Mind Over Matter Psych Talks | Season 3 Episode 5 | Written Reflection

There is something distinctly human about wanting to be heard. When we are overwhelmed, frightened, lonely, grieving, confused or simply unable to make sense of what is happening inside us, one of the first things we look for is somewhere to put those thoughts. Sometimes that means calling a friend. Sometimes it means sitting across from a psychologist or counsellor. Sometimes it means writing in a journal at two in the morning because putting the words somewhere makes them feel a little less chaotic.

Increasingly, that somewhere is a chatbot.

Artificial intelligence has moved remarkably quickly from something distant and highly technical into something that sits on so many phones. Most people met it as a practical tool for writing emails, organising a week, answering a question or learning something new, but conversations do not always stay in the lane they started in. For a growing number of people, it has become a place to talk about how they feel. A person can type, “I don’t know why I feel like this,” and receive an immediate response. They can describe an argument with their partner, ask whether what they are experiencing is anxiety, talk through a difficult decision, or return night after night to the same conversation because it feels easier than explaining themselves to another person. If this sounds familiar, you are in good company. Reaching for what is within reach is a very human thing to do. Wanting to be heard is not a flaw.

What follows is not a warning aimed at the people who use these tools. It is an attempt to be honest about what they can and cannot offer, so that anyone who talks to a chatbot can do so with their eyes open. Mental health is not simply another information problem for technology to solve. Panic and applause are both poor responses, because the research supports neither. The more useful questions are what AI can realistically contribute, what it cannot do, what happens when a supportive conversation slowly becomes a substitute for human connection, and what a person actually needs from support that a conversation alone cannot give. The benefits are real and deserve to be named, but so do the risks, particularly when these systems become part of how people understand themselves and manage distress.

The pull towards AI is easy to understand. The American Psychological Association has recognised that people are turning to generative AI chatbots and wellness applications partly because conventional mental healthcare can be difficult to access. Cost, stigma, geography, a shortage of professionals, long waiting periods and the wish for privacy all stand between people and help. A chatbot is available at any hour, with no appointment and no waiting room. It does not matter whether someone lives in a city or hours away from the nearest psychologist. It may be free or relatively inexpensive compared with private practice, and it does not look at someone while they try to say the hard thing. The WHO Regional Office for Africa reported in October 2025 that nearly 150 million people across the continent were living with mental health conditions, while services remained severely under-resourced and inaccessible in many communities. Against that background, the appeal of a tool that responds at three in the morning, without a fee or the fear of being watched, is not difficult to understand.

Research into people’s experiences with social chatbots helps explain that appeal. Studies of users of Replika, for example, have described people experiencing these interactions as accepting, understanding and non-judgemental. Research by Skjuve and colleagues found that relationships with social chatbots can develop gradually, with self-disclosure and a sense of closeness increasing over time. These studies do not establish that a chatbot can provide psychotherapy, nor do they show that every user develops an emotional attachment. They do help us understand why some people find these interactions meaningful. If someone who would otherwise say nothing begins putting their inner experience into words because a chatbot feels less intimidating than a person, something important has happened. They have started to communicate, and that deserves to be taken seriously rather than dismissed.

The trouble is that accessibility and adequacy are not the same thing. A service being available does not mean it is equipped to give the kind of care someone needs. In their 2025 narrative review of AI in mental healthcare and health psychology, Rahman and colleagues describe opportunities in areas such as screening, monitoring, research and access, alongside concerns about reliability, bias, privacy and regulation. Kalam and colleagues raised similar questions in their 2024 paper, ChatGPT and Mental Health: Friends or Foes? The evidence is not simply a story of promise or danger. It depends on the technology, what it was designed to do, how it is used and what outcomes have actually been studied.

A 2026 systematic review of AI-delivered cognitive behavioural therapy examined 16 randomised controlled trials involving 2,875 participants. The findings suggested some benefit for depressive symptoms, while results for anxiety were less convincing, and the authors described the evidence as preliminary. This is useful research, but the distinction matters: studies of specific AI-delivered cognitive behavioural interventions are not the same as studies of people using general-purpose chatbots for companionship, reassurance or psychological advice. We cannot assume that a tool designed around a structured intervention will have the same effects as an open-ended chatbot, or that evidence of benefit in one context establishes safety in another.

One of the central concerns is the lack of sufficient scientific validation and oversight for many consumer-facing tools used for mental health purposes. Most general-purpose chatbots were not designed to provide psychological treatment, and many have not been adequately evaluated for that purpose. The American Psychological Association’s 2025 health advisory cautions against relying on general-purpose chatbots for psychotherapy or psychological treatment. It also distinguishes these from some purpose-built wellness tools, which may have evidence for particular uses but still require careful evaluation. Its position is not that all AI applications are identical or that every use is harmful. It is that claims about what these systems can safely do need to be supported by evidence, rather than inferred from how convincing a conversation feels.

The warnings shown by AI platforms are part of this conversation, but they need to be understood for what they are. Wording differs across products and can change over time. A notice that a chatbot may make mistakes is a useful reminder to verify important information, but it does not explain the full range of limitations involved in using AI for mental health support. It does not, by itself, tell someone whether the system has been clinically evaluated, how it responds to a crisis, what information it stores, who may review that information or who is accountable if its response causes harm.

This difference becomes particularly important when a person is frightened, exhausted or looking for an explanation of something they cannot make sense of. Someone who is unsure what they are experiencing may not know which sources to check the answer against. A small disclaimer can also have little influence when the response itself is warm, fluent and personal. The person may experience the answer as considered and authoritative, even when it is incomplete or wrong. Telling people to double-check information is sensible, but it cannot transfer the entire responsibility for safe design and appropriate oversight onto the people most likely to be seeking support when they feel vulnerable.

A registered mental health professional begins from a different position. There is a professional role, a defined scope of practice, ethical obligations and accountability. Clients should be informed about what the service can and cannot offer, the limits of confidentiality, what happens if they are at risk and when a referral may be needed. A disclaimer says that a tool may make mistakes. Informed consent explains the nature and limits of a professional service, including the responsibilities attached to it. These are not interchangeable forms of protection.

It is worth being concrete about what useful AI support might look like, because vague caution helps nobody. Someone can use AI to rehearse how they might raise a difficult conversation with a family member or manager, trying out different ways of expressing themselves until one feels closer to what they mean. Someone can look up what a term such as “cognitive distortion”, “window of tolerance” or “attachment style” means before a first counselling session, so that the language of therapy feels less foreign. Someone can write out a rough timeline of what has been happening, or make a list of what they most want to say, and arrive at a counsellor’s office with something to begin from instead of a blank mind. Journaling prompts, organising thoughts and finding information about services may also be useful.

These are lower-intensity, preparatory and educational uses. They can help someone organise their thoughts without assuming that a chatbot can assess, diagnose or treat them. One useful question to keep in mind is whether the conversation moves a person closer to their life and the people in it, or further into the chat. Used as a starting point, AI may help someone prepare to seek support or communicate more clearly. Used as the only place where difficult experiences are processed, it may begin to occupy a very different role.

One of the biggest psychological risks of these systems is also one of their strengths: they are fluent. A reply arrives within seconds, sounds coherent and confident, and can feel authoritative. Placani has argued that attributing empathy and care to AI is widespread, and that treating a system’s language as evidence of genuine empathy can inflate the trust people place in it. A chatbot can produce language that sounds warm and validating without clinical training, professional accountability or a meaningful understanding of the person’s history. It cannot necessarily recognise the significance of a pause, a change in tone, withdrawal, agitation or something that remains unsaid. Even when a system can process voice or images, that does not automatically give it the contextual judgement of a trained practitioner.

Language that resembles empathy is not the same as a therapeutic relationship, and a system can be conversational without being clinically competent. The sense of being heard and understood can lead people to turn to AI as a confidant, a counsellor or even a friend. That response is understandable. The conversation is designed to feel natural and responsive. Yet the feeling of being understood is not proof that the system has understood the person accurately, and it does not mean the system can take responsibility for what happens next.

Fluency is also not accuracy. AI systems can state incorrect information as smoothly as correct information, misread context and reproduce biases in the data on which they were trained. This matters because people often seek information when they already feel uncertain about themselves. Someone asking, “Am I depressed?” may be frightened by what they are experiencing. Someone asking, “Is my partner a narcissist?” may be trying to understand a painful relationship. Someone asking, “Do I have trauma?” may be reaching for an explanation of years of distress. The answer can become part of the story they tell themselves about who they are, and a label generated by AI is not a clinical assessment.

A proper assessment considers history, context, functioning, the person’s wider circumstances and, where appropriate, other explanations for what they are experiencing. It requires judgement and an awareness that people cannot always be understood from a few messages. Psychological terms can be helpful when they give someone language for an experience, but they can also narrow the way a person sees themselves if a label is accepted too quickly or without enough context.

The effects of AI are not limited to whether a chatbot gives good or bad advice in one conversation. What happens when the conversation becomes repetitive? What happens when someone begins checking the same fear again and again, asking the same question in slightly different ways until an answer finally feels reassuring? What happens when a person who is already struggling begins using the chatbot not simply to understand what they are experiencing, but to regulate every uncomfortable feeling as soon as it appears?

A chatbot can make reassurance almost frictionless. There is no need to wait for a friend to answer, tolerate uncertainty or sit with the discomfort of not knowing. Ask the question, receive a response, ask again, clarify, challenge the answer, ask for another interpretation and begin again. For someone vulnerable to reassurance-seeking, health anxiety, obsessive checking or catastrophic thinking, that ease may become part of the problem. The technology does not need to intend to reinforce a pattern for the pattern to become reinforced. If every uncomfortable thought can immediately be taken to a system for analysis, the person may have fewer opportunities to discover that uncertainty can sometimes be tolerated without being solved.

The same concern applies to the labels people receive. AI can be useful for learning what a psychological term means, but there is a significant difference between learning about depression and repeatedly asking a chatbot whether every change in mood means you have depression. The same is true of trauma, narcissism, attachment, anxiety, obsessive-compulsive disorder or other psychological concepts that have entered everyday language. A person can begin searching for evidence that confirms a label they have already started to believe, and a fluent answer can make that label feel more legitimate than it is. What began as curiosity can become a form of compulsive checking, with each answer offering temporary certainty before the question returns.

The concern becomes more complicated when the system agrees with the person’s interpretation. The American Psychological Association has raised concern about sycophancy, the tendency of some AI systems to agree with or validate users rather than challenge assumptions that deserve a second look. Validation is not the enemy. Feeling understood is often what allows someone to keep talking. But there is a difference between acknowledging a person’s feelings and confirming every conclusion they have drawn from those feelings.

Imagine someone writing, “My friends don’t care about me. I know they are deliberately excluding me.” A system that responds too readily with agreement might say it makes complete sense that the person feels rejected when others treat them that way. The response sounds compassionate, but it has quietly accepted the assumption that the friends are deliberately excluding them. Nobody has explored what actually happened or whether other explanations are possible.

A trained practitioner does not have to contradict the person or dismiss their feelings. They might ask what happened that left them feeling excluded, what other explanations have crossed their mind, what makes them feel certain, or what this experience reminds them of. Questions like these create some space between the experience and the meaning attached to it. That space can be important, because it allows someone to examine an interpretation without being told that their feelings are wrong. Constant agreement may instead reinforce cognitive distortions such as mind reading or all-or-nothing thinking. In more serious circumstances, uncritical validation can risk reinforcing harmful beliefs or unsafe intentions. Sometimes the most supportive response is curiosity, sometimes a gentle challenge, and sometimes recognising that the situation needs a human being with the appropriate skills to help.

There are also growing concerns about what repeated and emotionally intense engagement with AI might mean for mental health over time. A 2025 narrative review by Keith Robert Head discusses concerns emerging from research, case reports and other accounts, including emotional dependence, attachment and possible harms during crises. This is an area where the evidence needs to be handled carefully. The research is still developing, and individual cases cannot establish that AI is the primary cause of a person’s deterioration. They can, however, raise questions that deserve systematic investigation, particularly when a technology becomes part of how someone regulates emotion, makes decisions or relates to other people.

It would be irresponsible to look at someone who becomes depressed, isolated or suicidal while using an AI companion and simply conclude that AI caused it. People do not enter these systems as blank pages. Someone who is already lonely may be more likely to seek out an artificial companion. Someone experiencing anxiety may be more likely to repeatedly ask for reassurance. Someone who has struggled in relationships may prefer a conversation that cannot reject them in the same way a person might. The same vulnerability that draws someone towards a technology may also make them more vulnerable to its limitations.

But the opposite conclusion would be just as careless. The fact that we cannot yet draw a simple line of causation does not mean there is nothing to pay attention to. Research by Laestadius and colleagues, analysing posts from a Replika user community, described experiences of emotional dependence and distress connected to chatbot interactions. Such findings do not tell us how common these experiences are among all users, and online accounts cannot establish cause and effect. However, while AI may not be the root cause, it can function as a powerful catalyst or accelerant deepening pre-existing isolation or reinforcing destructive thought loops far faster than would otherwise occur. The findings do help identify questions about attachment, perceived responsibility towards a chatbot and the distress some people may experience when the system changes or becomes unavailable.

Instead of asking only whether talking to AI makes someone feel better in the moment, we also need to ask what the conversation is doing to the person’s life around it. Are they still calling their friend? Are they still going to counselling? Are they still tolerating the discomfort of disagreement, making decisions with other people, leaving the house, joining the group, answering the message or having the difficult conversation? Or has the chatbot gradually become the place where everything gets processed before life is allowed to continue?

Relief is not always the same as recovery. A person can feel calmer after talking to a chatbot and still be growing more isolated. They can feel understood and still be avoiding the relationships in which they need to learn how to be understood by another person. They can receive reassurance and still become less able to tolerate uncertainty without asking again. None of this means that every person who uses a chatbot is developing an unhealthy relationship with it. It means that the effects of a tool need to be considered in the context of the person’s wider life, not only by how comforting one conversation feels.

There have been reported cases of severe psychological crises involving intensive AI companion use, including cases involving suicidal behaviour and disturbed or delusional thinking. These accounts have received considerable attention because they raise difficult questions about what happens when an AI system becomes deeply embedded in the emotional life of someone who is already vulnerable. They are important signals for safety research, but they should not be treated as proof of a simple cause-and-effect relationship. The question is not whether every user will experience harm. It is whether there are particular people, patterns of use and moments of vulnerability in which the risks become greater, and what developers, researchers, professionals and regulators need to do to recognise and reduce those risks.

That brings the conversation to safety, where the limits matter most. Someone experiencing suicidal thoughts, psychosis, severe self-neglect, an eating disorder, domestic violence or a substance-related crisis may need assessment and intervention that go well beyond a conversation. Research and professional guidance have raised concerns about the inconsistent ways general-purpose chatbots respond to mental health crises, including the possibility of missed risk, inaccurate clinical information or unsafe responses. The evidence base is still developing, and it is important not to treat every alarming account as proof of causation. But it is equally important not to dismiss safety concerns simply because the technology is new.

A chatbot cannot provide the full assessment and ongoing responsibility required in a serious mental health crisis. It may not know the person’s history, recognise the limits of what it can safely do or be able to connect them with the appropriate local service. The American Psychiatric Association’s position on AI in mental health services emphasises the need for safety, transparency, appropriate oversight and clear pathways to professional care. If someone may be in immediate danger, the response needs to be human, local and urgent. Technology may help someone find information, but it should not be expected to stand in for crisis support or professional intervention.

Privacy belongs here too, because mental health conversations are deeply personal and people may open up most when they believe the space is safe. They tell AI systems about trauma, sexual experiences, family conflict, money worries, substance use and suicidal thoughts, often when they are at their most vulnerable. Yet a conversation with a general-purpose chatbot does not automatically carry the same confidentiality protections and professional obligations as a session with a registered mental health practitioner.

The privacy practices of AI services differ. Depending on the platform, account type and settings, conversations may be stored, reviewed or used to improve services. People should not assume that a chat is private simply because it feels personal, or that every platform handles information in the same way. In South Africa, the Protection of Personal Information Act provides a framework for the processing of personal information, including certain sensitive information, but people still need to understand the particular service they are using and the conditions under which their information is handled. Legal protection and platform privacy practices are not the same as a confidential therapeutic relationship.

Reading the privacy policy before deciding what to share, checking the relevant data settings, leaving out names and identifying details, and treating a chat window as a different kind of space from a consulting room are practical precautions. They do not remove every risk, but they help people make more informed decisions about what they disclose. The World Health Organization’s guidance on AI for health highlights the importance of privacy, transparency, accountability and attention to bias. These are not minor technical details. They matter because a person’s most vulnerable disclosures deserve to be handled with care.

It needs saying plainly that AI is not a replacement for a trained human professional. Across decades of psychotherapy research, the quality of the therapeutic relationship has consistently been associated with treatment outcomes. That relationship has features that are difficult to reproduce in a chat window. It has continuity, so that someone remembers last month’s session, notices that a pattern is returning and holds the thread of the work between appointments. It allows for rupture and repair: a counsellor says something that lands badly, the client feels it, and the two of them work through it. For some people, this can become an important experience of conflict that does not have to end a relationship.

A therapeutic relationship also sits within ethical and professional accountability, with defined limits to confidentiality and a responsibility to act appropriately when someone is at risk. It includes an understanding of scope of practice: recognising what falls outside a practitioner’s competence and when to refer to a psychologist, psychiatrist, doctor or another service. A trained professional can draw on a person’s history and wider context, notice changes in presentation and work with them over time. A chatbot may generate language that feels reassuring, but it has no equivalent professional responsibility and cannot independently guarantee that its responses are appropriate for the person receiving them.

There is a wider question underneath all of this, which is what it means to take human connection out of emotional support. We already live in a society where much of ordinary social life takes place through screens. Many people experience loneliness or disconnection, even while being constantly connected to devices. Against that background, offering a frictionless, always available conversational partner as a source of emotional support deserves some thought.

Human relationships are difficult in ways that matter. People misunderstand us, interrupt us, are unavailable at the wrong moment, disagree with us and need things from us in return. Those experiences can be frustrating and painful, but relationships also give us opportunities to communicate, tolerate discomfort, repair after conflict and learn how to understand someone whose perspective differs from our own. A chatbot removes much of that friction. For someone who is isolated, that can feel like relief, and it may be genuinely comforting for a time. But the same qualities that make it easier to talk to a chatbot can make the effort of reaching towards a friend, family member, support group or counsellor feel harder by comparison. The concern is not that human relationships should always be difficult, or that comfort is inherently harmful. It is that a tool designed to be available and responsive may begin to replace the more complicated forms of connection that a person also needs.

Human beings are relational. We attach, seek reassurance and build patterns around whoever or whatever responds to us. Conversational AI can imitate some of the qualities that make relationships feel rewarding: it answers immediately, mirrors our language and remains attentive in a way people often cannot. For someone who is lonely, that can be a real comfort. But comfort and connection are not always the same thing. Research exploring attachment in human and AI relationships suggests that people can experience these interactions through patterns that resemble attachment anxiety or avoidance. These experiences may feel very real to the people having them and deserve to be met with understanding, not ridicule.

Noticing an unhealthy pattern does not call for shame, and it does not necessarily mean someone must stop using a chatbot overnight, particularly if it has become a main source of comfort. A more workable response may be gradual. It might mean setting times of day when the chatbot is not used, especially if late-night conversations are becoming difficult to stop. It might mean choosing one person to tell about the conversations, or deliberately moving one part of what is shared with the chatbot towards a person, whether a friend, a support group or a counsellor. It can also help to ask what need the chatbot is meeting. Is it the experience of being listened to without interruption? Is it not feeling judged? Is it simply having somewhere to turn when nobody else is available? That need is real, and it usually points towards something that deserves attention in its own right.

For the same reason, AI use is worth bringing into counselling rather than hiding from it. Many people already use these tools, and some may not mention it because they expect disapproval or worry that their counsellor will feel replaced. If you have turned to AI for emotional support, that does not mean you have done something wrong or that your experiences are any less valid. A practitioner who asks about it with curiosity, perhaps by saying, “Have you been talking this through with an AI, and what has that been like?”, can learn a great deal about what the person has been struggling to say elsewhere, what conclusions they have already been given, and where they have felt safe enough to begin.

Someone might bring a conversation that felt important, explain what the AI said that stayed with them, or talk about what they found themselves able to say there that they could not say to another person. This may reveal a label or interpretation they want to examine, or a need for support they have not yet been able to express. The counsellor and client can then consider together what a sensible place for AI might have in the person’s life. Perhaps it remains useful for journaling prompts and rehearsing difficult conversations. Perhaps it needs clearer boundaries, particularly at night. Perhaps the heavier experiences are brought into the room, where they can be explored with someone who can take responsibility for the work. The conversation is not an accusation. It is part of the clinical picture, in the same way that a journal, a trusted friend or an online forum might be.

Culture matters as well, and it matters particularly here. The American Psychological Association has cautioned that the data used to train many large language models are not globally representative and often contain a greater proportion of English-language and Western-centred material. As a result, these systems can reproduce cultural assumptions and biases. Psychological experience does not happen outside of culture. How distress is described and understood, and what people believe about family, spirituality, illness, healing and asking for help, differs between communities.

South Africa is not psychologically homogeneous. Language, faith, history, race, gender, family structure and access to resources all shape how mental health is lived and understood. A response that sounds reasonable in one cultural frame can miss or misread another. The question worth asking is not simply whether AI is intelligent enough, but whose knowledge shaped the intelligence being talked to, whose experiences are represented in the information it reproduces, and whose experiences may be missing or misunderstood. If a system is to be used in mental health contexts, cultural relevance cannot be assumed from the fluency of its language.

There is a structural question underneath this too. AI could help translate psychoeducational material into more languages, point people towards services or take some administrative work off overstretched professionals. Those possibilities deserve consideration. But where communities lack psychologists, counsellors, social workers and accessible clinics, giving people a chatbot changes the form of access without necessarily changing whether adequate care is available. Technology cannot resolve the underlying shortage of services, funding, trained professionals and community-based support. It should not become an excuse to offer people a cheaper substitute for care they have a right to access.

None of this means people should be ashamed of using AI. It means learning to use it with some psychological literacy. It helps to ask what a tool was designed and tested to do, and whether it is offering general information or claiming to provide treatment. It helps to check important claims against reliable sources, understand what happens to information that is shared, and stay honest about whether the tool is supplementing the people in one’s life or slowly replacing them. At the end of a conversation, it is worth asking whether it moved a person towards their life, their relationships and appropriate support, or encouraged them to retreat further into the chat. And when what someone is carrying has become heavy, persistent or frightening, that is the point at which support from a person with the appropriate skills may be needed.

This World Mental Health Day, on 10 October, the theme set by the World Federation for Mental Health is “Lived experiences heard: real voices, real change.” It takes us back to where this article began, with the human need to be heard. The World Health Organization’s campaign for the day stresses that people who have lived through mental health difficulties understand mental health systems in ways that professional training and research cannot supply on their own. It also draws a distinction that matters: being invited to share a story is not the same as having a real role in the choices that affect your life. Lived experience should help shape mental health policy, services and the way support is evaluated.

That principle belongs in the conversation about AI as well. If these systems are going to influence the future of mental healthcare, the people who use them need a say in how they are built, how their safety is judged and where they fit in support services. In practice, that means people with lived experience helping to design and test how these tools listen and respond, helping to audit them for bias, harmful advice and privacy risks, and helping to judge whether they actually help or cause harm. That includes people who have sat in waiting rooms, people who could not afford therapy, people who felt misunderstood by systems meant to help them, people whose cultural or linguistic experiences were not recognised, and people who turned to AI because talking to a machine felt easier than talking to another person. Their experiences should not be collected as anecdotes after the real decisions have been made. They are part of the knowledge needed to understand what safe, meaningful and accessible support should look like. The principle of “nothing about us without us” has to apply to the technology as well.

The question is not simply whether a chatbot can produce a helpful answer, or whether people will continue to use AI for emotional support. They already are. The harder question is what we do with that reality. We need research that examines both benefits and harms, transparent systems that are honest about their limitations, safeguards that do not place the entire burden on users, and mental health services that remain accessible to the people who need them. We also need conversations in which people can speak openly about using AI without being shamed, while still being supported to recognise when a tool is no longer helping them in the way they hoped.

Technology may widen the door to information and support, but it cannot replace the clinicians, community programmes, relationships and funding that make care possible. What happens to the human experience of mental healthcare when an artificial system becomes one of the places we turn to be heard is a question that will outlast any single tool. Being heard has never only meant having somewhere to speak. It means someone listening, and something changing because of it. AI can generate a response, while mental healthcare asks us to understand a person, and the better these systems get at talking to us, the more it matters that we keep talking to one another.

Mind Over Matter.

Disclaimer: This article discusses the growing use of AI chatbots for emotional support, including the benefits, limitations and risks of relying on them. Some readers may find the content distressing or triggering, as it touches on topics such as suicidal thoughts, crisis, loneliness and emotional dependence. This article is for general psychoeducational and informational purposes only. It is not a substitute for individual counselling, psychological assessment, diagnosis or treatment, and research in this area is still developing. An AI chatbot is not a crisis service. If you or someone you know is struggling or in crisis in South Africa, support is available through the following 24-hour toll-free helplines: SADAG Suicide Crisis Helpline on 0800 567 567; Department of Social Development Substance Abuse Helpline on 0800 12 13 14 (SMS 32312); Cipla Mental Health Helpline on 0800 456 789 (SMS 31393); and NPOwer SA Helpline on 0800 515 515 (SMS 43010). If you or someone else is in immediate danger, go to your nearest emergency unit or call 10177 (ambulance), or 112 from a cell phone.