For years, I have sat with a troubling paradox at the heart of my field. We know, with remarkable confidence, what good parenting support looks like. Intensive, in-person programs – delivered by trained facilitators, grounded in evidence, sensitive to culture – can measurably reduce violence against children, strengthen the bonds between caregivers and their kids, and set young lives on better trajectories. The science is solid.

The problem is that almost no one can access it.

Time, money, logistics, stigma: these are the walls that keep evidence-based parenting programs out of reach for the families who need them most. For a generation of researchers, the question has been whether digital technology – smartphone apps, chatbots, online content – can bridge that gap. The answer, to be honest, has been discouraging. Self-guided digital interventions have struggled with dismal engagement. Completion rates as low as 0.5% are not uncommon. A parent downloading a parenting app in a moment of anxiety rarely becomes a parent who finishes it.

So when my colleagues and I set out to test a blended digital-human parenting program in a preschool in Xinyu, a lower-middle-income city in central China, we were not naive. We did not think a chatbot on WeChat would transform family life on its own. Our hypothesis was more specific, and I believe it turned out to be correct: the technology was never the point. The human connection was.

What we did and what we found

The program, called Keyushiguang, was delivered through WeChat – the messaging platform that is near-universal in China – over two and a half months. It comprised 37 to 39 brief interactive modules covering topics like positive reinforcement, non-violent discipline, and responsive caregiving. Crucially, the chatbot content did not stand alone. Every week, preschool headteachers and social workers facilitated small group discussions in existing WeChat class groups, drawing on the same themes caregivers were encountering in the app. Children received small developmentally appropriate rewards linked to their parents’ engagement – a child-centered incentive that quietly transformed the program from a solitary exercise into a family affair.

Five hundred and forty-one caregivers of children aged three to six participated in this cluster randomised controlled trial – the first pragmatic RCT of its kind. What we found surprised even us.

Sixty percent of participants completed all chatbot modules. In a field where single-digit completion rates are the norm for digital interventions, that figure stopped us in our tracks. More importantly, it was not just engagement that improved. At the end of the program, caregivers in the intervention group were spending significantly more time in learning and stimulation activities with their children. Caregiver-perpetrated physical violence dropped by 29%. Attitudes endorsing corporal punishment shifted. Children showed fewer emotional and internalising behavioural problems. Parental anxiety fell. And when we followed up at six and twelve months, many of these gains had held.

Twenty-five percent of our participants were male – a demographic that traditional in-person parenting programs rarely reach, not for lack of trying but because of the structural and cultural barriers that make it hard for fathers and grandfathers to walk through a community center door. A chatbot they could open at midnight, quietly, with no one watching, was a different proposition entirely.

Why the human layer made the difference

I want to be careful here, because it would be easy to read our results as a story about technology. It is not.

The chatbot delivered the curriculum. But the headteachers provided the legitimacy. Parents in China have deep, trusted relationships with their children’s preschool teachers – relationships built on years of daily contact, school events, report cards, and informal hallway conversations. When a teacher said, in effect, “this program matters, I am here to support you through it,” that carried weight that no algorithm can replicate. Embedding the intervention within existing school-family structures helped destigmatise participation. It turned “seeking parenting help” into “engaging with the school.” For many caregivers, that reframe made all the difference.

The weekly group discussions served a different but equally vital function. They gave caregivers a space to make sense of what they were learning – to ask whether the advice about managing a tantrum actually works when your child is seven and the flat is small and you have just come off a twelve-hour shift. Human facilitators could contextualise digital content, catch misunderstandings, and hold space for the messy realities of family life that a scripted chatbot cannot anticipate.

And then there was the token economy – the small rewards that tied children’s experience directly to their parents’ engagement. This was, in retrospect, one of our most elegant design choices. It converted what might otherwise feel like a private self-improvement project into something the whole family was participating in together.

The equity dimension we did not expect

One finding from our study deserves particular attention, because it points toward something the parenting field has been slow to address.

Nearly a quarter of the children in our sample potentially had some form of functional difficulty – mobility, hearing, vision, cognitive, behavioural, or other. And nearly sixteen percent of caregivers themselves reported a functional difficulty. These are families who are routinely overlooked by parenting programs designed for the median participant, and who face disproportionately high risks of both poor child development outcomes and violence.

Our exploratory analyses showed that the program’s core principles – positive reinforcement, responsive caregiving, non-violent discipline – appeared to work across disability contexts. Children with functional difficulty were less likely to experience caregiver-perpetrated physical violence in families that received the intervention. Caregivers with functional difficulty showed similar improvements in parenting behaviours. The disability-inclusive adaptations we built into the program – inclusive language, representative visuals, flexible content – appeared to matter.

I do not want to overstate this. Family functioning improvements were weaker in households where the caregiver had a functional difficulty, suggesting the program was still not sufficiently tailored to the complex, layered needs these families face. But the signal is there, and it points toward what I would call a twin-track imperative: universal programs built on inclusive principles, alongside targeted support for families navigating functional difficulty. These are not competing priorities. They are complementary ones.

What this means for the field

I have spent much of my career working at the intersection of early childhood development, violence prevention, and global health – fields that have long struggled with a painful gap between what we know and what we can deliver. The conversation about digital technology in parenting support has often fallen into one of two camps: breathless optimism about apps as a solution, or scepticism that anything short of intensive face-to-face intervention can move the needle on entrenched behaviours.

Our study suggests both camps are wrong, or at least incomplete.

Digital tools, deployed carefully, adapted thoughtfully to local culture and context, and paired with meaningful human connection, can achieve outcomes comparable to selective or targeted interventions – at a fraction of the cost, with far greater reach, and with the flexibility that modern families actually need. Sixty percent module completion in a real-world setting is not a laboratory result. It is evidence from a public preschool in a lower-middle-income city, where parents are busy and stressed, and nobody was paid to finish the program.

The implications for how we design, fund, and scale parenting programs are significant. The answer is not to pour resources into ever-more-sophisticated AI, nor to insist on in-person delivery as the only credible model. The answer is to think carefully about what technology is genuinely good at – delivering consistent, personalised content at scale, at any hour, without stigma – and what human connection is irreplaceable for: trust, accountability, reflection, and the felt sense that someone who knows you is paying attention.

A note of humility

I am a researcher, not an advocate, and I want to be honest about what our study cannot tell us. Our evidence comes from a single preschool in one city in China. The trusted teacher-parent relationships that drove so much of our engagement may not transfer easily to settings where those relationships are weaker or more fraught. Cultural context – the deep respect for educational authority in Chinese family life, the ubiquity of WeChat – shaped our results in ways we cannot fully disentangle.

We also know that sustained behavior change is hard. Some of our gains on corporal punishment attitudes and parenting stress showed signs of attrition at the twelve-month mark. A two-and-a-half-month intervention cannot undo decades of socialized parenting patterns on its own. Booster sessions, ongoing community connection, and integration with broader social services will likely be necessary to sustain what programs like ours begin.

And there are the families who did not complete the program – the 40 percent who dropped off before finishing all modules. Understanding who they are, what got in the way, and how to reach them is as important as celebrating the majority who stayed.

The window that is always closing

Early childhood is not a policy priority that can be deferred. The brain is not infinitely plastic. The years between birth and age six are not interchangeable with the years that follow. Around the world, approximately 250 million children under five are at risk of not reaching their developmental potential – and behind each statistic is a family doing their best with the tools and support they have, which is often very little.

We have, for too long, offered those families a choice between an intensive program they cannot access and a digital app they will not finish. Our study suggests there is a third way – not perfect, not universal in its current form, but genuinely promising. A way that uses technology to extend human care rather than substitute for it.

The chatbot did not change how those parents raised their children. The teachers did. The community did. The children themselves did, in the small moments of reward and recognition that kept their parents opening the app each morning. The chatbot just made it possible for all of that to happen at scale.

That, I think, is the right way to understand what technology can do for children. Not replace the human beings who raise them. Help more human beings do it well.