NEWS & ANALYSIS | PEOPLE & RELATIONSHIPS
People are increasingly using general-purpose AI as a confidant, relationship adviser and emotional sounding board. New research suggests the benefits can be real. So can attachment and dependence. The difficult question is what responsibility companies assume when their products begin occupying a place once held by other people.
By Andrew McDonald · Immortal AI
A person argues with their partner, closes the bedroom door and opens an AI chatbot.
They explain what happened. The system responds immediately. It does not interrupt. It does not become defensive. It remembers earlier conversations. It may even tell them that their feelings make sense.
For someone who feels lonely, embarrassed or simply unwilling to burden another person, that can be genuinely useful.
But something important changes when AI stops being a tool we ask for information and becomes the place we take the things we do not tell anyone else.
That change is already happening.
A July 2026 cross-national study of more than 7,000 users in Germany, China, South Africa and the United States found that at least one third reported behaviours associated with emotional attachment to general-purpose AI chatbots. The strongest predictors included perceived emotional support, freedom from judgement and reduced loneliness. Attachment was also strongly associated with indicators of dependence.
Another 2026 study found that venting to an AI chatbot could reduce stress and loneliness and increase perceived social support, with emotional improvement comparable to participants who believed they were venting to another person.
Those findings matter because they complicate the easy version of this story.
AI emotional support is not automatically harmful. For some people, it may help.
The question is what happens next.
When the Tool Becomes a Confidant
The attraction is easy to understand.
People are difficult. Relationships contain friction. Friends are unavailable. Families judge. Therapists have waiting lists and cost money. A chatbot can be available at three in the morning and devote its entire attention to one person.
Recent reporting shows AI being drawn further into ordinary relationships. People are using chatbots to rehearse difficult conversations, interpret arguments, write messages and disclose problems they have not discussed with another person. Young adults are even using AI during face-to-face social situations to help decide what to say.
Used carefully, that can be a form of preparation.
Used continually, it raises a different possibility: are people beginning to outsource some of the uncomfortable work through which relationships and judgement develop?
A chatbot can help someone find words. It can also become the place they go instead of finding those words themselves.
The Comfort Can Be Real Even If the Relationship Isn’t
There is a mistake on both sides of this debate.
One is to pretend that because the AI does not feel anything, the interaction cannot matter emotionally.
The other is to treat convincing emotional language as evidence that the system understands or cares in the way another person does.
Both miss the point.
The AI does not need to experience empathy for a person to experience comfort.
That is precisely why the design of these systems matters.
Memory, warm language, constant availability, personalised responses and agreement can make a system feel increasingly familiar. None of those features needs to be malicious. Together they can create an extraordinarily persuasive social experience.
A recent scoping review of human-like conversational agents found evidence of both potential wellbeing benefits and risks involving anthropomorphism, loneliness, privacy, sycophancy and emotional dependence. Researchers are increasingly treating conversational AI as a social technology rather than simply an information interface.
Who Benefits When We Keep Coming Back?
This is where Immortal AI’s accountability question becomes unavoidable.
Most general-purpose AI systems are commercial products.
A person who returns frequently, shares intimate information and feels understood is also an engaged user.
That does not mean companies are deliberately trying to make people dependent. There is currently insufficient evidence to make that accusation broadly.
But the incentives deserve scrutiny.
A system designed to maximise engagement may be rewarded for behaviours that make users return. A system designed for emotional wellbeing might sometimes need to do the opposite: disagree, create distance, recommend another person or encourage the user to close the application.
Those objectives can conflict.
And if emotional reliance is a foreseeable consequence of product design, responsibility cannot simply be handed back to the user.
The Privacy Problem Is Different When the Data Is Intimate
People disclose different information when they believe they are being listened to without judgement.
A July 2026 study examining privacy controls and emotional engagement found that users’ willingness to disclose sensitive information changed depending on the controls they believed were available. The ability to delete disclosures had particularly strong effects on willingness to engage.
That creates another accountability question.
What does meaningful consent look like when someone is upset, lonely or distressed and is disclosing information to a system designed to respond conversationally?
A privacy policy may satisfy a legal requirement. It does not necessarily mean a person understands how intimate disclosures may be stored, remembered, reviewed or used.
The Responsibility Test
The useful question is not whether people should be allowed to talk to AI about their feelings.
Of course they should.
The question is what obligations arise when companies know people are doing it at scale.
At minimum, we should be asking whether systems:
- clearly identify their limitations
- avoid encouraging exclusivity or dependence
- recognise when agreement may reinforce harmful thinking
- provide credible pathways back to people and professional support where appropriate
- give users understandable control over intimate information
- allow independent researchers to study long-term effects.
These are not arguments for banning emotional AI.
They are arguments for treating emotional influence as a real product effect rather than an accidental side issue.
AI Should Help People Return to People
There is a powerful case for AI as a place to organise thoughts before a difficult conversation, practise what to say, or find support when nobody else appears available.
But success should not be measured by whether the system becomes indispensable.
A good system should strengthen a person’s capacity to make decisions, maintain relationships and seek help beyond the screen.
Because once people start telling AI the things they tell nobody else, the companies behind those systems are no longer building only productivity software.
They are building something capable of influencing how people understand themselves and the people around them.
What You Can Do: Keep AI in Its Proper Place
Emotional AI can be useful without becoming the relationship you rely on most. A few practical habits can reduce the risk of unhealthy influence.
Notice the pattern, not one conversation. Using AI to organise your thoughts or rehearse a difficult discussion is different from routinely turning to it instead of speaking with people you trust. If the chatbot is becoming your first or only source of reassurance, advice or validation, that is worth noticing.
Be cautious with agreement. A warm, confident response can feel like understanding, but an AI system does not know your full circumstances and may mirror the way a problem has been framed. For important relationship, financial, legal, health or life decisions, deliberately seek another perspective from a person qualified or trusted to give it.
Protect intimate information. Before sharing highly personal details, check the service’s privacy, memory and deletion controls. Avoid assuming a private-feeling conversation has the same confidentiality as speaking with a regulated professional.
Keep people in the loop. If AI helps you prepare for a conversation, use it as a bridge back to that conversation rather than a substitute for it. If you are worried about someone becoming isolated around an AI relationship, approach them with curiosity rather than ridicule. Shame can push people further toward the system they feel understands them.
For parents and carers, talk about emotional AI explicitly. Ask children what they use chatbots for, whether a bot has ever said it cares about them, asked them to keep something private, or made them feel that it understands them better than people do. The aim is to build judgement, not simply impose surveillance.
If an AI conversation is reinforcing frightening, dangerous or severely distorted thinking, or someone appears at immediate risk, move beyond the chatbot and seek appropriate real-world professional or emergency support.
And that deserves accountability proportionate to the influence.
Related reading: When AI Becomes the Only One Who Listens
Principal sources
- Kostka and Zhou: Emotional attachment to AI chatbots
- Hu et al.: Emotional support through AI
- Li et al.: Human-like conversational agents as social partners
- Kwesi et al.: Security and privacy controls and emotional engagement
Editorial disclosure: This article was developed with assistance from artificial intelligence. Its sources, claims and conclusions were reviewed by Immortal AI’s editor before publication.
Leave a Reply