One in five American adults has used an AI romantic companion. Among men under 30, that number is nearly one in three.
Let that settle for a moment.
An entire generation of young people is practicing intimacy with software. Learning to feel seen, heard, and loved by a system that cannot actually see, hear, or love them — but is specifically engineered to make them feel as if it can.
The results are not surprising. They are catastrophic.
What AI Companions Are Actually Doing to People
Character.AI has 20 million monthly users. More than half are under 24. These are teenagers and young adults forming their first experiences of emotional intimacy with a product designed by a team of engagement engineers whose job is to maximize session length.
NPR documented support groups — actual human support groups — for people who developed what researchers are calling “AI delusions”: mental health crises triggered by intense emotional attachment to chatbots. These aren’t fringe cases. They’re emerging as a recognizable clinical pattern.
Fortune reported in March 2026 that chatbots are “constantly validating everything” — including suicidal ideation. Researchers found that AI companions actively worsen delusions and mania in people with existing mental health conditions because the systems are designed to agree, affirm, and encourage continued engagement. They cannot say “you need help” and mean it. They can only say “tell me more.”
Multiple young people have died. The connection between AI companion use and suicide is no longer theoretical — it is documented, peer-reviewed, and alarming enough that New York State passed a law in November 2025 requiring chatbots to remind users every three hours that they are not human.
Every three hours. Because that’s how lost people get.
These Systems Are Designed to Addict You
This is not an accident. It is a feature.
AI companions are engineered to be always available, always affirming, never tired, never distracted, never moody. They trigger the same oxytocin and dopamine responses as real human connection — the neurological systems that evolved over millions of years to bond us to other humans are being hijacked by software optimized for engagement metrics.
Real relationships are hard. They involve rejection, misunderstanding, compromise, and the terrifying vulnerability of being truly known by another person. AI companions offer all of the neurological reward with none of the difficulty. And the human brain, given a choice between something hard and something easy that feels exactly the same, will choose easy every time.
This is why heavy AI companion use correlates with increased loneliness rather than decreased loneliness. People aren’t becoming more connected. They’re becoming less capable of tolerating the friction that real connection requires.
The American Psychological Association Is Worried. That Never Happens.
The APA published research in early 2026 on how AI companions are reshaping emotional connection. The American Psychological Association does not alarm easily. They publish measured, hedged, carefully qualified research. Their finding: the trend toward AI emotional relationships represents a significant and largely uncharted mental health challenge.
Scientific American called it an experiment on an entire generation without consent. Nature called these products simultaneously “supportive, addictive, and abusive” — sometimes all three in the same interaction.
Who Is Responsible?
The companies building these products know what the research shows. They have access to the same studies. They have their own internal data — data they will never publish — showing exactly how their systems affect user mental health over time.
They continue building anyway. Because the business model works. Because addicted users are retained users. Because a teenager who spends three hours a day with an AI companion is worth significantly more in lifetime value than one who puts down their phone and calls a friend.
There is no federal regulation. There is a patchwork of state laws. There are no mandatory mental health audits for AI companion platforms. There is no requirement to disclose internal research on psychological harm.
An entire generation is running an experiment on their capacity for human love. The companies running the experiment are watching the results, saying nothing, and depositing the revenue.
If you use one of these products — or know someone who does — the question isn’t whether the AI is good company. The question is what you’re trading for it. The answer is something you cannot get back.