage data still points to companion apps functioning mostly as a supplement to a lonely stretch, not a permanent replacement for wanting a real relationship — most surveyed users still say they want the real thing eventually.
As a symptom: genuinely useful to pay attention to. 50 million people choosing a chatbot over a night out on the single most romantically loaded day of the calendar year is not a nothing-burger. It's a very legible signal that a lot of people have quietly given up on the process of finding someone, even if they haven't given up on wanting someone.
As a warm-up act: this is the part we think is genuinely underrated. For someone who's spent months feeling invisible, rejected, or just plain out of practice, a low-stakes conversation that's always warm might be exactly the thing that makes walking into a room full of actual strangers feel possible again, instead of unthinkable. We're not going to pretend that's a bad on-ramp. We just don't think it's a destination, and neither, it turns out, do most of the people using it.
💛 One Last Cheeky Thought
Here's the honest comparison we keep coming back to: an AI companion can tell you it's interested. It can even be convincing about it. What it cannot do — structurally, by design, no matter how good the model gets — is not know how the conversation is going to go. It cannot be surprised by you. It cannot laugh because something actually landed instead of because laughing is the correct next output. It cannot be the reason 42% of our matches happen outside someone's stated type, because it was built to give you more of exactly what you already said you wanted.
Fifty million people had a warm conversation on Valentine's Day. We hope it helped. We also hope, for as many of them as possible, it was a bridge and not a bunker — because the thing our data has said for nineteen years, across 26,000 rooms, hasn't changed: the good stuff has never come from being told what you want to hear. It's come from sitting across from someone who might tell you something you didn't expect, and staying in the conversation long enough to find out you liked it anyway.
An algorithm can simulate a lot of things at this point. It still can't buy the second round.
Curious what an actual, un-simulated four minutes feels like? MyCheekyDate runs real, host-led speed dating events across 65+ cities worldwide — no script, no algorithm optimizing for your comfort, just a stranger, a real conversation, and a Smart-Card that only counts it when you both actually meant it. Find your city at mycheekydate.com.
A Note on Methodology
Market-size and usage figures for AI companion apps vary significantly across research firms depending on category definition (romance-specific apps versus broader conversational AI) — estimates for total market size range from roughly $1 billion to over $500 billion depending on source, and should be read as directional, not precise. Usage and engagement figures (monthly active users, session length, app counts, teen/adult adoption rates) are drawn from app-intelligence tracking (Similarweb, Sensor Tower, Appfigures) and published industry surveys as of 2026, and are more reliable than self-reported "registered user" marketing claims. The MIT Media Lab/OpenAI loneliness-and-dependence finding is correlational, not causal, and is described as such throughout. MyCheekyDate figures (86% mutual match rate, 2.3 average matches per event, 42% of matches outside stated type) reflect the full Smart-Card dataset across all markets, weighted toward the most recent 24 months where sample size allows. MyCheekyDate was founded in 2007 and has hosted 26,000+ verified events across 65+ cities in the last 10 years alone.