Consumer Neuroscience Foundations

Evaluate study validity and neural-data scope separately. The behavioural templates below generate hypotheses to test with behavioural outcomes; a brain-region label does not validate a design move.

When to Use

• An agency or vendor pitches a neuromarketing study, an "engagement index", or a neural or biometric ad pre-test.
• A team reports a lab signal (GSR spike, alpha asymmetry, fixation heatmap, emotion-AI score) and wants to act on it.
• A product will capture EEG, fNIRS, HRV, GSR, eye movement, face or voice affect, and someone asks what law applies.
• Someone is designing a copy, framing or narrative test and needs a behavioural hypothesis template (see the table below).

Skip it and use the owner instead:

• Pricing, defaults, anchoring, dark-pattern and DMCC/CMA enforcement → foundations-behavioral-economics.
• Generic score or survey reliability and validity with no physiological signal → foundations-measurement-theory.
• Causal lift of a shipped change → foundations-causal-inference or an A/B test.
• Usability or attention research without biometrics → software-ux-research.

Workflow

1. Classify the ask: study purchase or evaluation, signal interpretation, neural-data law scope, or a behavioural design question (route the last to the templates below or to foundations-behavioral-economics).
2. For a study or vendor: apply the four-part buy rule and the quick reference, then the vendor questions in references/neuro-study-validity.md.
3. For any neural or biometric finding: fill the three evidence rows and refuse reverse inference.
4. For data capture: classify the signal (neural vs downstream) before choosing the legal gate.
5. Turn what survives into a behavioural hypothesis and an in-market test.

Evidence Contract

When a recommendation relies on a neural mechanism or biomarker, record three separate rows:

• mechanism evidence: the process exists.
• measurement evidence: this measure tracks that process in this task and population, at this n.
• product-effect evidence: the intervention changes the user outcome in the field.

A biomarker association does not establish the last two. Behaviour-first advice that makes no neural claim needs no neural rows. State the behavioural outcome, the test design and the causal limit instead. Full record, confound table and regression checks: references/evidence-to-product.md.

No reverse inference. Going from "region or signal X was active" to "users felt Y" needs evidence that X is selective for Y in this task (Poldrack 2011). Insula, DMN, mu-band suppression, N400, GSR and pupil dilation are all non-selective. Report the observation and test the behavioural hypothesis; do not name a mental state.

Should We Buy a Neuro Study?

Buy only if all four hold. Otherwise run a self-report pre-test plus an in-market holdout or A/B test.

1. Decision fit. The decision is between creative or stimulus variants that cannot be field-tested cheaply or quickly (for example TV cuts before a media buy). If an A/B test can answer it within the decision window, the A/B test is the ground truth.
2. Published reliability at your n. The primary metric has published test-retest or split-half reliability in a comparable ad task at the planned sample size. Generic claims ("EEG measures attention") do not count.
3. Out-of-sample validity of any composite. A vendor's proprietary index (engagement, emotion, "neuro score") must show validity against held-out market or behavioural outcomes, not only correlations inside the calibration set. No validation data means the index is untested and is not evidence.
4. Preregistered decision rule. The primary metric, n, exclusion rules and the threshold that changes the decision are written down before data collection.

Why this bar: head-to-head evidence is specific to its tested task. Venkatraman et al. (2015, JMR 52(4):436–452) compared several methods (self-report, implicit, eye-tracking, biometrics, EEG, fMRI; method count unsourced — verify) on 30-second TV ads against market-level ad elasticities. fMRI measures explained the most variance beyond traditional measures, and ventral striatum activity was the strongest predictor. That is one study, one ad format and a lab-to-market link, not a licence for any vendor's metric.

When NOT to buy: the choice is reversible and testable in-market; the vendor will not disclose reliability or validation data; the metric is frontal alpha asymmetry or a webcam emotion score used as the primary outcome; the sample is below the metric's published requirement; or the product-effect question is really a causal-lift question.

Detail, vendor-evaluation questions and reporting template: references/neuro-study-validity.md. Study protocol and n by modality: assets/playbooks/study-design.md.

Quick Reference: Metric Validity

Metric	What the evidence supports	Use as primary?
EEG inter-subject correlation (ISC)	Excellent between-sample reliability for video ads (van Diepen, Boksem & Smidts 2025). A cited 2025 BMC Psychology review of ISC and attention is unsourced — verify before relying on it	For video attention only with task-specific validation; still needs a behavioural outcome
EEG alpha / beta power	Rated good reliability in the same study	Secondary
Frontal alpha asymmetry (FAA)	Poor between-sample reliability in the same video-ad study; this does not establish preference or purchase validity	No. Secondary only, baseline-normalised, never to classify an individual
GSR / EDA, HRV	Autonomic activation; statistically distinct from subjective affective arousal (BAAS, Nat. Commun. 2025); valence unknown	No. Pair with a valence and a behavioural measure
Eye-tracking fixations	Where people looked; not liking, comprehension or intent	For attention allocation only
fMRI ventral striatum / NAcc, MPFC	Neuroforecasting of aggregate outcomes; which region generalises depends on the domain (Genevsky et al. 2025; Srirangarajan et al. 2026). A study's successful sample size is not a transferable minimum	Research partnership only
Facial coding / webcam emotion AI	A facial configuration is not proof of felt emotion; consumer tools vary widely in accuracy	No
ML classifier on neuro signals	Prediction accuracy does not establish mental-state construct validity; participant or stimulus leakage can inflate it	Only with independent cross-study validation and an independently justified target label

Sample size: the van Diepen et al. abstract reports that repeated viewings improved reliability and that 30–40 participants were needed for most metrics to assess an ad overall; temporal tracking needed more. Read the full text for the metric, number of viewings and reliability threshold before using a minimum.

Neural-Data Law Scope

Before capture, check the current state and EU neural-data statutes: read each legislature's enacted text and amendments for its neural-data definition, consent or limitation rights, entity exemptions and effective date, using a state-privacy tracker to locate them. Scope distinctions that stay stable: the statute's definition (CNS-only versus CNS and PNS), exclusions for signals inferred from nonneural information, and purpose limits.

• Practical consequence: assess EEG against each statute's definition and purpose limits. Do not automatically classify fNIRS as direct neural measurement: it measures haemodynamic changes indirectly, so inference and nonneural-information exclusions need counsel's analysis. For GSR, HRV, eye-tracking, facial coding and voice affect, assess downstream-signal exclusions separately. Then check ordinary personal-data, health-data and biometric rules by jurisdiction and purpose; these signals do not all automatically qualify under BIPA or GDPR Art 9. Record the lawful basis and any consent required even where a neural-data statute does not apply.

EU and UK. Classify the purpose first. Sensor capture alone does not make a system emotion recognition, biometric categorisation or high-risk. Workplace or education emotion inference raises AI Act Art 5(1)(f); check the medical or safety exception. Before assigning a compliance deadline, consult the European Commission's AI Act timeline, distinguishing Art 5, Art 50 and high-risk obligations. The classification checklist is in references/ethics-operational-checklist.md. Under GDPR, physiology is not automatically special-category data: record an Art 6 basis and assess whether health inference or unique identification brings in Art 9. DMCC and CMA choice-architecture enforcement is owned by foundations-behavioral-economics. Route final classification to qualified counsel; when source access is unavailable, report the affected classification or deadline as unverified.

Behavioural Hypothesis Templates

Each template gives a definition, misuse boundary, inputs and outputs, failure modes and a worked example. Use them to write a testable behavioural hypothesis. Neural background is context only.

When choosing or combining templates, load the composition guide for the shared misuse boundaries and recipe selection before opening the selected playbooks.

#	Template	Behavioural question to test	Evidence caution
1	Attention & Salience	Does the priority element get seen and acted on?	First fixation ≠ liking
2	Arousal Physiology	Does intensity help or hurt task completion and comfort?	GSR ≠ felt arousal; Yerkes-Dodson is not a design law
3	Social Bonding	Do real care signals change perceived care and retention?	The registered replication found no main oxytocin effect under its tested conditions (Declerck et al. 2020)
4	Narrative Transportation	Does a story format change transportation score, recall and attitude?	Measure transportation by scale (Green & Brock 2000; van Laer et al. 2014), not DMN
5	Regulatory Focus & Fit	Does message–goal fit beat a mismatched frame for this decision?	Regulatory focus ≠ BIS/BAS; referral source is not a classifier
6	Mirror Systems & Emotional Contagion	Do verified testimonials with faces beat none?	Mu suppression is not an MNS readout
7	Neuroaesthetics	Does the visual variant change preference and credibility?	Polish without delivery backfires
8	Interoception & Somatic Markers	Does body-state framing inform, or does it pressure?	Never manufacture somatic urgency
9	Memory Consolidation	Does timing or spacing change recall and retention?	No clock-defined consolidation window
10	Reward Anticipation	Do reveals raise wanting without lowering liking?	Measure wanting and liking separately; cap loops
11	Embodied Cognition	Does a congruent metaphor improve comprehension?	Social-priming demos failed replication
12	Predictive Processing	Does format consistency or a primed reveal reduce errors and churn?	No "surprise budget" can be measured in UX

Multi-template recipes (anxiety-relief journey, reading bond, daily cadence, mixed-audience landing page, trust repair, ethics audit, AI assistant UX) and the misuse-boundary table: references/composition-recipes.md.

Harm Test

A technique is legitimate only if it:

1. steers users towards outcomes they would endorse on reflection;
2. can be easily overridden or opted out of;
3. does not rely on concealment, manufactured signals (fake urgency, fake warmth, fake social proof) or exploitation of vulnerability.

Wellness, anxiety, spiritual and financial-distress audiences get the stricter default: No, unless there is a documented harm-test pass. Operational gates (pre-study, pre-deployment, US neural-data consent, vulnerable-user bans): references/ethics-operational-checklist.md.

Anti-Patterns

Anti-pattern	Why it fails	Instead
"Region X lit up, so users felt Y"	Reverse inference without selectivity	Report the signal; test the behaviour
FAA or a webcam emotion score as the decision metric	Poor or unvalidated reliability	ISC or behavioural outcome as primary
Buying a vendor composite index on face value	No out-of-sample validity shown	Demand holdout validation, or do not buy
Neuro study instead of an A/B test for a testable choice	Lab-to-field gap; small n	A/B test is ground truth; neuro can only pre-screen
Consumer EEG headband reported like research-grade ERPs	Hardware and processing can distort signals and change reliability	Require metric-specific validation on the proposed hardware; ISC and broad bands are not automatic exemptions
Personalising by referral source as a "BIS/BAS" proxy	Channel carries intent and traffic-quality confounds	Test message–goal fit directly
"Oxytocin-driven trust" copy	No replicated main effect	Real care mechanics; measure perceived care
Notifications timed to a "consolidation window"	No clock window is established	User-selected timing, quiet hours, randomised cohorts
Treating GSR, eye or face data as outside all privacy law	Out of neural-data statutes does not mean unregulated	General biometric, sensitive-data and GDPR analysis
Assuming a hormone or consolidation timer sets an interface deadline	Neural response timing and hormone kinetics do not set UX deadlines — there is no cortisol, oxytocin or consolidation timer for UX	Use measured behavioral timing, not a claimed neural window

More traps with sources (consumer-EEG validity, ML-on-neuro overfit, social priming): references/patterns-scenarios-traps.md.

Navigation

File	Read when
references/neuro-study-validity.md	Evaluating a vendor, pitch or completed neuro study
references/evidence-to-product.md	Turning any neural or biometric finding into a product recommendation
references/biomarker-signal-dictionary.md	Interpreting a specific signal: what it can and cannot index, confounds, misreads
references/ethics-operational-checklist.md	Before any study or launch that captures physiological data; neural-data statute lookup
references/instrumentation-vendor-landscape.md	Choosing a modality: signal reference and decision tree (no vendor list; confirm vendor terms directly)
assets/playbooks/study-design.md	Designing a study: n by modality, within/between, preregistration
references/composition-recipes.md	Combining behavioural templates into a flow; misuse boundaries
assets/playbooks/signal-to-design-cookbook.md	"We observed X": behavioural hypotheses and verification
references/frameworks-meta.md	Choosing an organising frame; anti-frameworks (triune brain, "buy button")
references/formal-theory-map.md, references/primitives-overview.md	Theory background and limits per template
data/sources.json	Primary sources

• Key sources: Poldrack 2011 (reverse inference); Venkatraman et al. 2015; van Diepen et al. 2025; Genevsky et al. 2025; Srirangarajan et al. 2026; BAAS 2025; Declerck et al. 2020, Nat Hum Behav 4(6):646–655; Higgins 1997, 2000; Motyka et al. 2014; Green & Brock 2000; van Laer et al. 2014.

Related Skills

• foundations-behavioral-economics: framing, defaults, reinforcement schedules, DMCC and dark-pattern enforcement.
• foundations-measurement-theory: reliability and validity of any score or scale.
• foundations-causal-inference: field lift and holdouts.
• marketing-cro, marketing-paid-advertising, marketing-content-strategy: apply templates 1, 4 and 5 through their desire-segmentation.md references.
• software-ui-ux-design, software-ux-research, product-management: attention, consistency and reward-loop hypotheses.

Learnings Loop

When prior decisions or pitfalls are relevant, consult learnings.consolidated.md if present; use learnings.md only for needed history or as the available fallback. Otherwise skip both.

After applying it, if you encountered a pattern worth remembering, a mistake worth preventing, or a domain fact that surprised you, append one dated bullet to learnings.md via agents-skills-feedback-loop/scripts/append_learning.py. Do not modify SKILL.md itself.