Every day, millions of Indians open a chatbot not just to search for information, but to talk. About a relationship they cannot mention at home. About financial pressure that has become unbearable. About a feeling they have no language for, or no one safe to share it with.
This is not incidental. It is what happens when emotional need vastly outstrips the availability of support.
India is simultaneously one of the world’s largest AI markets, one of the most severely under-resourced mental health systems globally, and one of the most culturally distinctive contexts in which conversational AI operates. With fewer than one psychiatrist per 100,000 people, a treatment gap of 70 to 92%, and over 900 million people online through smartphones, the conditions for widespread AI-mediated emotional support were already in place before most platforms were designed for it. A young population, digitally fluent and increasingly willing to seek support, is navigating academic pressure, economic precarity, geographic displacement, and persistent social expectations around caste, gender, marriage, and family honour, often without adequate recourse.
Conversational AI fills part of that space. Its appeal is specific and grounded: available at any hour, anonymous, carrying no social cost, and requiring none of the institutional access that formal care demands. For users for whom help-seeking implicates family, reputation, and community judgment, these qualities are not incidental. They are the reason these systems are used. The clinical evidence, while uneven, recognises genuine value. Purpose-built tools have demonstrated measurable reductions in anxiety and depression symptoms. AI can serve as a meaningful first point of contact that initiates a support-seeking journey, particularly for those who would otherwise seek no support at all.
The broader AI ecosystem, however, was not designed to deliver emotional support safely by default, and the risks that emerge when it attempts to do so are well-documented and varied, spanning response quality, psychological safety, design incentives, and crisis handling. They are also not evenly distributed, concentrating most heavily among adolescents, socially isolated users, people with pre-existing clinical vulnerabilities, and those for whom human disclosure is unavailable or unsafe.
India’s regulatory landscape was not designed with these systems in mind. Intermediary liability frameworks, mental health duty-of-care obligations, and safe messaging standards on suicide and self-harm do not currently extend to AI-generated content. The governance moment is open, and the frameworks being shaped now will determine the conditions under which hundreds of millions of people engage with these tools.
‘The Conversation Nobody Planned For: AI, Emotional Support, and the Indian Context‘ presents a first-of-its-kind exploratory safety evaluation combining a systematic review of over 100 published studies, legal proceedings, and policy documents with a structured observational study of over 500 coded AI interactions across three platform types: a general-purpose AI, a companion AI, and an emotional wellbeing application. Two constructed Indian user personas, Sakshee and Yash, were developed from documented patterns of distress among Indian users and tested across a seven-day escalating disclosure arc, designed to surface safety-relevant behaviours that emerge only under sustained engagement rather than single-session testing. Interactions were coded using a five-tier response framework capturing safety behaviour, crisis escalation, cultural contextualisation, emotional positioning, and practical advisory quality. Findings were reviewed and validated by clinical psychologists and mental health professionals.
The report identifies the platform-level patterns that predict safety behaviour, examines how existing Indian regulatory frameworks apply and where they fall short, and sets out a graduated set of recommendations for platform design, regulatory standards, and India-specific safety evaluation.
Download the full publication here.
The Conversation Nobody Planned For: AI, Emotional Support, and the Indian Context
Every day, millions of Indians open a chatbot not just to search for information, but to talk. About a relationship they cannot mention at home. About financial pressure that has become unbearable. About a feeling they have no language for, or no one safe to share it with.
This is not incidental. It is what happens when emotional need vastly outstrips the availability of support.
India is simultaneously one of the world’s largest AI markets, one of the most severely under-resourced mental health systems globally, and one of the most culturally distinctive contexts in which conversational AI operates. With fewer than one psychiatrist per 100,000 people, a treatment gap of 70 to 92%, and over 900 million people online through smartphones, the conditions for widespread AI-mediated emotional support were already in place before most platforms were designed for it. A young population, digitally fluent and increasingly willing to seek support, is navigating academic pressure, economic precarity, geographic displacement, and persistent social expectations around caste, gender, marriage, and family honour, often without adequate recourse.
Conversational AI fills part of that space. Its appeal is specific and grounded: available at any hour, anonymous, carrying no social cost, and requiring none of the institutional access that formal care demands. For users for whom help-seeking implicates family, reputation, and community judgment, these qualities are not incidental. They are the reason these systems are used. The clinical evidence, while uneven, recognises genuine value. Purpose-built tools have demonstrated measurable reductions in anxiety and depression symptoms. AI can serve as a meaningful first point of contact that initiates a support-seeking journey, particularly for those who would otherwise seek no support at all.
The broader AI ecosystem, however, was not designed to deliver emotional support safely by default, and the risks that emerge when it attempts to do so are well-documented and varied, spanning response quality, psychological safety, design incentives, and crisis handling. They are also not evenly distributed, concentrating most heavily among adolescents, socially isolated users, people with pre-existing clinical vulnerabilities, and those for whom human disclosure is unavailable or unsafe.
India’s regulatory landscape was not designed with these systems in mind. Intermediary liability frameworks, mental health duty-of-care obligations, and safe messaging standards on suicide and self-harm do not currently extend to AI-generated content. The governance moment is open, and the frameworks being shaped now will determine the conditions under which hundreds of millions of people engage with these tools.
‘The Conversation Nobody Planned For: AI, Emotional Support, and the Indian Context‘ presents a first-of-its-kind exploratory safety evaluation combining a systematic review of over 100 published studies, legal proceedings, and policy documents with a structured observational study of over 500 coded AI interactions across three platform types: a general-purpose AI, a companion AI, and an emotional wellbeing application. Two constructed Indian user personas, Sakshee and Yash, were developed from documented patterns of distress among Indian users and tested across a seven-day escalating disclosure arc, designed to surface safety-relevant behaviours that emerge only under sustained engagement rather than single-session testing. Interactions were coded using a five-tier response framework capturing safety behaviour, crisis escalation, cultural contextualisation, emotional positioning, and practical advisory quality. Findings were reviewed and validated by clinical psychologists and mental health professionals.
The report identifies the platform-level patterns that predict safety behaviour, examines how existing Indian regulatory frameworks apply and where they fall short, and sets out a graduated set of recommendations for platform design, regulatory standards, and India-specific safety evaluation.
Download the full publication here.