Designing for Human Emotions: A Research Perspective

Usability tells us whether a system works.

Emotion tells us how it lands.

In HCI and UX research, emotional experience is often treated as secondary, something inferred after task success, satisfaction scores, or usability metrics. But emotion is not an outcome layered on top of interaction. It is part of the interaction itself.

If we don’t explicitly research emotional experience, we don’t get neutral products, we get emotionally unmanaged ones.

Why Emotions Are Harder to Study (But Necessary)

Emotion has traditionally been sidelined in HCI because it is:

🤓Contextual rather than universal

🤓Dynamic rather than static

🤓Harder to quantify than task success or time-on-task

Yet decades of research in psychology and affective computing show that emotions shape:

• Decision-making

• Trust and compliance

• Cognitive load

• Memory and long-term product perception

From a research standpoint, the challenge is not whether emotion matters, but how to study it rigorously.

What We Can Actually Measure

Emotions are not directly observable, but they leave traces. In UX research, these traces show up across self-report, behavior, physiology, and language.

1. Self-reported emotional data

Still essential, but often underused.

Examples: Emotion scales (e.g. valence/arousal, PANAS, SAM); Post-task questions like:

• “Where did you feel uncertain?”

• “When did you feel most confident?”

• “Which part felt emotionally demanding?”

Key insight: timing matters. Asking during or immediately after interaction captures more accurate emotional states than retrospective surveys alone.

2. Behavioral signals

Emotion often manifests as behavior before it is verbalized.

What researchers can look for:

• Hesitation, pauses, re-reading content

• Abandonment or backtracking

• Over-checking, repeated confirmations

• Apologies or self-blame in think-aloud studies (“I’m probably doing this wrong”)

These moments often indicate anxiety, lack of trust, or fear of error, not usability failure.

3. Linguistic and conversational cues

Language is one of the richest emotional data sources.

Examples:

• Politeness or hedging (“I think… maybe…?”)

• Frustration masked as humor

• Over-formal tone in systems meant to feel supportive

• Emotional labor placed on the user (“Please explain why you need help”)

In AI and conversational interfaces especially, analyzing user phrasing over time reveals shifts in trust and confidence.

4. Physiological and biometric methods (when appropriate)

Used more in academic or high-stakes contexts.

Examples:

• Heart rate variability (stress)

• Skin conductance (arousal)

• Eye-tracking (cognitive and emotional load)

These methods are less scalable, but valuable when emotional cost is critical (e.g. healthcare, safety-critical systems).

Designing Research Around Emotional States

A useful reframing is to study emotional transitions, not isolated moments.

For example:

• Entry state: anxious, rushed, uncertain

• Interaction state: overloaded, reassured, confused

• Exit state: confident, frustrated, resigned

Research questions then shift from:

“Can users complete the task?”

to:

“How does this system change how users feel about themselves and the task?”

Why This Matters Now

As systems become more autonomous and AI-driven, emotional effects scale faster than usability issues.

AI interfaces express confidence, authority, and tone — all of which directly shape:

• User trust

• Over-reliance

• Self-doubt

• Perceived accountability

Without emotional research, we risk designing systems that are efficient but psychologically costly.

Share