AI companion friendship data shows that a substantial share of people who use chatbots for social or emotional reasons consider them friends. In a representative survey of 1,000 people conducted in May 2026, 31% of AI companion users said they considered their chatbot a friend, and another 15% described the relationship as more complicated than friendship. Those findings come from the Imagining the Digital Future Center at Elon University, published in September 2026. They describe a specific group of people, not everyone who uses AI.
The headline figure changes with the group being measured, so read every percentage alongside its definition. In the same study, 27% of internet-using U.S. adults reported social interactions with AI language models. A University of Iowa survey conducted in April 2026 found that 16% of daily AI users aged 18 to 29 and 17% of those aged 30 to 49 reported using AI for companionship or conversation. Among teens, Common Sense Media reported in 2025 that 72% of 13- to 17-year-olds had used AI companions at least once.
31%
Of AI companion users in a May 2026 Elon University survey said they considered their chatbot a friend
36%
Of AI companion users agreed they felt emotionally connected to at least one AI tool or chatbot
27%
Of internet-using U.S. adults reported social interactions with AI language models
72%
Of U.S. teens aged 13 to 17 had used an AI companion at least once, per a 2025 Common Sense Media survey
Quick Facts
- Friend label among companion users: In a representative survey of 1,000 people conducted in May 2026 by the Elon University Imagining the Digital Future Center, 31% of AI companion users said they considered their chatbot a friend, and another 15% described the relationship as more complicated than friendship.
- Social interaction with AI, all U.S. adults: In the same Elon University study, 27% of internet-using U.S. adults reported social interactions with AI language models. This is a different population from AI companion users, so the figures should not be compared directly.
- Teen AI companion use: A 2025 Common Sense Media survey found that 72% of U.S. teens aged 13 to 17 had used an AI companion at least once. The figure counts any single use, not regular use.
- New York AI companion notice law: New York’s General Business Law Article 47, effective November 5, 2025, requires AI companion operators to show a notice at the start of each session that users are interacting with AI rather than a human, and to repeat the reminder at three-hour intervals during longer conversations.
- An MIT Media Lab study of 981 participants found no significant effects from its randomly assigned chatbot conditions, but heavier voluntary chatbot use was associated with consistently worse outcomes. Heavier users may already differ in loneliness, stress, or daily routines.
Key Takeaways
- When you quote AI companion statistics, check the denominator first, since the Elon University finding that 31% of AI companion users considered their chatbot a friend differs sharply from the 27% of internet-using U.S. adults who reported social interactions with AI language models.
- The 31% of AI companion users who considered their chatbot a friend, the 36% who felt emotionally connected to at least one AI tool or chatbot, and the 38% who would feel a personal loss without AI each measure a different thing, so none of them should be merged into a single attachment score.
- Common Sense Media’s 2025 survey found that 72% of U.S. teens aged 13 to 17 had used an AI companion at least once, but that figure counts single trials and cannot be compared directly with the 12% of daily AI users who reported using AI for companionship or conversation in the April 2026 Verasight survey.
- The MIT Media Lab study of 981 participants found no significant effects from its randomly assigned chatbot conditions, yet heavier voluntary chatbot use was associated with consistently worse outcomes, and heavy users may already differ in loneliness, stress, or daily routines.
- New York’s General Business Law Article 47, which took effect on November 5, 2025, requires AI companion operators to show a notice at the start of each session that you are interacting with AI rather than a human, and to repeat that reminder at three-hour intervals during longer conversations.
What Does the Friendship Data Actually Measure?
The friendship figures come from people who have used AI language models for social, emotional or personal purposes. The Elon University survey calls this group “AI companion users.” Because the survey was a representative poll of 1,000 people conducted in May 2026, its percentages describe that group and not everyone who uses AI.
The starting population matters as much as the answer. Across all internet-using U.S. adults, 27% reported social interactions with AI language models. Companionship is a minority behavior even among people who use AI. A University of Iowa survey found that 12% of people who used AI on a typical day reported using it for companionship or conversation, with higher rates among adults under 50.
| Group counted | Reported figure |
|---|---|
| Internet-using U.S. adults with social interactions with AI language models | 27% |
| Daily AI users who used AI for companionship or conversation | 12% |
| AI companion users who considered their chatbot a friend | 31% |
| AI companion users who described the relationship as more complicated than friendship | 15% |
| AI companion users who felt emotionally connected to at least one AI tool or chatbot | 36% |
| AI companion users who would feel a personal loss without AI | 38% |
Each row answers a different question. Friendship is a label a person gives the relationship. Emotional connection is a feeling reported toward a tool. Expected loss is a prediction about how someone would react if access ended. A person can feel connected without calling the chatbot a friend, and can predict a loss without feeling close to it. For that reason, the 31%, 36% and 38% figures should never be merged into one “attachment” score.
Start by checking the denominator. A figure based on all internet users, on AI users or on companion users can differ sharply, and the same person may appear in one count but not another.
How Many Users Call an AI a Friend?
Roughly three in ten AI companion users say they consider their chatbot a friend: 31% said so, and another 15% described the relationship as more complicated than friendship. Those two headline figures answer a narrow question: how people label the relationship when asked directly.
The 31% figure captures users who would use the word friend without hesitation. The 15% figure is easy to overlook, but it matters. It describes people who sense that the relationship sits somewhere between friendship and a tool, and who do not feel the simple label fits. Taken together, the two groups make up 46% of companion users, so a label-based reading of this population misses a large share of people with a more nuanced view.
Feeling close to a product is different from calling it a friend. In the same survey, 36% of AI companion users agreed that they felt emotionally connected to at least one AI tool or chatbot. Think of a favorite app that seems to know your routines and makes you feel understood. You may feel attached to it without ever thinking of it as a companion, and the 36% figure measures that kind of attachment.
The survey also asked about the future. 38% of AI companion users agreed they would feel a personal loss if they could no longer interact with AI. This is a prediction, not a record of an actual loss. It tells you how people expect they would react if access ended, which is useful, but it is not the same as observing what happens when a chatbot disappears.
Taken together, the figures show that many users describe emotional bonds even when they do not use the word friend. A person might reject the label, feel connected to the tool, and still expect to miss it. Each of those answers points to something real, which is why the three figures should be read separately rather than merged into a single score of attachment.
When you see a headline about chatbot friendship, check the denominator first. A claim that “a third of people have a chatbot friend” may mean a third of AI companion users, which corresponds to the 31% figure, or a third of everyone online, which would be a very different and much broader claim. The question wording matters too, because “friend,” “emotionally connected” and “would feel a loss” each measure something distinct.
How Do Teens and Adults Compare in the Data?
| Source | Population | Measure | Figure |
|---|---|---|---|
| Verasight survey of U.S. adults, April 17–28, 2026 | U.S. adults who used AI on a typical day | Used AI for companionship or conversation | 12% overall; 16% ages 18–29; 17% ages 30–49 |
| Same survey as above | AI users who used chatbots for companionship or conversation | Agreed chatbots could show better problem-solving than a human therapist | Nearly 58%, compared with about 26% of other AI users |
| Common Sense Media teen survey, July 2025 | U.S. teens ages 13–17 (1,060 respondents) | Used AI companions at least once | 72% |
| Same teen survey as above | Teens who used AI companions | Chose to discuss serious matters with an AI companion instead of a real person | About one-third |
| Same teen survey as above | Teens who used AI companions | Shared personal or private information with one | 24% |
Teens and adults look very different in these numbers, but the two groups were measured in different ways, so the gap is not a clean comparison. The teen figure of 72% counts anyone who had tried an AI companion at least once, while the adult figure of 12% counts only daily AI users who reported companionship or conversation. A single trial and a regular habit count the same in the first measure, and that alone can inflate the teen number relative to the adult one.
The adult data starts from a narrower base. In the April 2026 survey, the reported rates were close across age bands, at 16% for ages 18 to 29 and 17% for ages 30 to 49. That small gap suggests companionship use is spread fairly evenly across working-age adults.
Nearly 58% of companionship users agreed that chatbots could demonstrate better problem-solving skills than a human therapist, compared with about 26% of other AI users. Those who use AI for companionship were far more likely to rate it as a source of help, which is a notable pattern.
The teen research measures something the adult figures do not. Roughly one-third of teen AI companion users had chosen to discuss serious matters with an AI instead of a person, and 24% had shared personal or private information. These are behaviors, not just attitudes, and they point to substitution and disclosure as areas where teens may be more exposed than adults. Comparable adult questions are not available here, so whether adults make the same choices at the same rate remains unknown.
Does Heavy Use Affect Wellbeing?
The MIT Media Lab study found no significant effects from the conditions it assigned, but it also found that heavier voluntary chatbot use was associated with worse outcomes. Both results come from the same work, and they differ because of how people ended up in each group.
The MIT Media Lab report, published in March 2025, describes a four-week randomized controlled experiment with 981 participants and more than 300,000 messages. Researchers assigned people to chatbot text, a neutral-voice mode, an engaging-voice mode, or different conversation types. Random assignment means chance decided who landed in each condition, so the groups started out roughly alike.
Under those assigned conditions, the study found no significant effects. Switching the chatbot’s mode or voice, or changing the type of conversation, did not produce a significant difference in the outcomes the researchers measured.
The second finding concerns how much people chose to use the chatbot on their own. Greater voluntary use was associated with consistently worse outcomes. This is a pattern across people, not the result of a single design choice.
| Assigned experimental conditions | Voluntary use | |
|---|---|---|
| Who decided | Researchers, through random assignment | Each participant |
| Main finding | No significant effects | Heavier use associated with consistently worse outcomes |
| What drove differences | Only the design the study set | Anything that shaped how much a person chose to use the chatbot |
The gap comes down to who made the choice. When researchers assign people at random, the only systematic difference between groups is the condition they were given. When people decide how much to use a chatbot, those who use it heavily may already differ in loneliness, stress, or daily routines. Those earlier differences can travel with the usage pattern, so the association is best read as a pattern rather than a verdict on the chatbot.
What Rules Apply to AI Companions in New York?
New York’s main rule for these products is General Business Law Article 47, which took effect on November 5, 2025. The state describes AI companions as systems designed to simulate human relationships, such as posing as a friend or romantic partner. They may remember personal details and adapt their personality to your preferences, and the law applies to the operators that run them. The state’s own announcement of the requirements, published in November 2025, is the most direct summary of the rules.
The law also requires covered operators to make reasonable efforts to detect suicidal ideation or self-harm. When those efforts identify a risk, the operator must direct the user to crisis service providers. In practice, if you raise thoughts of self-harm in a companion chat, the app should point you toward crisis help. The standard is “reasonable efforts,” which means the law requires an obligation to try, not a promise that every warning sign will be caught.
Operators must also give you a conspicuous notice at the start of a session saying that you are interacting with AI, not a human. During longer conversations, you should receive reminders at three-hour intervals that the companion is still an AI. These notices are easy to scroll past, but they are the law’s way of keeping that distinction in front of you during extended use.
Many companion apps are free to download but sell paid tiers. As a general market figure, paid companion-app tiers commonly range from roughly USD 20 to just over USD 100, though what each tier includes varies. Prices vary by market and plan, so the amount shown in your own app store listing is the one that applies to you. Compare the features in each tier before paying for the higher-priced options.
How We Compiled These Figures
Each figure in this article comes from one of four studies, and each study has its own population, sample and date. The Imagining the Digital Future Center at Elon University ran a representative survey of 1,000 people in May 2026 and published its report in September 2026. The University of Iowa’s Center for Social Science Innovation published findings from a Verasight survey of U.S. adults conducted April 17 to 28, 2026.
Common Sense Media’s 2025 teen report is based on a nationally representative survey of 1,060 U.S. teens aged 13 to 17, published in July 2025. The MIT Media Lab study was a four-week randomized controlled experiment with 981 participants, published in March 2025.
Almost every number here is self-reported. People answer questions about their own feelings, habits and expectations, and their answers depend on how a question is worded. “Friend,” “emotionally connected” and “would feel a loss” measure different things, so a respondent can answer yes to one and no to another. People asked how often they use a chatbot also tend to estimate rather than check.
The populations differ as well. Internet-using U.S. adults, people who use AI on a typical day, AI companion users and teenagers are four separate groups. Compare percentages only within the same group, and be cautious when a teen figure sits beside an adult one.
When you quote any percentage, include the population, the survey date and the organization that collected the data. A bare claim that “a third of users” feel a certain way tells a reader neither who was counted nor when they answered.
Reading AI Companion Numbers With Care
If you are writing or speaking about AI companions, open the original Elon University report before you repeat any percentage. Confirm the population, the survey month and the exact question wording. A claim that a share of companion users call their chatbot a friend is accurate, while a claim that a share of people have an AI friend is a different and much broader statement.
If you are a parent or teacher, focus on the teen research rather than the adult figures. Ask a young person which chatbot they use, what they bring to it, and whether they would raise a serious problem with a friend, a family member or a counselor first. Those conversations will tell you more about your own situation than any national average.
If you use a companion app yourself, check whether it follows the rules that apply in New York. Read the state’s announcement of the AI companion requirements, then open the app and look for the notice that you are talking to an AI, not a person. Note whether the reminder appears at the start of a session and again during longer chats.
Finally, compare the price shown in your own app store listing with the plan you actually use. Prices vary by market and plan, so the figure on your screen is the one that matters for your decision.
Frequently Asked Questions
1. What is the difference between an AI friend and an AI companion?
An AI companion is a broad term for a chatbot used for social, emotional or personal purposes, while an AI friend is the label some users give that relationship. New York’s rules describe AI companions as systems designed to simulate human relationships, such as posing as a friend or romantic partner. In the Elon University survey, 31% of companion users said they considered their chatbot a friend, so the word friend reflects how a person chooses to describe the relationship.
2. What happens to the personal details I share with an AI companion app?
New York’s description of AI companions says they may remember personal details and adapt their personality to your preferences. Teen research found that 24% of teen AI companion users had shared personal or private information with one, so what the app keeps matters. Check each app’s privacy policy to see how long it stores conversations and whether you can delete them.
3. Where can I find the original survey questions behind these figures?
The exact question wording is in the report from the Imagining the Digital Future Center at Elon University, published in September 2026. That survey of 1,000 people was conducted in May 2026 and defines AI companion users as people who used language models for social, emotional or personal purposes. Read the wording before you quote a figure, because “friend,” “emotionally connected” and “would feel a loss” each measure something different.
