AI should bring people together, not replace them

Open almost any app this year and something will offer to keep you company. A chatbot that remembers your day. A companion that always replies, never gets tired and never has a bad week of its own. If you’re lonely, or frightened about work, or just worn out by people, that offer can feel like relief. There’s no shame in taking it.

It isn’t a niche idea any more. In July 2025, Common Sense Media reported that 72% of US teenagers had tried an AI companion. That April, Meta’s chief executive, Mark Zuckerberg, suggested on a podcast that AI could help fill the gap many people feel in their friendships, while saying it probably wouldn’t replace in-person relationships.

We went the other way. At Everconnected, technology’s job is to help people find each other, then get out of the way. Here’s our thinking on what AI should be for when it comes to people.

Why AI companions appeal, and why that’s okay

Let’s be fair to the other side, because the evidence deserves it. In a series of studies, Julian De Freitas and colleagues at Harvard Business School found that a short chat with an AI companion eased loneliness in the moment about as much as talking to another person, and more than watching YouTube. The key ingredient was feeling heard. If a chatbot helped you through a bad night, that comfort was real.

The longer picture is less reassuring. In early results from a four-week trial with 981 people, published in March 2025, researchers at the MIT Media Lab and OpenAI found that heavier daily chatbot use went along with more loneliness, more emotional dependence on it and less socialising with people. That’s a correlation: lonely people may simply reach for it more. But it suggests that something always available isn’t the same as a life with more people in it. We look at this more closely in AI can’t replace friends.

What can’t an AI friend give you?

Here’s where the evidence ends and our view begins.

Friendship isn’t only being listened to. It’s being needed back. It’s someone with their own bad days, their own opinions and their own life, who could have said no and chose you anyway. When a friend makes time for you, it costs them something, and that cost is part of what makes it matter. A system designed to please you can’t offer that, however warm it sounds.

Friendship also takes time, and the time is part of it. In a 2018 study, Jeffrey Hall at the University of Kansas estimated that it takes about 50 hours together to go from acquaintance to casual friend, around 90 hours to become friends and more than 200 hours to become close friends. They’re rough, survey-based estimates. Still, the point stands: there’s no shortcut, and nobody can put in those hours for you.

So what should AI be for?

Our answer: introductions. Think of the friend who says, “You two should meet. You’d get on.” That’s a genuinely useful thing to do, and it’s the part of connection where sorting through lots of possibilities can help. What comes after (the talking, the awkward pauses, the slow build of trust) belongs to people.

That’s how we’ve built Everconnected. In the background, we match you with someone who shares your values and interests. Everything after that is the two of you.

Does matching people on shared values actually help?

There’s good evidence that similarity draws people together. In 2008, psychologists Montoya, Horton and Kirchner pooled 313 studies and found that both actual similarity and perceived similarity were strongly linked to attraction, in the psychologist’s sense of liking someone. Notice the second one: it isn’t only what’s on paper. Feeling that you have things in common matters too.

But matching only gets you into the room. In Arthur Aron’s well-known 1997 study, pairs who had never met took turns asking each other 36 increasingly personal questions, and ended up feeling closer than pairs who asked 36 small-talk ones. Aron’s team reported that this worked even for pairs who didn’t share important attitudes. Our reading: shared values can make it easier to start, but it’s the conversation that does the work.

Why voice, and why no photos?

Many apps that connect people run on text and pictures. We chose voice, partly because of research like this. In 2021, Amit Kumar and Nicholas Epley found that people expected a phone call with an old friend to feel more awkward than an email. When they were randomly assigned to actually make the call, they felt a stronger bond, and it wasn’t more awkward. In a separate experiment, people felt more connected when they talked than when they typed.

No photos is a choice, not a research finding. Our hunch is that when you meet someone as a voice and a set of answers, you listen differently.

Try this: use AI without letting it crowd out people

1. Notice why you’re reaching for it. If you open a chatbot because you’re lonely, that isn’t a failure. It’s information. Let it nudge you to message one person too.
2. Let tech find the door, then walk through it yourself. Use tools to find a local group, a class or a volunteering slot. Then turn up in person.
3. Call instead of typing, once this week. It will probably feel less awkward than you expect.
4. Ask a real question. Swap “How’s work?” for something like “If you could relive one ordinary day of your life, which day would it be?”
5. Give it hours. If friendship takes around 50 hours to get going, a weekly walk or call adds up faster than you’d think.

What we built, and what we won’t do

Everconnected is a place to meet like-minded people, find community and have genuine conversations, one to one, by voice. It isn’t a crisis or mental health service. It’s free to join: create an account at app.everconnected.com, or get the app from the App Store or Google Play.

We won’t pretend an app can fix loneliness on its own. It can make an introduction. The listening, the laughing and the slow trust are yours.

Questions people ask

Sources

1. Common Sense Media (2025, 16 July). Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions. Nationally representative survey of 1,060 US teens aged 13-17. www.commonsensemedia.org/research/talk-trust-and-trade-offs-
2. Dwarkesh Patel (2025, 29 April). Mark Zuckerberg interview, transcript: the section on AI friends. www.dwarkesh.com/p/mark-zuckerberg-2
3. De Freitas, J., Uguralp, A. K., Uguralp, Z. O., & Puntoni, S. (2024). AI companions reduce loneliness. arXiv:2407.19096 (working paper; later published in the Journal of Consumer Research). arxiv.org/abs/2407.19096
4. Fang, C. M., et al. (2025, March). How AI and human behaviors shape psychosocial effects of chatbot use: A longitudinal randomized controlled study (version 1). MIT Media Lab and OpenAI. arXiv:2503.17473v1. arxiv.org/abs/2503.17473v1
5. Hall, J. A. (2019). How many hours does it take to make a friend? Journal of Social and Personal Relationships (online 2018). University of Kansas news release, 6 March 2018. news.ku.edu/2018/03/06/study-reveals-number-hours-it-takes-m
6. Montoya, R. M., Horton, R. S., & Kirchner, J. (2008). Is actual similarity necessary for attraction? A meta-analysis of actual and perceived similarity. Journal of Social and Personal Relationships, 25(6), 889-922. researchportal.murdoch.edu.au/esploro/outputs/journalArticle
7. Aron, A., Melinat, E., Aron, E. N., Vallone, R. D., & Bator, R. J. (1997). The experimental generation of interpersonal closeness: A procedure and some preliminary findings. Personality and Social Psychology Bulletin, 23(4), 363-377. doi.org/10.1177/0146167297234003
8. Kumar, A., & Epley, N. (2021). It’s surprisingly nice to hear you: Misunderstanding the impact of communication media can lead to suboptimal choices of how to connect with others. Journal of Experimental Psychology: General, 150(3), 595-607. doi.org/10.1037/xge0000962
9. University of Texas at Austin. Phone calls create stronger bonds than text-based communications (news release on Kumar & Epley). news.utexas.edu/phone-calls-create-stronger-bonds-than-text-