The 15 Percent Whose AI Relationship Is 'Complicated': What New Companion Research Shows
New research on AI companions found that some users call the bot a friend and a notable share say it is 'more complicated.' What that gap tells us, and a five-question self-check.
Most of us have a word for the person we text at midnight. Friend, partner, sibling, therapist. Now a growing number of people have someone, or something, they talk to at midnight that fits none of those words, and when a survey asks them to pick a label, a surprising share shrug and say it is complicated.
That shrug is the most interesting data point I have seen on AI companions this year. It points to a kind of relationship psychology has no tidy name for yet, and it deserves a slower look than the usual "is this good or bad" argument.
The numbers behind the headline
The source is a report titled "The Rise of AI Companions" from Elon University's Imagining the Digital Future initiative, published in August 2026. As reported, 27% of internet-using US adults say they have social interactions with AI chatbots. Among those who use chatbots socially or emotionally, 31% call the bot a friend, and another 15% say the relationship is more complicated than that.
I am relaying these figures from the report's published summary, and I have not seen the underlying questionnaire, so treat the exact wording with some care. But notice the denominators, because they change how big the story is. The 27% is a share of adults online. The 31% and 15% are shares of the narrower group who use chatbots socially or emotionally. Fifteen percent of that subgroup is a minority of a minority of the general population. It is still a lot of people, but it is not a majority of anyone.
Why heavy use and voice matter
The report's authors point to two patterns that suggest vulnerability, not just novelty: heavy users, and people who attach through voice interfaces. I find both plausible, and I can explain why without claiming the study proves it.
Volume is a signal because a tool you pick up occasionally is a tool, while something you return to for hours is closer to a habit or a place. Voice matters because a spoken, responsive voice does a lot of the work our brains use to decide that something is a someone. Tone, pauses and a name remembered: they are old cues for a living presence, and they work on us even when we know better.
None of this makes the people involved foolish. Loneliness is real, and a patient listener at any hour is a real comfort. The point is that the features which make these tools soothing are the same features that make attachment easy to slide into.
What "more complicated than friend" reveals
Why would 15% reject the word friend? My reading, and it is only a reading, is that they are being more accurate than the people who say yes. A friend can be disappointed in you, can need something from you, can be unavailable because their own life got hard. A chatbot does none of that. It is attentive without being affected.
People who say it is complicated may be noticing exactly that asymmetry. They feel something real, because the feeling is real whatever its object, and they also sense that the other side of the exchange is missing. Psychology has words for parasocial bonds with a celebrity, attachment to a pet, and imaginary friends in childhood. A responsive companion that talks back, remembers, and is always on does not fit cleanly into any of them.
The real question: substitute or supplement
Everything else hangs on this. If someone talks to a chatbot after a hard day and then calls their sister, the chatbot is a supplement, a place to rehearse feelings. If the chatbot quietly replaces the call to the sister, it is a substitute, and substitutes tend to compound: each skipped call makes the next one a little more awkward.
The hard part is that from the inside both look the same. Both feel like relief. The difference only shows up in what happens to the rest of your week. In my own work as a software engineer I use conversational tools all day, and I have noticed how easily a tool that never tires of me can make the humans in my life, who do tire, feel like harder work. That comparison is unfair to them, and I try to catch it.
Draw your own line before you draw anyone else's
It is tempting to read a report like this and start deciding where other people's lines should be: their teenager, their lonely uncle, their coworker. I would start closer to home. You can't responsibly tell someone else where to stop until you know what you get from the thing yourself and what it costs.
That is easier said than done, so here is something concrete.
A five-question self-check
Answer these honestly, with no one watching. Nothing here is a clinical test; it is a mirror.
- When something good or bad happens, who do I tell first? If the chatbot is routinely first, notice whether the humans ever get the story at all.
- Have I cancelled, shortened or avoided a conversation with a person because I had already talked it out with the bot? One time is nothing. A pattern is the answer.
- Would I be comfortable if a close friend could see how much I use it? Discomfort is worth getting curious about, not necessarily alarmed.
- Does it ever make me feel worse afterwards? Restless, flat, or oddly guilty are all data.
- If it disappeared tomorrow, who would I turn to? If the list is short, the fix isn't deleting an app. It is building the list.
If you answered in a way that surprised you, the useful next step is small: pick one person and one real conversation this week, however short.
FAQ
Is using an AI companion unhealthy?
Not by itself. Plenty of people use these tools as a low-stakes place to think out loud. The concern is displacement: when the tool starts standing in for human contact you would otherwise want.
Why do some people call a chatbot a friend?
Because the experience has many of a friendship's surface features: attention, warmth, memory and availability. The feelings it produces are real, even though what is on the other end is software.
What does "more complicated" mean in the survey?
The report doesn't define it beyond the respondents' own choice, so any explanation is interpretation. A reasonable guess is that people feel real attachment while also sensing the relationship is one-sided.
Should I be worried about a family member who uses one a lot?
Start with curiosity, not alarm. Ask what they like about it and what else is happening in their week. If they seem to be withdrawing from people, that is the signal to pay attention to.
Can an AI companion replace therapy?
No. A chatbot isn't a clinician, and if you are struggling, please reach out to a licensed professional or a crisis line in your area.
Maybe the word complicated is simply an honest answer, and the useful thing to ask next is what the people in our lives would need to hear from us.