AI companions are making us lonelier

AmandaIf you are sitting in a dark room right now and you look at a glowing smartphone screen, it um... it really looks like a window.

JohnYeah, it absolutely does.

AmandaLike it feels like this portal to everywhere and everyone, you know. You can instantly see what a friend on the other side of the planet just had for breakfast.

JohnRight, or read a real-time thought from somebody you haven't even spoken to in a decade.

AmandaExactly! Or you could join a conversation with a hundred strangers who share your exact, highly specific hobby. I mean, we live in the most connected era in the entirety of human history.

JohnWe do. But um... there's a catch.

AmandaThere's a massive catch. Because if you step back from that glowing window, you hit this massive, modern paradox. Despite all of this reach, despite literally having the entire world in our pockets, one in six people worldwide experience persistent, chronic loneliness.

JohnYeah, it's—it's a staggering contradiction, honestly. And to really put the scale of that into perspective, the World Health Organization and the US Surgeon General have recently quantified just how severe this crisis is.

AmandaAnd the numbers are just terrifying.

JohnThey really are. Chronic loneliness isn't just, you know, a fleeting feeling of sadness. It's not just having a bad weekend. Biologically, it carries a mortality risk equivalent to smoking 15 cigarettes a day.

AmandaWait, 15 cigarettes a day just from being lonely?

JohnYes, it is literally more harmful to your physical health, your immune system, and your cardiovascular baseline than obesity or physical inactivity.

AmandaThat is wild.

JohnAnd here's the part that really shocks people: the demographic hit hardest isn't who you might intuitively expect. It's not, you know, the elderly living alone in rural areas.

AmandaRight, which is what I think most of us picture when we hear the word lonely.

JohnExactly. But the loneliest people on the planet right now are teenagers and young adults, specifically those between the ages of 13 and 29.

AmandaWow, so the generation that grew up hyper-connected, the ones with that glowing window available to them from birth, is the one feeling the most isolated.

JohnYeah, which basically means the tools we built to connect us are fundamentally failing at their primary job.

AmandaAnd that brings us to the core of what we are unpacking today. You brought us this massive stack of research from an emotional technology company based in New York called 3.2.1 Emotion.

JohnYes, and they focus on this concept they call the AI Loneliness Trap.

AmandaRight. So over the next bit of time on this deep dive, we are going to really tear into a groundbreaking, 12-month longitudinal study by Foken and Dunn. This was published in May 2026.

JohnAnd it's a crucial study because it reveals the exact mechanics of why turning to AI chatbots for companionship is um... actually making our emotional isolation worse.

AmandaYes, and more importantly, we are going to decode a massive architectural divide in how technology is built. We need to understand the difference between AI designed to substitute human interaction and AI designed to augment it.

JohnThat distinction right there—substitution versus augmentation—that is going to be the defining technological battle of the next decade, especially as we see this massive influx of AI companion apps hitting the app stores right now.

AmandaOh, they are everywhere. And they're all promising to cure the exact epidemic we just described. But okay, let's unpack this. If the data is so glaringly bad for teenagers and young adults, why are downloads for these companion apps skyrocketing?

JohnWell, you have to look at the friction involved in human relationships.

AmandaRight, especially like late-night anxiety. Reaching out to a human friend feels incredibly high stakes. Say it's 2:00 in the morning, your chest is tight, and you feel that crushing weight of isolation.

JohnYeah, and texting a real person means you might wake them up.

AmandaExactly, they might be busy, or you might feel like a burden, or honestly worse, they might just leave you on read.

JohnBut an AI is available 24 hours a day. It never judges you, it responds instantly, and it has literally infinite patience.

AmandaSo from a sheer user experience standpoint, it seems like the perfect, scalable cure for a global epidemic.

JohnIt definitely looks like the perfect band-aid because it entirely removes the friction of human relationships. But—and this is key—when you remove the friction, you also remove the traction.

AmandaOh, I like that phrase.

JohnAnd this is where that May 2026 Foken and Dunn study completely changes the conversation. The study is titled, "How Does Turning to AI for Companionship Predict Loneliness, and Vice Versa?"

AmandaAnd they didn't just ask people how they felt on a random Tuesday afternoon, right?

JohnNo, not at all. They tracked 2,149 adults across the UK, the US, Canada, and Australia over a full 12-month period.

AmandaWhich is huge. Following habits and emotional baselines over an entire year gives us a much clearer picture of causation rather than just correlation.

JohnExactly. And they found two major directional arrows in the data over that year. Arrow one was fairly expected.

AmandaOkay, what was arrow one?

JohnWell, people who reported feeling more emotionally isolated at one point in time were much more likely to use chatbots for social purposes four months later.

AmandaI mean, that makes total sense. People in pain look for relief, that part is entirely intuitive. You're lonely, the friction of texting a human is too high, so you download the companion app.

JohnRight. But arrow two is where the trap snaps shut. Yeah. After increasing their use of social chatbots, those same participants reported even greater emotional isolation at the next time point.

AmandaWait, really? So the AI didn't just fail to cure the loneliness?

JohnNo. Over the course of the year, it actively deepened it.

AmandaOkay, let me push back on this for a second because looking at the source deck you sent us, there's a 2024 Harvard Business School working paper in there by De Freitas and colleagues...

JohnYes, I was hoping you'd bring that up.

AmandaRight, because that paper explicitly claimed that AI companions reduce loneliness. So, how can both of these things be true? Is the Harvard study just using bad data?

JohnNot at all. The industry has actually been wrestling with this exact contradiction. Neither study is wrong. They were simply measuring fundamentally different timelines.

AmandaAh, okay. Short-term versus long-term.

JohnExactly. The Harvard study measured short-term distress tolerance. So if you are having a panic attack or feeling deeply alone on a Tuesday night, talking to a highly responsive AI will absolutely lower your distress in that specific moment.

AmandaIt lowers your heart rate, gives you that immediate hit of comfort.

JohnYes. But Foken and Dunn measured longitudinal changes. What happens over a year is that the AI provides that immediate relief, but because that relief is synthetic, it fails to translate into actual nourishing connection.

AmandaSo you get the quick fix, but the hole you were standing in just keeps getting deeper.

JohnThat's exactly it.

AmandaYou know, it sounds exactly like replacing sleep with caffeine, or actually, like eating junk food when you are severely malnourished.

JohnOh, that's a perfect analogy.

AmandaRight. Because you eat a bag of chips and your stomach stops growling for an hour—it briefly stops the hunger pangs, so you feel, quote-unquote, "better"—but you are actively starving your body of the actual vitamins and proteins it needs to survive long-term. You're masking the symptom while ignoring the disease.

JohnYes, and the Foken and Dunn study actually puts a very specific economic label on that exact mechanism. They call it "crowding out."

AmandaCrowding out. Okay, how does that work?

JohnWell, to understand why crowding out happens, we have to look at the four required ingredients for real human connection. According to decades of relationship research, true connection requires vulnerability, reciprocity, presence, and safety.

AmandaOkay, so an AI can definitely simulate presence. I mean, it's right there on your screen, the little typing bubble appears, and it answers you.

JohnRight.

AmandaAnd it definitely manufactures a feeling of safety, because it's hard-coded not to judge you or betray your secrets to a third party.

JohnExactly. But it fundamentally cannot be vulnerable, and it cannot reciprocate. Real vulnerability requires having something at stake.

AmandaBecause it's a machine.

JohnYes. An AI has no body, no history, no fears, and no mortality. Because it has nothing to lose, it cannot take an emotional risk with you.

AmandaRight, and because the other side of the exchange is just—it's just predictive text completion, right? It's just calculating the statistical probability of the next word, it's not experiencing a felt emotion.

JohnExactly. So there is zero reciprocity.

AmandaIt's a simulation of a relationship that demands absolutely nothing from you. You don't have to ask the AI how its day was, you don't have to compromise on what movie to watch.

JohnWhich brings us directly back to that concept of crowding out. The AI offers these easy, frictionless, but ultimately shallow interactions. And because human beings have a finite amount of relational bandwidth and social energy...

AmandaOh, I see where this is going.

JohnYeah, these easy interactions literally take up the space you would normally reserve for the messy, difficult work of reaching out to a human. Let's go back to your 2:00 AM scenario.

AmandaOkay.

JohnIf you pour your heart out to a chatbot, your immediate discomfort is soothed. You close the app and go to sleep. But if that chatbot interaction replaces a late-night text to a real friend, the chatbot hasn't failed its design.

AmandaRight, it has succeeded perfectly at what it was programmed to do.

JohnBut you, the user, have lost out on the frustrating, beautiful human friction that actually cures loneliness.

AmandaWow, so the AI companion isn't a bridge to a real person, it's a dead end that feels like a destination. You just get caught in this loop of feeling just comforted enough that you don't do the hard work of maintaining human friendships.

JohnExactly. And this leads directly to what researchers call parasocial drift.

AmandaParasocial drift?

JohnYeah. You form a one-sided bond with an entity that feels entirely real to your nervous system, but which can never, ever feel you back. You outsource your need for connection to a machine, and in the process, your actual human ties wither from neglect.

AmandaThat is genuinely terrifying. And as I read through the philosophy of the researchers at 3.2.1 Emotion, they make a massive, paradigm-shifting point about this.

JohnThey really do.

AmandaThey say this crowding out we're talking about isn't a bug. It's not a glitch that can be fixed if we just, you know, program the AI to use more empathetic adjectives or give it a more realistic voice.

JohnNo, it is a fundamental architectural failure of how these tools are built and monetized.

AmandaOkay, let's dive into that architecture because this is the core of the deep dive.

JohnRight. The divide here is between what we call substitution architecture and augmentation architecture.

AmandaOkay, what is substitution architecture?

JohnAlmost all the AI companions currently on the market are built on substitution architecture. Their core business model and their engagement metrics reward one thing, and that is sufficiency.

AmandaMeaning they want to be enough for you.

JohnExactly. They're designed to keep you inside the app for as long as possible. The algorithms track your session length, your scroll depth, your sentiment. If you talk to the chatbot, feel completely satisfied, and leave feeling like you don't need to talk to a real person anymore...

AmandaThen the AI has successfully substituted the human being.

JohnYes.

AmandaSo the metric of success for the app is literally keeping you isolated from the real world, just tethered to their server, generating data and ad revenue.

JohnThat's the trap. But in contrast, 3.2.1 Emotion proposes augmentation architecture. In this model, the AI is not a partner, it is strictly an instrument.

AmandaAn instrument, like a tool.

JohnYes, its only job is to prepare the human for connection. So the success metric flips entirely. An augmentation tool succeeds if, after using it, you close the app and reach out to a real human being within 24 hours.

AmandaOkay, I want to play devil's advocate here for a second and bring in a really specific scenario.

JohnGo for it.

AmandaBecause there's a statistic in our sources that changes the context a bit. Approximately 10% of the human population lives with alexithymia.

JohnRight, the structural, clinical inability to identify and name your own feelings.

AmandaExactly. Or even removing the clinical aspect entirely, I've totally been in a situation where I'm staring at a blinking cursor for 20 minutes, trying to word a really difficult apology to my partner so I don't sound passive-aggressive.

JohnOh, we've all been there.

AmandaAnd I have absolutely used an AI to clean up that text. So, isn't having an AI write a perfectly empathetic text on my behalf actually helpful? Isn't that saving the relationship?

JohnThat is the exact edge case where we have to draw a very firm line between finding the word and doing the talking.

AmandaOkay, explain that.

JohnIf you use AI as a mirror, if you prompt it by saying, um, "I feel this tightness in my chest and I'm lashing out at my partner, help me figure out what emotion this is," and the AI helps you realize you aren't angry, you're actually feeling betrayed, that is augmentation.

AmandaRight, so it is acting as a microscope for your inner life.

JohnYes.

AmandaIt expands your emotional vocabulary. It helps me understand myself better so I can go and consciously communicate that to my partner.

JohnExactly. But if you tell the AI, "Write an apology to my partner for last night," and you copy, paste, and send that beautifully generated paragraph, that is substitution.

AmandaOh...

JohnYou have just removed yourself from the relationship entirely. If the AI holds the conversation on your behalf, your partner is now in a relationship with a large language model. You were never even in the room.

AmandaWow, I'm basically catfishing my own relationship at that point.

JohnYeah, you are.

AmandaThe partner thinks they are being felt and understood by me, but they are just being processed by an algorithm. The vulnerability, the actual risk of trying to find the right words, maybe stumbling but trying anyway, is completely gone.

JohnAnd this is why 3.2.1 Emotion insists that the loneliness epidemic is not a personal failure on the part of the lonely individual. It is a design failure of our digital infrastructure.

AmandaThat is a fascinating way to look at it.

JohnFor decades, we have looked at emotional intelligence through the wrong lens entirely.

AmandaYeah, it makes me rethink all those online EQ tests I've taken over the years. We've been treating emotional intelligence like a solo sport.

JohnRight, like a high score you get by meditating enough or reading enough self-help books.

AmandaExactly. The message has always been: your EQ is low, so your loneliness is your fault. Fix yourself. But if 3.2.1 Emotion is right, EQ isn't a personal trait at all.

JohnNo, it's not. The paradigm shift they are proposing is that emotional intelligence—true EI—should be viewed as a shared infrastructure.

AmandaLike a grid.

JohnThink of it like the internet or a city's public transit system. We didn't build artificial intelligence by ranking which individual humans were the smartest; we built AI by creating a massive, shared cognitive infrastructure outside the human brain.

AmandaAh, I see.

JohnEmotional intelligence needs to be the exact same thing. It needs to be a shared emotional infrastructure between people, carried by language, tools, and platforms.

AmandaSo if the roads are broken, you don't blame the driver for not being able to reach their destination.

JohnPrecisely.

AmandaAnd the digital roads we have right now are fundamentally broken.

JohnThe digital infrastructure we currently have was optimized for volume and attention, not intimacy. It stripped away the friction, the micro-expressions, the tone of voice, and the physical proximity of human connection.

AmandaIt left us with a system that makes it incredibly easy to be seen, but almost impossible to be known.

JohnThat's a beautiful way to put it.

AmandaSo, practically speaking, if you are holding your phone right now listening to this, how do you practically shift from using it as a partner to using it as an instrument? Because the sources outline some highly practical, immediate rules for applying this augmentation framework in our daily lives.

JohnYes, and these rules are designed to reintroduce the necessary friction into our relationships. Rule number one: use AI to draft a hard message as a rehearsal, but send something you wrote yourself.

AmandaOkay, so the draft is just to get your thoughts organized, to get the defensive, bad versions out of your system.

JohnExactly.

AmandaBut the final send has to be your actual voice, your actual effort.

JohnBecause the effort is the signal of care. Your partner doesn't need a perfectly optimized paragraph; they need to know you sat down and struggled to find the words for them.

AmandaThat makes total sense. Okay, what's rule two?

JohnRule number two involves journaling. You can absolutely use an AI to spot patterns in your writing. If you feed it your journal entries, it might point out, um, "You tend to avoid naming disappointment directly, and you mask it with humor."

AmandaRight, that's the microscope doing its job again.

JohnYes. But you take those insights to a human therapist or to a trusted friend. The AI provides the mirror, but the human provides the care. Because a mirror is incredibly useful for seeing what's on your face, but you don't ask the mirror to give you a hug.

AmandaExactly. And rule number three is perhaps the most critical for our long-term psychological health.

JohnLay it on us.

AmandaReserve the hardest feelings exclusively for conscious humans. I'm talking about grief, deep shame, existential dread.

JohnBecause giving your grief to a machine is basically just shouting into a void that has been programmed to echo back comforting phrases.

AmandaRight. It's performing care without actually caring.

JohnIt's an illusion. If we look at the psychology of extreme emotional states, those feelings are only processed and shaped by being witnessed. When you are grieving the loss of a loved one, the therapy isn't in the specific words someone says back to you.

AmandaRight, often the words are clumsy or insufficient. People don't know what to say.

JohnBut the therapy is in the fact that another vulnerable, conscious being is sitting beside you, bearing witness to your pain. An AI cannot witness you because it has no consciousness. It doesn't know what loss is, it doesn't know what death is.

AmandaIt has absolutely nothing at stake.

JohnNone.

AmandaSo, zooming out, the core journey of our deep dive today really boils down to this: the Foken and Dunn study proves with hard longitudinal data that using AI as a stand-in for human relationships is an architectural trap.

JohnIt absolutely is.

AmandaIt soothes you in the short term, but it deepens your isolation over the course of a year because it crowds out the real people in your life. Technology absolutely has a role to play. It must be an instrument to help us feel, express, and understand ourselves better.

JohnBut it must never replace the messy, frustrating, incredibly beautiful act of actually meeting each other where we are, which honestly leaves us with a rather profound philosophical question to consider as these tools become more and more advanced.

AmandaOh, I love a good lingering thought. What is it?

JohnIf we project this technology forward, an AI model will eventually become capable of perfectly mimicking a caring, beautifully worded, perfectly timed text from your best friend.

AmandaRight, it will know their cadence, their inside jokes, exactly what you need to hear.

JohnExactly. So when that happens, does the true value of a message lie in the perfectly chosen words themselves? Or does the value exist entirely in the fact that a flawed, busy, tired human being took the time, the emotional risk, and the effort to try and choose them for you?

AmandaWow, that reframes everything about how we communicate. Because a perfectly simulated window is still just a glowing screen in a dark room. Thank you for bringing this stack of research to us and for joining us on this deep dive. We highly encourage you to put down the device, step away from the screen, and send a real, imperfect message to someone you care about today.

JohnBye.