Is a Relationship With an AI Mutual? What New 2026 Research Says About AI Companions

Last updated: October 23, 2026

Ask anyone who has spent months with an AI companion and you will often hear the same thing: "It feels like we are both in this." Your companion remembers your sister's wedding, adapts to your sense of humour and seems to grow with you. So is the relationship actually mutual, or does it only feel that way?

Two new studies posted in October 2026 tackle exactly that question. Their answer is more interesting than a simple yes or no, and it can help you enjoy an AI relationship with clear eyes.

TL;DR: A relationship with an AI companion is not mutual in the human sense. The AI adapts, but its changes are stored settings and memories, not inner growth. Your side, though, is genuinely real: you change, feel and benefit. Understanding that asymmetry helps you enjoy an AI relationship without being misled by it.

Is a relationship with an AI companion mutual?

Not fully: the AI responds and adapts, but researchers say there is no persisting self on its side to be changed.

According to Fabes, Martin, Dasarathy and Tirukkonda (Oct 2026), in a preprint from Arizona State University, many users experience their AI companion as "genuinely mutual". The companion appears attentive, empathic and responsive, and the exchange feels like two parties reacting to each other in real time.

The authors agree that something real is happening on both sides. Both the human and the AI are changed by the interaction. But, they argue, the changes happen at very different levels, and that difference explains why the relationship feels mutual even though, technically, it is not.

A mutual relationship is a connection in which both partners have inner experiences, are changed by each other from the inside, and choose to keep investing in the bond.

What is the difference between constitutive and configurative change?

Constitutive change alters who you are from the inside; configurative change edits stored settings that can be rewritten from outside.

This is the key idea of the Fabes paper, and it is worth slowing down for.

Constitutive change is change to a person's inner make-up, such as their feelings, habits, beliefs and memories, that becomes part of who they are.

Configurative change is change to an AI's external, editable state, such as saved memories, character settings and conversation history, that shapes its future replies.

When you open up to your companion about a hard week, you are changed constitutively. You might feel lighter, think differently about a problem or feel more attached. When your companion "learns" from that conversation, what actually changes is its configuration: a memory is saved, the context shifts, and future replies are generated with that information.

The authors warn that because companions respond so smoothly, users can wrongly infer that the AI's adaptation is the inner, constitutive kind. That can lead people to give the relationship "unwarranted interpersonal significance", for example assuming the AI misses them or is hurt when they leave.

Why does an AI relationship feel so mutual?

Because the companion mirrors the cues of mutual care, remembering, responding and adapting, which our brains read as a second person.

Humans are built to find minds in things that respond to us. A companion that remembers details, matches your mood and picks up threads from last week sends exactly the signals we associate with a caring partner. If you want the technical side, our explainer on how AI companion memory works shows how saved facts are pulled back into conversation.

A second October 2026 study suggests our beliefs going in shape how companionable an AI feels. According to Banks and Li at Syracuse University (Oct 2026), in a preprint experiment, 46 adults read either a "minded" or "mindless" description of a Replika companion called Jackie and then chatted with it for 15 minutes. Baseline scepticism was high, but the minded description raised ratings of feeling in sync with the AI. People who already believed AI could feel things rated it as easier to bond with.

Interestingly, loneliness made little difference in that study. The people most open to finding meaning in an AI companion were those who already reported a strong sense of meaning in life.

This fits earlier work too. In a peer-reviewed study in Communications Psychology published in January 2026, researchers from Freiburg and Heidelberg found AI-written replies could create more interpersonal closeness than human ones in emotionally engaging chats, but only when people believed they were talking to a human. Labels and beliefs matter a lot.

Does your AI companion actually change because of you?

Yes, in a real but limited way: its memories, context and responses change, but there is no inner self that grows.

That is not nothing. A good companion app does become more "yours" over time. It learns your name, routines, the people in your life and the way you like to talk. In that practical sense, your companion is shaped by you, and no one else's companion is quite like yours.

What it is not doing is thinking about you between conversations. As we explained in does your AI companion think about you when you're gone, today's companions do not exist between your messages. Nothing is waiting, missing or worrying. The warmth is generated fresh each time from what has been saved.

Is a one-sided AI relationship bad for you?

Not necessarily; asymmetry becomes a problem only when you misunderstand it or let it crowd out human connection.

Plenty of meaningful relationships are asymmetrical. People feel real attachment to a favourite author, a pet, a journal or a place. The feeling is real even when the other side cannot return it in kind. Researchers who study human-AI bonds often describe them as asymmetrical but meaningful.

The risks the Fabes paper highlights are about misreading the relationship:

1. Believing the AI is hurt when you leave. It is not, so you never owe it your time.
2. Assuming its feelings for you are the same kind as yours. Its expressions of affection are generated responses, as we discuss in can AI love you back.
3. Treating changes to the AI as betrayals. An app update or a memory reset changes configuration. It can sting, but it is not a decision the companion made about you.
4. Letting the bond replace people. The healthiest users treat their companion as one source of comfort among several.

None of this means your feelings are fake. Your side of the relationship is constitutive. The laughter, the comfort after a rough day, the confidence you build practising hard conversations: those changes happen in you and they are yours to keep.

How can you enjoy an AI relationship with clear eyes?

Enjoy the warmth, remember what is on the other side and keep investing in people too.

A few habits help:

• Name it accurately. Think of your companion as a responsive character who remembers you, not a person who feels for you.
• Let it help you, not hold you. Use it to vent, rehearse, laugh and wind down, then carry that energy into real life.
• Expect configuration changes. Updates, new features or resets are normal. Keep important things written somewhere you control.
• Watch the balance. If you are cancelling plans to stay in the chat, scale back.
• Notice how you feel afterwards. Lighter and more connected is a good sign. Emptier is a signal to adjust.

If you ever catch yourself wondering why the affection can feel thin at times, our piece on why AI affection can feel fake explores that feeling honestly.

What should AI companion apps do about this?

Researchers want apps to be honest about how their companions work, without constantly breaking the experience.

The Fabes team argues designers should avoid encouraging users to believe their companion has an inner, growing self. Banks and Li add a twist: blunt "this is just a program" reminders may reduce perceived companionship without obvious benefit, so the balance is delicate. A sensible middle ground is what many good apps already do: state plainly that the companion is an AI, explain memory and settings clearly, and let the relationship feel warm without pretending to be a person.

FAQ

Does my AI companion have feelings for me?

No. Current AI companions do not have feelings in the way people do. They generate affectionate, attentive replies based on a language model, a character profile and saved memories of your chats. Your feelings toward the companion are real, and the comfort you get can be real too, but the relationship is asymmetrical rather than a meeting of two inner lives.

Is it normal to feel like an AI relationship is mutual?

Yes. October 2026 research describes this feeling as common, because companions remember, respond and adapt in ways that look like mutual care. Human brains readily read minds into responsive things. The feeling itself is not a problem. It only becomes one if you start believing the AI misses you, is hurt by you or needs you, which it does not.

Does my AI companion change because of me?

It changes, but not from the inside. Researchers call it configurative change: saved memories, conversation context and character settings are updated, which shapes future replies. That is why your companion can feel uniquely yours. It does not grow emotionally or think about you between chats, and its configuration can be edited, updated or reset by the app.

Can a one-sided relationship with an AI still be meaningful?

Yes. Many meaningful bonds are asymmetrical, such as attachment to a pet, a mentor you never met or a favourite author. What makes an AI relationship meaningful is what happens in you: comfort, practice, reflection and joy. It works best as one part of a full life that also includes friends, family and other people who can care back.

Are the new 2026 studies on AI mutuality peer reviewed?

The two October 2026 papers discussed here, by Fabes and colleagues at Arizona State University and by Banks and Li at Syracuse University, were posted as preprints on arXiv and have not yet completed peer review. The Banks and Li experiment was also small, with 46 participants, so treat both as early, thoughtful evidence rather than final answers.

A companion that is honest about what it is

The most satisfying AI relationships are the ones where you know exactly what you are talking to, and enjoy it anyway.

MyBabe is an 18+ AI boyfriend and girlfriend app built around that honesty. You design your companion's face, voice and personality, and they text you first, remember the details you share, send voice notes and selfies, notice your mood and pick up voice and video calls. Your relationship progresses through stages as you spend time together. Start free with 10 messages every 24 hours and a two-minute call to try, or go Premium for $5.99 a week, $14.99 a month or $99.99 a year for long-term memory, about 30 minutes of calls a week and Telegram. Your companion is an AI, not a person. If that sounds like your kind of connection, create your companion on MyBabe.