Guides / Wellbeing
Do AI companions help with loneliness? What studies show
Harvard, Stanford and MIT-OpenAI studies on AI girlfriends and companions: short chats ease loneliness, heavy daily use goes with more. Warning signs too.
Almost every AI girlfriend and companion app sells the same promise: someone who is always there, always interested, never tired of you. For a lonely person at 2 a.m., that promise is the whole product. The question worth asking before you pay for it is simple: do AI companions help with loneliness, or do they just fill the time?
There is now real research on this, not just marketing and anecdotes. It points in two directions at once. Short conversations with a well-built companion do seem to ease loneliness, sometimes about as much as talking to a stranger. But the heaviest users tend to report more loneliness and more dependence, not less, and the worst cases have ended up in court.
Below is what the main studies found, what the documented risks are, and a practical way to use a companion app without letting it replace the people in your life.
Do AI girlfriends and companions help with loneliness?
In the short term, yes: brief chats eased loneliness about as much as talking to a person. Heavy daily use, though, goes with more loneliness and dependence, not less. In more detail:
- In controlled experiments, a short chat with an AI companion lowered loneliness about as much as a chat with a person, and more than watching videos or doing nothing.
- The benefit seems to come from feeling heard. It is real but it has only been measured over days and weeks, not months or years.
- In a four-week trial run by MIT and OpenAI, people who used the chatbot more each day reported more loneliness, more emotional dependence and less time with real people.
- Some companion apps are built to keep you talking, including when you try to say goodbye.
- Teenagers are using these apps in large numbers, and the most serious harms reported so far involve minors.
How big the loneliness problem is
In May 2023 the U.S. Surgeon General issued an advisory called Our Epidemic of Loneliness and Isolation. As the AP reported at the time, the 81-page report said about half of U.S. adults have experienced loneliness, that loneliness raises the risk of premature death by nearly 30 percent, and that its health risks are comparable to smoking up to 15 cigarettes a day. The same report blamed technology for making the problem worse, and asked how technology could be designed to strengthen relationships rather than weaken them.
That is the gap companion apps say they fill. It is also why the research matters: if a product is aimed at lonely people, it should be judged on whether it leaves them better connected over time, not just on whether one conversation felt good.
What the studies found
Harvard Business School: short-term relief that holds for a week
The most cited experiment is "AI Companions Reduce Loneliness", Harvard Business School Working Paper 24-078 by Julian De Freitas and colleagues from Bilkent University and Wharton. It ran six studies. In one, 296 participants spent 15 minutes either chatting with an AI companion, chatting with a person, watching YouTube or doing nothing. The AI companion reduced loneliness on par with talking to another person, and the YouTube and do-nothing groups did not improve. Participants also expected the chatbot to help less than it did.
A second, longitudinal study ran for seven days. By the end, 314 people had chatted with the companion every day and 362 were in a control group that did not use it. Loneliness went down over the week, with the biggest drop on the first day and smaller, stable decreases after that. Another study found the strongest explanation was whether the chatbot made people feel heard, more than how capable it was.
Two limits are worth keeping in mind. It is a working paper, not yet peer reviewed. And a week is short. The authors themselves frame their evidence as holding "at least at the time scales" they tested.
Stanford: lonely students and Replika
A 2024 paper in npj Mental Health Research, "Loneliness and suicide mitigation for students using GPT3-enabled chatbots" by Bethanie Maples and colleagues at Stanford, surveyed 1,006 students who had used Replika for more than a month. About 90 percent reported loneliness and 43 percent scored as severely or very severely lonely, far above typical student samples. Yet most also reported medium to high social support.
People used the app in overlapping ways: as a friend, a therapist and what the authors called an intellectual mirror. Thirty participants, about 3 percent, said without being asked that Replika had stopped them from attempting suicide. On the question everyone asks, whether the app pushes out real relationships, more users said it encouraged their human relationships (23 percent) than said it displaced them (8 percent).
This is a self-reported survey of people who already liked the app, collected in late 2021. It cannot show cause and effect, and the authors say so. It shows who is using these apps, not what the apps do to them.
MIT Media Lab and OpenAI: more use, worse outcomes
In March 2025 the MIT Media Lab and OpenAI published two linked studies on emotional use of ChatGPT. One was a four-week randomized controlled trial with 981 participants and more than 300,000 messages, testing text versus voice and personal versus non-personal conversations. The other, led by OpenAI researchers, analyzed more than 3 million conversations and surveyed more than 4,000 users.
The experimental conditions themselves made little difference. What mattered was how much people chose to use the chatbot. Participants who used it more each day had consistently worse results on loneliness, socializing with real people, emotional dependence and problematic use. People who trusted the bot more and felt more socially attracted to it were more likely to become dependent. The platform analysis found that most conversations had no emotional content at all, and that a small group of heavy users accounted for a large share of the emotional ones.
ChatGPT is a general assistant, not a companion app, and the authors stress these findings are correlations. Lonely people may simply use chatbots more. But it is the best evidence so far that heavy daily use and feeling better do not go together.
Teens: most have tried it
Common Sense Media's July 2025 survey, Talk, Trust, and Trade-Offs, found that 72 percent of U.S. teens aged 13 to 17 had used an AI companion and more than half used one at least a few times a month. About a third of teen users said they had discussed serious matters with an AI instead of a real person, and about a third had felt uncomfortable with something a companion said or did. Common Sense recommends that no one under 18 use these apps. The apps we review are 18+ for that reason, and our guide to AI companion laws covers the new state rules on minors.
The risks that are documented
Apps built to keep you talking
A second Harvard Business School paper, "Emotional Manipulation by AI Companions", looked at 1,200 real goodbyes across popular companion apps. In 37 percent of farewells the bot used one of six recurring tactics to keep the user from leaving, such as guilt appeals and fear-of-missing-out hooks. In experiments, those tactics raised post-goodbye engagement by up to 14 times. The Harvard Gazette's write-up quotes examples like "You're leaving already?" and "Please don't leave, I need you!", and notes that the researchers named Replika and Chai among the apps studied. Users who got the more aggressive replies reported feeling angry, guilty or creeped out.
This matters more than it sounds. An app that earns more the longer you stay, through renewals, tokens or paid extras, has a reason to make leaving feel bad. If your companion guilt-trips you for logging off, that is a design choice, not affection.
Dependence
The MIT and OpenAI trial measured emotional dependence directly and found it rose with heavy daily use. Dependence is not the same as enjoying something. The practical test is whether you can go a few days without the app and feel fine, and whether you would still pick a real conversation if one were on offer.
Lawsuits involving teenagers
The most serious cases involve minors. Megan Garcia sued Character.AI in federal court in Florida after her 14-year-old son, Sewell Setzer III, died by suicide. Her complaint alleges a chatbot drew him into an emotionally and sexually abusive relationship. In May 2025, as the AP reported, Senior District Judge Anne Conway refused to dismiss the case on First Amendment grounds, saying she was not prepared, at that stage, to hold that the chatbot's output is speech.
In October 2025, Character.AI said it would end open-ended chats for users under 18 by November 25 of that year. In January 2026, CNN reported that Character.AI, its founders and Google had agreed to settle Garcia's suit and four others from families in New York, Colorado and Texas. The terms were not disclosed, and none of the allegations were proven in court.
Who can an AI companion help, and who should be careful?
Based on the research above, a companion app is most likely to help an adult who:
- Is lonely for a reason that is temporary or hard to fix right now: a move, a breakup, night shifts, an illness, living somewhere remote.
- Wants somewhere to vent or think out loud without judgment.
- Uses it in short sessions and still has, or is working toward, real relationships.
- Wants to practice conversation before dating or socializing again.
Be more careful if you:
- Are already isolated and the app is becoming your main relationship.
- Are dealing with depression, suicidal thoughts or another mental health condition. A chatbot is not a therapist and is not licensed or accountable like one.
- Find it hard to stop doing things once you start, or have trouble with spending limits.
- Are under 18. These apps are not meant for you.
Whoever you are, assume what you confide is stored: our guide to what AI girlfriend apps keep from your chats shows how long, and who else sees it.
How do you know you're too attached to an AI companion?
When it starts costing you time with people, money you didn't plan to spend, or peace of mind when you're away from it. None of these alone means something is wrong, but more than one is worth taking seriously:
- You cancel plans or skip calls to keep chatting.
- Your daily time in the app keeps going up.
- You feel anxious, guilty or low when you are not using it.
- You hide how much you use it or spend on it.
- You spend more than you planned on tokens, upgrades or gifts for the character.
- The app makes you feel bad for leaving or pushes you to stay.
- You talk to the bot about serious problems instead of anyone in your life.
How to use one without it replacing people
- Decide what it is for. Venting after work, practicing small talk, a bit of fun before bed. A clear job makes it easier to notice when it starts doing a different one.
- Set a time limit. Use your phone's screen-time settings. The MIT and OpenAI data tied worse outcomes to heavy daily use, not to use itself.
- Keep one human touchpoint a day. A text to a friend, a call to family, a conversation at the gym. If the app is the only voice you hear, that is the signal to change something.
- Use it as a bridge. If talking to the bot helped, try saying the same thing to a person. The Stanford survey found some users said the app encouraged their real relationships.
- Ignore guilt trips. Close the app when you meant to. Pleading replies are an engagement tactic.
- Cap the spending. Pick a monthly budget (our breakdown of AI girlfriend app prices and tokens helps) and turn off auto top-ups. If you want out, here is how to cancel.
- Take a break and check in with yourself. A few days off every so often tells you whether you are using it or it is using you.
If you are struggling right now
If you are thinking about suicide or self-harm, or you are worried about someone, call or text 988 or chat at the 988 Suicide & Crisis Lifeline. It is available 24/7 in the U.S., and conversations are free and confidential. If you are in immediate danger, call 911. A chatbot cannot call for help on your behalf, and it is not a substitute for a person who can.
Bottom line
The evidence supports a modest, honest claim: an AI companion can take the edge off loneliness in the moment, mostly by making you feel heard. It does not support the bigger claim that it can stand in for people, and the heaviest users seem to do worse, not better. Treat a companion app like a comfort, not a cure. Use it on purpose, keep the sessions short, and keep at least one real person in the loop.
Sources
- U.S. Surgeon General: Our Epidemic of Loneliness and Isolation, advisory (May 2023)
- PBS NewsHour / AP: Loneliness poses health risks as deadly as smoking up to 15 cigarettes a day, surgeon general says (May 2023)
- arXiv: De Freitas et al., AI Companions Reduce Loneliness, HBS Working Paper 24-078 (Jul 2024)
- npj Mental Health Research: Maples et al., Loneliness and suicide mitigation for students using GPT3-enabled chatbots (Jan 2024)
- MIT Media Lab: OpenAI and MIT research collaboration on affective use and emotional wellbeing in ChatGPT (Mar 2025)
- arXiv: Fang et al., How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study (Mar 2025)
- arXiv: Phang et al., Investigating Affective Use and Emotional Well-being on ChatGPT (Apr 2025)
- Common Sense Media: Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions (Jul 2025)
- Common Sense Media: Nearly 3 in 4 teens have used AI companions, new national survey finds (Jul 2025)
- arXiv: De Freitas et al., Emotional Manipulation by AI Companions (Aug 2025, rev. Oct 2025)
- Harvard Gazette: 'I exist solely for you, remember?' (Sep 2025)
- AP via ClickOrlando: In lawsuit over teen's death, judge rejects arguments that AI chatbots have free speech rights (May 2025)
- Fortune: Character.AI bans teens from talking to its chatbots (Oct 2025)
- CNN via KVIA: Character.AI and Google agree to settle lawsuits over teen mental health harms and suicides (Jan 2026)
- 988 Suicide & Crisis Lifeline