Emotion AI is the commercial label for systems that infer emotional states from observable signals — most commonly facial expression, sometimes voice, text or physiological data. It is the applied, product-facing subset of affective computing.
What it is
Typical implementations capture video, detect a face, extract landmarks, classify Action Units, and output a probability distribution across a small set of emotion categories, usually the Ekman seven.
Market estimates put emotion AI at roughly $4bn in 2026 with strong projected growth, spread across automotive, advertising research, consumer devices and education.
Why it matters
Regulatory exposure. Since February 2025, EU AI Act Article 5(1)(f) has prohibited AI systems that infer emotions from biometric data in workplace and education contexts. Anything sold as emotion AI into those markets is directly affected, regardless of whether processing happens on-device.
A common confusion
'Emotion AI' and 'engagement analytics' are used interchangeably in marketing and are legally very different. Engagement derived from behavioural signals involves no emotion inference and no biometric data.
Related
See also affective computing, emotion inference and biometric data. See the full glossary for the rest, or the EU AI Act guide for the wider context.
See where your content loses people
Book a walkthrough and we will show you the engagement data on your own content.