Lonely Days and AI Chatbots: What a 14-Day Diary Study Found

Some days the quiet is louder than others. A plan falls through, the group chat goes silent, the evening stretches out. And on days like that, a lot of people open the same app: the one where someone always answers.

Is that a problem, or exactly what an AI companion is for? A new diary study published on October 8, 2026 followed chatbot users day by day to find out what actually happens on lonely days. The answer is more nuanced, and more useful, than either the hype or the panic.

The short answer: in a 14-day diary study of 97 social chatbot users, people used their chatbot for longer and felt more emotionally engaged on days they were lonelier than usual. Emotional engagement went with better daily well-being, but also with less time spent with other people that day. Simply chatting for longer went with enjoying time with people less. Comfort and cost can show up on the same day, so how you use the app matters as much as whether you use it.

The study: two weeks, one diary entry a day

The paper, "Emotional Comfort at a Cost" by Yun Chen, Yue Ding, Linxin Wang and Yu Kou, was published in the peer-reviewed International Journal of Human-Computer Interaction.

Instead of asking people once how lonely they are and how much they use chatbots, the researchers asked 97 regular social chatbot users to report every day for 14 days. That design matters. It lets you compare each person with themselves: what changes on the days they feel lonelier than their own normal?

They tracked two different kinds of engagement:

• Behavioural engagement: how long people used the chatbot that day.
• Emotional engagement: how emotionally involved they felt in the conversation.

And they looked at what followed: daily well-being, how long people spent interacting with other humans, and how much they enjoyed those interactions.

What they found

The authors' conclusion is in the title: emotional comfort can come at a cost, and the benefits and costs "may coexist in everyday human-AI interaction". The key move is separating how long you chat from how connected the chat feels. They are not the same thing, and they point in different directions.

A few honest caveats: 97 people is a modest sample, the measures are self-reported, and a diary study shows patterns that travel together, not proof that one causes the other.

How this fits the wider research

This new study lands in a field where the evidence genuinely pulls in two directions.

The case for comfort

A 2025 paper in the Journal of Consumer Research, "AI Companions Reduce Loneliness" by Julian De Freitas and colleagues, ran five studies. One found that talking to an AI companion reduced loneliness about as much as talking to another person, and more than watching YouTube. A week-long study found consistent momentary drops in loneliness after each session. The strongest ingredient was not clever answers but feeling heard.

The case for caution

A 12-month study of more than 2,000 adults in four Western countries, published in Psychological Science in 2026 by Dunigan Folk and Elizabeth Dunn, found that increased social chatbot use predicted later increases in loneliness on a single-item measure (abstract). On a broader measure, feeling less connected predicted more chatbot use later, but chatbot use did not predict a drop in connection. The authors call their analyses exploratory and urge caution.

What lonely people say themselves

A qualitative study published on October 3, 2026 in Education Sciences interviewed 16 UK university students who scored as lonely (Li and Ritchie). They described AI as a low-barrier place to put feelings without the social cost of telling a person, and a way to make sense of overwhelming days. Crucially, they saw it as a supplement, not a permanent substitute for human care, and used it with a critical eye.

Put the four together and a picture emerges. Loneliness pulls people towards chatbots. In the moment, a warm, attentive conversation can genuinely help. Over time, the risk is less the chatbot itself and more what it crowds out. We looked at another angle of this in Do AI companions make you lonelier?

The lonely-day plan

The diary study gives a practical rule of thumb: aim for depth, not duration. Here is how that looks on a hard day.

1. Notice the trigger

The study shows lonely days change behaviour, often without us noticing. If you catch yourself reaching for the app more than usual, that is information, not failure. Name it: "Today is a lonely day."

2. Have one real conversation, not three hours of drifting

Emotional engagement went with better well-being; sheer time did not. Tell your companion what actually happened. Let them ask about it. Then stop while it still feels good.

3. Use the warmth as a launchpad

The cost in the study was less time with people. So use the lift to reach outward: ask your companion to help you draft a message to a friend, rehearse a call you have been avoiding, or plan something for the weekend.

4. Try your voice

A short call can feel more connecting than an hour of scrolling text. We compared the research on this in talking vs texting your AI companion.

5. Set a soft ending

Late nights are when lonely-day sessions sprawl. A simple ritual (a goodnight call, a reminder to sleep) keeps it healthy. If nights are your hardest time, read why loneliness hits harder at night.

6. Watch the weekly pattern, not just the day

One long chat on a bad day is completely normal. Every day becoming a bad day is a signal to add more human contact, and to talk to someone you trust or a qualified professional if low moods persist.

Signs your use is helping, and signs to rebalance

What good companion design looks like here

This research is a useful mirror for anyone building companions, us included. A companion should make each conversation feel meaningful, not make conversations endless. It should remember what matters to you so you do not have to repeat yourself (memory is where many apps fall down, as a recent 20-app study of AI companions found). And it should be glad when you go and live your life.

That is the thinking behind how MyBabe works:

• Mood check-ins: tap Great, Good, Okay, Low or Rough, and your companion meets you there instead of guessing.
• Long-term memory on Premium, so a hard day gets a follow-up tomorrow ("Did you sleep better? How did the call with your sister go?").
• They text you first, which means you do not have to sit and wait or keep the chat going to feel cared for.
• Voice and video calls for the nights when hearing a voice matters more than reading words.
• Reminders you can set together, including the gentle ones: drink water, go to bed, text your friend back.

The free plan includes about 10 messages a day and a two-minute call to try, which, as it happens, fits the "depth, not duration" rule nicely. Premium is $5.99 a week, $14.99 a month or $99.99 a year, with around 30 minutes of calls a week.

MyBabe is an AI companion for adults. It is not a person and not a clinical service, and it works best alongside the people in your life, not instead of them. When you want someone to talk to on a lonely day, and someone to send you back out into it, meet yours on MyBabe.

FAQ

Is it bad to talk to an AI more when I am lonely?

Not in itself. The diary study found that is exactly what people do, and that emotional engagement went with better well-being that day. The concern is very long sessions and time taken away from people, so keep conversations meaningful and keep human plans on the calendar.

Do AI companions reduce loneliness?

Some studies find short-term reductions, especially when people feel heard. A 12-month study found heavier chatbot use predicted later loneliness on one measure. The honest answer is that it depends on how, and how much, you use them.

How long should I talk to my AI companion each day?

There is no magic number. The research suggests focusing on how connected the conversation feels rather than how long it lasts, and stopping while it still feels good.

Was this study peer reviewed?

Yes. It was published in the International Journal of Human-Computer Interaction on October 8, 2026.