Changing habits is tough – can AI help?

The Challenge

Behaviour change is hard. We’re creatures of habit, and even when we know what’s good for us, we often struggle to follow through. Health advice is everywhere, yet many of us continue with the same routines — not because we don’t care, but because breaking established habits is one of the toughest things to do.

From a behavioural science perspective, that’s exactly the problem. Health messages don’t always work because they’re trying to fight against deeply ingrained patterns. Some people will act on good advice, but many others will understand the message, even want to change, and still end up doing the same old thing.

Onmi’s Approach

At Onmi, we think the only way to build new behaviours is to first disrupt the old ones. That’s where our “Do’s” come in — small, personalised behavioural prompts that challenge you to try something new.

These micro-actions are designed to gently interrupt the routines that lead to unhealthy choices. Over time, these small experiments can spark a bigger shift, helping people build healthier patterns without it feeling like a massive life overhaul. It’s not about telling people what’s right — it’s about helping them experience something different, and learn from it.

Engagement

Of course, building a smart behaviour change program is one thing — getting people to actually pay attention to it is a whole other challenge.

In today’s world of addictive apps and constant distractions, grabbing someone’s attention is like trying to shout over a hundred other voices in a crowded room. And even when you do get through, there’s no room for error. One off-message notification or generic tip can make people tune out or lose trust.

That’s why relevance matters. People respond when they feel something was made just for them.

Large Language Models

The good news? We finally have the tools to do personalisation at scale. With large language models (LLMs), we can craft tailored content that fits different personalities, moods, and goals — without needing a team of writers working around the clock.

But AI isn’t magic. The tricky part is making sure the messages are still accurate, helpful, and aligned with what we actually want to say.

FlexAI in COMFORTage

So, we have to take a careful approach. We have two ideas about this. Firstly, we could break things down into small, specific steps — giving the AI less room to make mistakes. Alternatively, we can write the core of the behavioural prompts ourselves, then let the AI adapt the details for each person. That way, we get the benefits of scale without losing control of the message.

In COMFORTage, we’ll be testing both methods to see what gives the best mix of confidence and personalisation. We’re working on an AI-based system now and will soon start testing with the University of Manchester. After that, we’ll keep improving it with input from real users.

Because if we want people to change their behaviour, we need to meet them where they are — and that starts with the right message, at the right time, in the right voice.

**Article written by ONMI B.V., a key partner in the COMFORTage project.