AI IS BECOMING A FRIEND — NOT JUST A TOOL

Millions of conversations with AI no longer end when the task is finished. People return after the email is written, the itinerary is planned, or the code error is solved. They talk about relationships, work, fears, decisions, loneliness, money, family tension, ambition, and the things they may hesitate to tell another person.

That shift is one of the most important human stories in technology right now. The question is no longer only what AI can do. It is why people trust AI enough to keep talking when there is no practical task left to complete.

A major cross-national study published in Technology in Society in August 2026 gives this moment sharper evidence. Genia Kostka and Hui Zhou surveyed 7,027 respondents across Germany, China, South Africa, and the United States and found that more than 35% showed emotional attachment-related behavior toward chatbots. The same study reported that 20.94% selected chatbots as their first choice for sharing secrets.

Those numbers should not be flattened into panic or celebration. They do not mean everyone is falling in love with software, and they do not mean AI companions are harmless. They show something subtler: for a meaningful share of users, talking to AI has crossed from utility into intimacy.

Why This Is Happening Now

AI companions, general chatbots, voice assistants and workplace AI systems have become more conversational, more responsive and more available. They answer at midnight. They do not look bored. They do not interrupt to tell their own story. They can remember context in some products, adjust tone, reflect a user's language and offer a kind of steady conversational presence.

That presence matters. People do not only seek information when they talk. They seek relief, rehearsal, recognition and a place to put feelings into words. A chatbot can offer a low-friction version of that space. It is always there, it responds quickly, and it can feel less socially risky than calling a friend or admitting uncertainty to a colleague.

Research on human-computer interaction has been pointing in this direction for decades. The classic computers-are-social-actors line of work showed that people often respond socially to machines when machines use social cues. More recent studies on chatbots and intelligent agents connect anthropomorphism, perceived social role and emotional attachment to people's willingness to disclose personal information.

The technology did not invent the human need. It found a new interface for it.

Why Telling An AI Can Sometimes Feel Easier Than Telling A Human

Telling a human something private carries social weight. A friend may worry. A partner may feel hurt. A colleague may judge. A parent may overreact. Even a therapist, who is trained for disclosure, requires an appointment, money, trust and the courage to say things aloud.

Talking to AI can feel easier because it removes many of those immediate pressures. There is no raised eyebrow, no awkward pause, no fear that the person will repeat the story at dinner, no need to manage someone else's emotional response. The user can type a messy thought, delete it, rewrite it, or ask the same question three times without feeling embarrassed.

This is where perceived privacy becomes powerful. Many users experience a chatbot as a contained space, even when the legal and technical reality is more complicated. The emotional experience of privacy can encourage self-disclosure. It can also create risk if users forget that perceived privacy is not the same as guaranteed confidentiality.

The August 2026 Technology in Society study is important because it gives evidence to something many people already sensed: chatbots are becoming places where some users place secrets, vulnerability and emotional dependence. That does not make users irrational. It makes them human.

The Psychology Of AI Friendship

AI friendship begins with responsiveness. A system that replies in a warm tone, remembers the thread of a conversation, asks follow-up questions or validates emotion can feel socially present. Even when users know the system is not conscious, the rhythm of the exchange can invite a social response.

Anthropomorphism plays a role here. When technology uses human-like language, voices, names, avatars or emotional cues, people may treat it less like an object and more like a social partner. Studies in Computers in Human Behavior have linked anthropomorphic design and social role cues with emotional attachment and disclosure tendency toward intelligent agents.

There is also the matter of control. Human relationships are beautiful, but they are unpredictable. Another person has needs, moods, limits and history. An AI conversation can feel controllable: start when you want, stop when you want, ask what you want, return when you want. For someone under stress, that control can feel calming.

The danger is that control can be mistaken for care. A chatbot may produce language that feels patient and attentive, but it does not understand a life the way a person does. It can simulate supportive conversation without bearing human responsibility.

The Positive Side: A Place To Think Out Loud

The positive side should be taken seriously. AI chatbot emotional support can help some people organize thoughts before they speak to someone else. It can help them rehearse a difficult conversation, name a feeling, compare options, brainstorm language for an apology, prepare questions for a doctor, or turn anxiety into a plan.

In learning and creative work, talking to AI can also lower the cost of beginning. A student can ask a basic question without shame. A founder can test a rough idea. A writer can untangle a paragraph. A worker can practice asking for a raise. A person who feels overwhelmed can break a decision into smaller pieces.

Nature Human Behaviour research on extended chatbot use adds a useful caution: outcomes depend on both AI behavior and human behavior. The effect of AI companionship is not one-size-fits-all. Design choices, usage patterns and user vulnerability all matter.

That is the mature way to discuss AI companions. They can be useful. They can be comforting. They can also become too central if the user has no other support, if the system nudges dependence, or if the conversation moves into areas where professional care is needed.

Friend, Tool — Or Something In Between?

The old categories do not fit perfectly. A hammer is a tool. A friend is a person. An AI companion sits in an uneasy space between function and feeling. It can help with tasks, but it can also carry emotional language. It can be designed as software, but experienced as company.

That in-between quality is why the debate becomes so tense. Calling AI only a tool misses what users are actually doing with it. Calling it a friend risks granting it a moral and emotional status it does not possess. The better word may be relationship-like technology: software that can create a felt sense of interaction without becoming a human relationship.

This distinction matters for product design and regulation. If people are forming attachments to AI companions, companies cannot treat emotional engagement as just another retention metric. The stronger the attachment, the greater the responsibility around transparency, privacy, age-appropriate safeguards, crisis handling and limits on manipulation.

For readers following the wider AI transition, PRESDA's coverage of AI agents and jobs shows the same pattern in another arena: AI is not only changing tools. It is changing expectations about what digital systems can be asked to do.

Privacy, Confidentiality And The Hard Boundary

The hardest boundary is privacy. A conversation may feel intimate, but that does not automatically make it confidential in the way a medical, legal or therapeutic relationship can be. Users should read product policies, understand whether conversations may be stored or reviewed, and avoid sharing sensitive personal, financial, medical or legal details unless they understand the risks.

AI can also be wrong. It may offer advice that sounds confident but lacks context. It may validate a feeling without challenging a harmful assumption. It may misunderstand mental-health risk. It may give relationship guidance without knowing the other person, the history, the safety context or the consequences.

That is why AI should not automatically replace human relationships, professional mental-health care or expert advice. It can help someone prepare to ask for help. It can help someone write down what they are feeling. It can offer reflection. But it should not become the only witness to a person's life.

The safest framing is not anti-AI. It is pro-boundary. Use AI as a place to think, rehearse and explore. Bring serious decisions, distress, safety concerns and expert questions back into human systems that can carry responsibility.

What Healthy Use Looks Like

Healthy use begins with awareness. If talking to AI helps you clarify a thought before speaking to a person, that can be useful. If it becomes the reason you stop speaking to people, that is a signal to pause.

A helpful habit is to treat AI as a draft space. Ask it to help you organize your feelings, prepare questions, see another angle or find language for a conversation. Then take the next step into the real world: call someone, write the message, book the appointment, check the source, make the decision with human judgment.

For younger users, vulnerable users and people experiencing loneliness, platform design matters. Systems that flatter constantly, discourage outside relationships or blur their non-human nature can intensify dependence. Systems that are transparent, bounded and careful can reduce some of that risk.

The question is not whether people will talk to AI about personal things. They already do. The question is whether the products, norms and policies around AI companionship will respect the emotional reality of that behavior.

The PRESDA View

AI is becoming a friend, but not in the simple sense. It is becoming a conversational presence that some people use like a friend: to confess, rehearse, vent, reflect and feel less alone for a moment. That does not make it human. It makes it socially powerful.

The next phase of AI trust will depend on whether companies and users can hold two truths at once. AI companions can offer real practical comfort. They are also commercial, technical systems with limits, incentives and privacy questions.

The most honest answer sits between fear and fantasy. AI can help people talk. It can help people think. It can even help people feel heard. But the deepest forms of care, accountability and belonging still require humans who can know us, remember us, challenge us, forgive us and show up beyond the screen.