Should I use facial coding or the Immersion SDK to measure emotional response in my app?
A camera watches the face and claims to classify expressions into emotion labels. Biometric identifier + emotion recognition. The Immersion SDK measures the body's response to what the brain is doing, on the device, with no biometric identifier, no health data leaving the device, and no emotion-recognition system.
| Facial coding | Immersion SDK | |
|---|---|---|
| What it measures | Visible facial muscle movement, mapped to labels like "joy" or "anger." | Neurochemical signatures in cardiac rhythm, decoded into Value and Psychological Safety (0–100). |
| What it collects | Video of the user's face, usually a facial-geometry template. | Nothing from the user. Your app passes sensor samples in; the Immersion SDK scores them in memory and discards them. It requires no HealthKit or Health Connect permission of its own. One activation call carrying your app credentials. No PII. |
| Passive | Partly | Yes |
| Timing | Seconds after the internal response; only when the expression surfaces. | Instant. |
| Biometric identifier | Yes | No |
| Health data leaves device | No | No |
| Emotion-recognition system | Yes | No |
| Predicts behavior | Unreliable. The largest review of the evidence (Barrett et al., 2019) found facial movements are neither reliable, specific, nor generalizable indicators of emotion. | Strong. Value predicts memory and behavior with 85–98% accuracy in published work. |
Biometric identifier + emotion recognition
Facial geometry is a biometric identifier under Illinois BIPA and Texas CUBI (written consent, retention policy, private right of action in Illinois; $650M Facebook, $1.4B Meta, $1.375B Google). Under GDPR Article 9 it’s special-category data once it identifies. Under the EU AI Act, inferring emotion from a face is an emotion-recognition system: banned in workplaces and schools since February 2025, subject to a duty to inform users since August 2026, and high-risk from December 2027. California's ADMT rules name "analysis of facial expression" expressly.
In practice: A consent screen before the camera turns on, a biometric retention policy, an AI Act risk file, and a per-user disclosure that you’re reading their face, which is also the moment the reading stops being reliable.
- FPF on EU AI Act Art. 5(1)(f)
- ABA on BIPA litigation trends
- ICO, special category data
- Barrett et al. 2019, Emotional Expressions Reconsidered (PSPI)
- Texas AG: $1.4B Meta settlement (CUBI)
- Texas AG: $1.375B Google settlement
- ABA: In re Facebook BIPA, $650M settlement
- CPPA: CCPA regulations on ADMT, risk assessments (text)
- Digital Omnibus on AI (Reg. (EU) 2026/1744): high-risk deadline moved to 2 Dec 2027
A prediction engine, not an emotion-recognition system.
No biometric identifier, so BIPA and its cousins don’t apply. No health data stored or transmitted, so nothing crosses a border that Apple, Google, or GDPR Article 9 would care about. No named emotions inferred, so it’s not an emotion-recognition system under the EU AI Act. It calculates from whatever sensor your app already reads and computes two behavioral predictors inside your app: a prediction engine that draws no inference from a face, a voiceprint, or a brain scan.
In practice: The Immersion SDK asks your users for nothing. It requires no HealthKit or Health Connect permission and no entitlement of its own: your app hands it samples from whatever source it already has (a BLE fitness sensor, a watch, a file) and it hands back Value and Safety. A BLE fitness strap feeding fitness data into your app needs no health permission at all. If your app reads heart rate through HealthKit or Health Connect, that permission is your app's requirement, not the Immersion SDK's. No biometric consent, no emotion-recognition disclosure, no health data leaving the device.
Where facial coding persists, and why that doesn’t help
You’ll still find it in media-testing labs and ad-research panels, because a camera is already pointed at the participant. That doesn’t make the reading valid. Barrett and colleagues reviewed roughly a thousand studies and found that people scowl in only about 30% of anger episodes and scowl for many other reasons, so a label read off a face is a guess dressed as a measurement. Consent solves the legal problem. Nothing solves the science problem.
- Barrett, L. F., Adolphs, R., Marsella, S., Martinez, A. M., & Pollak, S. D. (2019). Emotional expressions reconsidered: Challenges to inferring emotion from human facial movements. Psychological Science in the Public Interest, 20(1), 1–68. doi.org/10.1177/1529100619832930
Regulatory summaries are for orientation, not legal advice. Confirm classification for your own app with counsel.
Related
All guides- Why would heart rate tell you anything about the brain?The oxytocin → vagus nerve → cardiac rhythm mechanism, with citations.
- Heart rate vs HRV vs ImmersionWhat each metric can and can’t tell you, with five worked moments.
- How do I measure emotion in my app?Five ways to do it. Four make you liable for something. One doesn’t.
- Immersion vs voice prosodyVoiceprint plus emotion recognition.
- Immersion vs HRVHealth data plus emotion recognition, without the science.
- Immersion vs surveysLawful, simple, and late.