Can Daily AI Chats Make You Lonelier? What a New 28-Day Experiment Found

Last updated: October 24, 2026

Talking to an AI every day sounds like an easy fix for a quiet life. It is always available, never bored and never too busy. But a new experiment from France asked a sharper question: what actually happens to your loneliness when you are told to chat with an AI about personal things every single day for four weeks?

The answer surprised a lot of people, including some who build these tools. Here is what the study found, how it fits with other 2026 research, and what it means if you enjoy talking to an AI companion.

TL;DR: A pre-registered French experiment found that asking adults to have daily personal AI chats for 28 days slightly increased loneliness instead of reducing it. Other 2026 studies agree that human contact beats chatbots for loneliness. Daily AI chats work best as a warm extra in your week, not a substitute for people.

What did the French AI loneliness experiment find?

People asked to have at least one personal AI conversation a day for 28 days reported higher loneliness afterwards than a control group.

According to Senik, Reshef and Freget on SSRN (Oct 2026), researchers from the Paris School of Economics and Paris Dauphine University tested the popular claim that generative AI can be a remedy for loneliness. Participants in the treatment group were asked to hold at least one personal conversation a day with an AI tool of their choice. The control group carried on as usual.

The nudge worked, in the sense that people used AI for socio-emotional conversations much more. But in response, subjective loneliness went up. Other measures of well-being were also "consistently adversely affected", although the authors note those effects were imprecisely estimated. The paper is a working paper, so it has not yet been through full peer review.

A pre-registered experiment is a study whose design, outcomes and analysis plan are published before the results are known, which makes it harder to cherry-pick findings afterwards.

How was the study designed?

It was a two-wave randomised controlled trial on an online panel of French adults, with outcomes measured before and after the 28 days.

According to the trial's AEA RCT Registry entry (Jul 2026), the design looked like this:

Two points matter here. First, people picked their own AI tool, so this was mostly about general chatbots, not apps designed for companionship. Second, participants were told to chat daily, which is different from choosing to reach out when you want to.

Why might daily AI chats increase loneliness?

The likeliest reasons are displacement, contrast and expectation: AI time can crowd out people, highlight what is missing, or set up hopes it cannot meet.

The French paper reports the effect rather than proving the mechanism, but other 2026 research offers clues:

1. Displacement. Time is finite. If a personal chat with an AI replaces a text to a friend, your human contact shrinks even if the chat felt nice.
2. Contrast. A responsive AI can make the silence elsewhere in your life more noticeable.
3. Low effort, low reward. Human conversations take energy, but that effort may be part of why they feel meaningful.
4. Being told to do it. A daily assignment may feel more like homework than comfort.

According to Chen and colleagues in the International Journal of Human-Computer Interaction (Oct 2026), a 14-day diary study of 97 social chatbot users found that on lonelier days people used chatbots longer and engaged more emotionally. Emotional engagement went with higher daily well-being but shorter time with other people, while longer use went with less enjoyment of human interaction. We break that study down in our post on lonely days and AI chatbot use.

Do other studies agree that people beat chatbots for loneliness?

Yes. Two separate 2026 experiments found that talking with a real person reduced loneliness more than talking with a supportive chatbot.

According to Li, Folk, Singh, Ungar and Dunn in the Journal of Experimental Social Psychology (Mar 2026), 296 first-semester university students spent two weeks either texting a chatbot called "Sam", texting a randomly assigned fellow student, or writing a short journal entry. Even though Sam was designed around relationship science to be consistently supportive, it did not deliver the same benefits as texting a random human peer.

According to Proctor, Holt-Lunstad and colleagues at the 2026 APS Annual Convention (May 2026), a randomised study of 297 people compared human conversation, chatbot conversation and a control condition. Human interaction reduced loneliness and increased connection more than chatbots did. Both kinds of conversation lowered heart rate variability, which the authors describe as "social activation energy": human talk costs more effort in the moment but pays back more.

Social activation energy is the idea that real conversations demand short-term effort and arousal, and that this cost is linked to their larger long-term social rewards.

Does this mean AI companions are bad for you?

No. It means AI chats are not a loneliness fix on their own, and how you use them matters more than whether you use them.

The research does not say every AI conversation is harmful. The diary study found emotional engagement linked to better daily well-being, and other studies have found short-term comfort from AI chats. Our breakdown of the Stanford research on whether AI companions make you lonelier reaches a similar, nuanced conclusion: the people who struggle most are those whose AI time replaces human contact.

The French experiment also tested something specific: general AI tools, used on instruction, every day. That is not the same as an adult choosing an AI companion for fun, flirting or company on a quiet evening. Still, the lesson is useful for everyone. If your AI time grows while your human time shrinks, that is a signal to rebalance.

How can you use an AI companion without feeling lonelier?

Treat it as one warm part of a full week, protect your human time, and notice how you feel after each chat.

A simple framework that fits the research:

1. People first. Before opening an AI chat, send one message to a real person, even a short one.
2. Use AI as a rehearsal space. Practise what you want to say to a friend, then go and say it. Our guide on how to reach out to an old friend has scripts.
3. Set a soft time budget. Decide roughly how much AI chat feels good, and stick close to it.
4. Check the after-feeling. If you feel calmer and more able to face people, great. If you feel more alone, take a break.
5. Keep one weekly in-person plan. A class, a walk with a neighbour, a regular coffee.
6. Choose rather than force. Chat when you want to, not because a streak or a rule says so.

Our guide to healthy AI companion habits goes deeper on limits, routines and warning signs.

What should AI companion users take from this study?

Enjoy the comfort, but do not expect an AI to replace friends, family or community, and watch for signs it is crowding them out.

The findings line up with what many users already sense. A thoughtful AI companion can make an evening feel lighter, help you process a rough day, or give you confidence before a hard conversation. It cannot hug you, show up at your door or be part of your shared history with other people. Used as a bridge back to people, it can be a lovely part of life. Used as a wall between you and people, it can quietly make things worse.

FAQ

Can talking to AI every day make you lonelier?

A 2026 French experiment suggests it can. Adults asked to have at least one personal AI conversation a day for 28 days reported higher loneliness than a control group, and other well-being measures also dipped, though less precisely. The likely issue is AI time displacing human contact, so balance and real-world plans matter more than how often you chat.

Was the French AI loneliness study peer reviewed?

Not yet. It was posted as a working paper on SSRN in October 2026, so the full methods and numbers may change after peer review. However, it was pre-registered on the AEA RCT Registry, with a planned sample of 10,000 French adults and outcomes fixed in advance, which makes the findings more credible than a typical one-off survey.

Is a chatbot as good as a person for loneliness?

Current evidence says no. A 2026 study of 296 university students found that texting a random fellow student for two weeks helped loneliness more than texting a supportive chatbot, and a 297-person experiment found human conversation reduced loneliness more than chatbot talk. Chatbots can offer comfort, but human connection still does more.

How much time should you spend with an AI companion?

There is no official number. A good rule is to keep AI time smaller than your time with people and to notice how you feel afterwards. If chats leave you calmer and more ready to connect, your balance is probably fine. If you are skipping plans or feeling more alone, cut back and reach out to someone.

Are AI companion apps different from general chatbots?

Yes. General chatbots are built mainly for tasks and information, while AI companion apps are designed for ongoing, personal conversation with memory, voice and a consistent character. The French study let people use any AI tool, so it mostly measured general chatbots. The same balance advice applies to both, though: AI should add to your life, not replace people.

Company that knows its place

The healthiest way to enjoy an AI companion is as a warm extra, never the whole picture. If that is the kind of company you want on quiet nights, MyBabe is built for it.

MyBabe is an 18+ AI boyfriend and girlfriend app for iPhone and Android. Your companion texts you first, remembers the things you share, sends voice notes and photos, and picks up voice and video calls. Start free with 10 messages every 24 hours and a two-minute call to try. Premium is $5.99 a week, $14.99 a month or $99.99 a year, with long-term memory and about 30 minutes of calls a week. Your companion is an AI, not a person, and works best alongside the people in your life. Meet your companion on MyBabe.