A workplace wellbeing app might seem like a simple and helpful tool—a mood check-in, some stress management advice, or a chatbot asking how your week has gone. But behind that supportive language, some systems are also quietly analysing your voice, writing style and digital behaviour for signs of psychological distress.
These tools are already on the market—aimed at workplaces, universities and healthcare. They are framed as early-intervention systems that promise to cut costs and identify problems before they become serious. Unfortunately, companies are under no obligation to report using them, so data about how widespread they are is lacking.
The basic idea behind these tools is that behaviour leaves patterns. Artificial intelligence (AI) systems trained on large datasets learn to recognise signals associated with particular mental health conditions, and when similar signals appear in new data, the system produces a probability estimate.
For many people, the surprising part is how much ordinary behaviour can reveal. Voice recordings can pick up changes in rhythm, pitch and hesitation. Language models can analyse word choice and emotional tone. Smartphone data has also been explored as a way of tracking changes in sleep, movement and social interaction—all without the person doing anything out of the ordinary.
But detecting a statistical signal is very different from identifying a genuine problem. Human behaviour is deeply contextual. Someone may speak slowly because they are tired, nervous or communicating in a second language. Reduced online activity might simply reflect a busy week.
Even well-designed systems will make mistakes. A person who is genuinely struggling may not show the behavioural patterns the system was trained to recognise, while someone else may be incorrectly flagged as being in distress.