foundations-consumer-neuroscience

Consumer Neuroscience Foundations

12 canonical consumer-neuroscience primitives for product, content, interface, and retention design. Each primitive is domain-agnostic and ethically bounded. Primitives 1–8 cover engagement-time neural responses (salience, arousal, bonding, narrative, regulatory orientation, social mirroring, aesthetics, interoception). Primitives 9–12 cover temporal and predictive mechanisms (memory consolidation, reward anticipation, embodied cognition, predictive processing). Primitive #10 (reward anticipation, Berridge "wanting" vs "liking") is intentionally distinct from foundations-behavioral-economics primitive #13 (reinforcement schedules / dopamine prediction-error): that skill covers schedule-of-reinforcement design; this skill covers anticipatory dopamine as a separate design lever — countdown UX, drop reveals, daily-card open, pre-purchase excitement. Primitive #12 (predictive processing & active inference) is the unifying primitive that grounds attention (#1), interoception (#8), and narrative (#4) under one prediction-error-minimization frame: the brain continuously generates predictions; violations of priors incur a prediction-error cost that must be "earned" by the design.

Ethical obligation: every primitive in this skill operates on pre-conscious or sub-deliberative neural systems. The manipulation risk is higher than for behavioral-economics nudges, because users cannot easily introspect on the mechanism. Read the Misuse Boundary subsection in each playbook before applying any technique. The test from Thaler and Sunstein: "Would you be embarrassed if the technique appeared on the front page of a newspaper?" If yes, it is exploitation, not design. The DMCC Act 2024, in force from 6 April 2025, makes online choice architecture and dark patterns directly actionable by the CMA with fines up to 10% of global annual turnover.

When to Apply

Apply consumer-neuroscience when:

• Attention/salience design — first-7-second hook, visual hierarchy, modal vs inline
• Anxiety-driven engagement loops (cosmic, dating, status apps) — needs DMCC ethical audit
• Parasocial / narrative-led conversion (creator content, branded characters)
• Daily-cadence retention with timing-sensitive triggers (consolidation windows, wake-time)
• Trust repair, reciprocity, or oxytocin-bond design in social/community products

Skip and use simpler alternatives when:

• Pure pricing/defaults/anchoring question — foundations-behavioral-economics is sufficient and cheaper
• Audience has no measured anxiety/arousal/attention baseline — neuro framing is decoration, not insight
• B2B SaaS with rational-buyer mode dominant — emotional primitives mostly noise; use behavioral-econ + decision-theory
• The proposed mechanism manipulates without genuine user benefit — fails DMCC Act 2024 ethical gate; do not ship
• Required signals (eye-tracking, GSR, fMRI) aren't available AND no biomarker proxy exists — claim is unfalsifiable
• Causal lift question — use foundations-causal-inference to measure; neuro primitives suggest mechanisms, not effect sizes

Contents

• Quick Reference
• Primitive Index
• Formal Supporting Theory
• Ethical Bounds
• Misuse Boundaries
• Decision Checklist
• Anti-Patterns
• Composition Recipes
• Knowledge Base & Operational Guides
• Workflow
• ASCII Flow
• Navigation
• Related Skills
• Fact-Checking

Quick Reference

Primitive Index

Each primitive has a full playbook: Definition / When to use / Misuse boundary / Inputs / Outputs / Failure modes / Worked example / Sources.

Formal Supporting Theory

Use references/formal-theory-map.md when the task needs source assumptions, ethical boundaries, or a distinction between observed neural response and normative welfare.

Ethical Bounds

The Harm Test

A neural design technique is legitimate if it:

1. Steers users toward experiences or decisions they would endorse on reflection.
2. Can be easily overridden or opted out of.
3. Does not exploit pre-conscious neural mechanisms to act against the user's interests.

The same lever — arousal, oxytocin warmth, reward anticipation — can be legitimate or manipulative depending on whether the underlying offer genuinely serves the user.

Manipulation vs Legitimate Design

UK Regulatory Context (August 2026)

DMCC Act 2024 entered into force 6 April 2025, revoking the CPRs 2008 outright (s.251(1), commenced by SI 2025/272) and succeeding them with ss. 226 (misleading actions), 227 (misleading omissions), and 228 (aggressive practices), plus the Sch. 20 list of banned practices. The successor provisions are redrafted, not a restatement — old CPRs regulation numbers do not map across cleanly, so cite DMCC sections. The CMA has direct civil-enforcement power and can fine up to 10% of global annual turnover without requiring a court order.

Enforcement is now live, not prospective — the first two infringement decisions both concerned online choice architecture rather than advertising content:

• 18 November 2025: CMA opened its first DMCC enforcement actions against 8 firms (drip pricing, default opt-ins, pressure selling) and issued approximately 100 advisory letters across 14 sectors.
• 18 June 2026: second infringement decision — Marks Electrical fined £720,000 (£1.2m reduced 40% for early settlement) and ordered to refund ~£600,000 to ~40,000 customers, for pre-selected extra charges (customers auto-opted into paid recycling and unwrapping services). Conduct covered April–November 2025. This is the clearest signal of the enforcement floor: a mid-size retailer, a single default-opt-in pattern, a seven-figure headline penalty plus consumer redress.

April 2025: CMA published procedural guidance on DMCC enforcement. December 2025: CMA published price transparency guidance under DMCC.

Online Choice Architecture (dark patterns) now directly actionable under DMCC, including:

• Confirm-shaming (manipulative framing on decline options)
• Pre-ticked defaults that benefit the operator at user expense
• Drip pricing (incremental price reveal late in purchase flow)
• False urgency ("Only 2 left!" when stock is unconstrained)
• Forced continuity (auto-renew without prominent disclosure)

Secondary regulatory anchors:

• ASA CAP Code: misleading advertising, fabricated testimonials, manufactured social proof
• DMCC Act 2024 s.228 (aggressive practices — replaced CPRs 2008 Reg. 7, revoked 6 April 2025)
• UK GDPR: biometric and neuro-physiological signal capture (GSR, HRV, eye-tracking, fNIRS) constitutes special-category data in many use cases; requires explicit consent and lawful basis (Article 9)

EU Regulatory Context (August 2026)

For products serving EU users, the EU AI Act is the parallel anchor to DMCC and applies on top of GDPR.

• Article 5 prohibitions in force from 2 February 2025: AI systems that deploy "subliminal techniques beyond a person's consciousness" or "purposefully manipulative or deceptive techniques" causing significant harm are prohibited outright. AI systems that exploit vulnerabilities (age, disability, socio-economic situation) are also prohibited. This directly captures the manipulation column of the table above when AI is in the loop. Unaffected by the 2026 delay below — the prohibitions bind now.
• Emotion-recognition prohibition (workplace and education): AI inference of emotions from facial expression, voice, GSR, HRV, or any biometric stream is prohibited in workplace and education contexts (Article 5). Commercial deployment outside those contexts is not prohibited but is regulated.
• High-risk classification DELAYED to 2 December 2027 (was 2 August 2026): the AI Digital Omnibus was published in the Official Journal 24 July 2026 and entered into force 27 July 2026, deferring standalone Annex III high-risk obligations — which include commercial emotion-recognition and biometric-categorisation systems — by 16 months. Annex I (AI embedded in products under EU product-safety law) moves to 2 August 2028. Providers and deployers must still meet data-governance, transparency, human-oversight, robustness, accuracy, and post-market monitoring requirements, but the compliance deadline is December 2027. Treat this as schedule relief, not repeal: systems in design now will ship into the regime.
• Article 50 transparency obligations remain on the original 2 August 2026 schedule — they were not delayed. Users exposed to emotion-recognition or biometric-categorisation systems must be explicitly informed, now. A four-month grace period (to 2 December 2026) applies only to the Article 50(2) watermarking duty for systems already on the market. This is the live EU obligation for affect-inference products as of August 2026.
• GDPR continues to apply: lawful basis (typically Article 9 explicit consent for biometric data) is a precondition; the AI Act adds requirements on top. GDPR is unaffected by the Omnibus delay and is the binding constraint in the interim.

For UK-only products, DMCC + UK GDPR are sufficient. For EU users or shared-stack products, both regimes apply and the stricter rule binds.

US Regulatory Context (August 2026)

Four states have enacted neural-data-specific privacy laws (Colorado, California, Montana, Connecticut), and nine further bills were introduced across six states in the first six weeks of 2026 alone (Alabama, California, Illinois, New York, Vermont, Virginia). Treat this as a live patchwork, not a settled regime.

Scope caution — these laws are narrower than "any biometric signal." Most define neural data as signals from the nervous system measured directly, and several explicitly exclude the downstream physiological signals this skill most often uses. Montana SB 163 carves out "nonneural information … the downstream physical effects of neural activity, including but not limited to pupil dilation, motor activity, and breathing rate" — which excludes GSR and eye-tracking. California SB 1223 requires neural data be "not inferred from nonneural information," likely excluding facial coding and voice affect. Colorado's definition reaches only data used for identification purposes, excluding most consumer applications. Connecticut has no explicit carve-out, leaving GSR and eye-tracking ambiguous there. Practical consequence: EEG and fNIRS are squarely in scope; GSR, HRV, eye-tracking, facial coding, and voice affect are mostly out of neural-data statutes — but remain covered by general state biometric/sensitive-data law, BIPA-style statutes, and GDPR for EU users. Do not use a neural-data-law exemption as a reason to skip consent; check the general privacy regime instead.

• California SB 1223 (effective 1 January 2025): amends CCPA to classify "neural data" (signals from central or peripheral nervous system, not inferred from nonneural information) as sensitive personal information. Opt-in consent required; right to delete and restrict sharing apply. Primary source
• Colorado HB 24-1058 (effective 7 August 2024): amends Colorado Privacy Act to include "neural data" within "biological data" as sensitive data. First US law to define and protect neural data. Scope limited to data used or intended for identification. Primary source
• Montana SB 163 (effective 1 October 2025): adds neurotechnology data to Montana's Genetic Information Privacy Act. The most extensive of the four: detailed express-consent requirements for collection, marketing and research use, disclosure, transfer, and sale — often requiring separate informed consent per purpose and per third party. Explicitly excludes nonneural downstream signals. If a product captures true neural data from US users, Montana sets the strictest operative bar.
• Connecticut SB 1295 (signed 24 June 2025; effective 1 July 2026): amends CTDPA to add neural data as a sensitive data category; processing requires express consumer consent; selling sensitive data without consent prohibited. Primary source
• Vermont H.814 / Act 101 (signed 18 May 2026; effective 1 July 2026): correction — do not overstate this law. As enacted, H.814 was substantially narrowed in the Senate: it recognises a largely declaratory statement of "neurological rights" (mental privacy, freedom of thought, non-discrimination in neurotechnology), but the consent requirement and private right of action were stripped before passage. Enforcement rests exclusively with the Vermont Attorney General; there is no consent gate for businesses. Its main forward hook is a commissioned study reporting to the next legislative session. Vermont's binding neural-data framework is Vermont S.71 (neural data as sensitive data), effective 1 January 2028 — track that bill, not H.814, for compliance planning. Treat H.814 as a signal of legislative direction, not a live consent gate. Primary source
• UNESCO Recommendation on the Ethics of Neurotechnology (adopted 12 November 2025): first global non-binding framework covering neural data across commercial uses. Non-binding but widely cited in board-level compliance discussions and DPA engagement. Primary source
• US MIND Act 2025 (proposed): would direct FTC to study neuromarketing as a named use case; not yet law but signals federal regulatory attention. Document FTC-readiness posture if product involves neuromarketing explicitly.

Practical implication: any product capturing genuine neural signal (EEG, fNIRS) from US users must run a per-state consent analysis — California CCPA sensitive PI from 1 January 2025, Montana's per-purpose express consent from 1 October 2025, Colorado and Connecticut in parallel. For GSR, HRV, eye-tracking, facial coding, and voice affect, the neural-data statutes mostly do not bite; the governing constraints are general sensitive-data and biometric law plus GDPR Article 9 for EU users. See references/ethics-operational-checklist.md US Neural Data Laws section.

Vulnerable-User Note

CMA enforcement priorities specifically name "aggressive sales practices which take advantage of vulnerability." Wellness, anxiety-relief, and astrology/spiritual audiences are explicitly in scope as vulnerability-risk contexts. EU AI Act Article 5 reinforces this with an outright prohibition on AI systems that exploit vulnerabilities of specific groups (age, disability, socio-economic situation) to materially distort behaviour. Any application in these categories must apply the stricter column of the manipulation table — not the middle ground. Manufactured urgency, oxytocin-proxy warmth without genuine care mechanics, and reward-anticipation loops targeting financially or emotionally vulnerable users are highest-risk under both regimes.

Misuse Boundaries

Check references/patterns-scenarios-traps.md before applying primitives to production user flows.

Decision Checklist

• Attention earned: Is the design earning attention through genuine relevance or novelty, not bottom-up hijacking? → attention & salience (#1)
• Arousal calibration: Is the engagement intensity appropriate for the decision being made? Will the arousal level impair or support the user's goal? → arousal physiology (#2)
• Warmth signals: Are trust and affiliation signals real? Is any warmth mechanic backed by genuine social data? → social bonding (#3)
• Narrative accuracy: If the experience transports users emotionally, is the content accurate? Does immersion serve or obscure the user's interests? → narrative transportation (#4)
• Regulatory orientation: Does the audience skew BIS (prevention) or BAS (approach)? Is the primary message tone matched to the audience's dominant orientation? → approach-avoidance (#5)
• Social proof quality: Are testimonials, reactions, and contagion signals from real users in verified data? → mirror systems (#6)
• Aesthetic-to-delivery ratio: Does visual quality match functional delivery? Is aesthetic reward being used to compensate for a weak product? → neuroaesthetics (#7), reward anticipation (#10)
• Interoceptive framing: Is any body-state or "gut feel" framing based on real signals? Is it used to inform, not to manufacture anxiety? → interoception (#8)
• Consolidation timing: Are push notifications and reminders timed to consolidation windows (evening, post-sleep) rather than maximum interruptibility? → memory consolidation (#9)
• Wanting vs liking balance: Is reward anticipation matched by hedonic payoff? Is the anticipation loop capped to prevent compulsion? → reward anticipation (#10)
• Embodied language: Does copy use body-state metaphors congruent with the product experience? → embodied cognition (#11)
• Prediction-error budget: Does the design surprise users only when it has earned the attentional cost? Are established priors preserved during routine use? → predictive processing (#12)
• Ethical gate: Does each technique pass the harm test? Does it survive DMCC scrutiny for vulnerable-user contexts? → ethical bounds section

Anti-Patterns

Composition Recipes

Recipe 1: Anxiety-Relief Consumer Loop (pre-purchase)

Goal: guide an anxiety-experiencing user through a reassurance journey to a confident purchase decision, without manufacturing or amplifying anxiety.

Stack:

1. Arousal physiology (#2): Detect or assume elevated arousal state (wellness/anxiety audience). Design the entry experience to begin deescalating arousal — calm visual pacing, low-contrast background, short sentence length. Do not spike arousal at entry.
2. Predictive processing (#12): Establish clear product-structure priors immediately. Anxious users have a high prediction-error cost; predictability is reassurance. Consistent layout, no hidden elements.
3. Narrative transportation (#4): Use a "person like me" story (brief, first-person, past-tense) in which anxiety was the starting state and resolution was the outcome. DMN engagement with a self-relevant arc reduces threat appraisal.
4. Social bonding (#3): Introduce real human warmth — a named support person, a real community count, a genuine care statement backed by operational reality (response time, refund policy). Oxytocin half-life ~3–5 min; warmth must be re-encountered across the session, not front-loaded only.
5. Interoception (#8): Close with a body-state check cue ("How do you feel right now?") that invites somatic attention; let the user register their own shift. This is the somatic marker that encodes the product association positively.

Ethical-bound check: The anxiety being relieved must be real. Do not manufacture anxiety (#2 misuse) to then relieve it. DMCC vulnerable-user test must pass: would the CMA say this practice takes advantage of vulnerability?

Fail signal: "felt scammed" or "felt manipulated" qualitative reports; CSAT drop post-purchase; CMA/ASA complaint volume rising.

Inputs: Baseline anxiety trigger (product category, entry surface, referral source); relief mechanism (narrative arc, warmth signal, somatic check-in); time-to-relief target (default: ≤90s from entry to perceived deescalation); audience retention metric (Day-7 and Day-30 re-engagement rate); persona arousal profile (high-BIS prevention-dominant vs. moderate arousal). Rules: Relief must be initiated within 90s of entry trigger — cortisol arousal curves peak and begin recovery in this window; delay beyond 90s risks entrenchment. Avoid intermittent reinforcement schedules in the relief journey (no random resolution timing) — variable-ratio schedules for an anxiety audience create compulsive re-checking, not relief. Ethical gate: relief must address a genuine user need; manufactured anxiety to then relieve it fails the DMCC harm test and the EU AI Act Article 5 prohibition on exploiting vulnerabilities. Outputs: Trigger-to-relief interaction sequence (step-by-step UX flow with timing); measurable anxiety reduction signal (PSS-style 1–5 self-report at session close, target mean shift ≥1 point); ethical pass/fail flag (CMA vulnerable-user test + DMCC harm test result documented before ship).

Recipe 2: Parasocial Reading Bond (purchase)

Goal: generate a genuine reading bond between user and content (horoscope, tarot, interpretive reading) that drives purchase and repeat engagement without deception.

Stack:

1. Narrative transportation (#4): Open with a brief orienting narrative that primes the DMN. The reading itself should use second-person, present-tense framing to maximize self-referential processing in vmPFC.
2. Mirror systems (#6): Include at least one face or depicted emotional state that matches the emotion the user is likely experiencing. FFA activation and MNS simulation generate social presence with a non-present author.
3. Social bonding (#3): "Others who received this reading reported..." — real cohort social framing; affiliative warmth through shared experience, not manufactured intimacy.
4. Embodied cognition (#11): Copy uses body-state metaphors grounded in the product's actual experience ("a weight lifts," "clarity settles in") — not random metaphors.

Ethical-bound check: Content accuracy: predictive or interpretive content must be labeled as such (ASA CAP Code; no false claims of scientific accuracy). Social data must be real. Face imagery must be genuine or clearly illustrative.

Fail signal: Low share rate despite high session time — narrative bond did not activate social-contagion desire; revisit mirror system (#6) and real social proof (#3).

Inputs: Content type (horoscope, tarot, interpretive reading, personalized narrative); audience persona arousal profile (DMN engagement baseline — emotionally primed vs. neutral); social proof artifacts available (verified cohort testimonials, face assets, share-rate data from prior sessions); notification timing options (session-start cue window, post-read follow-up timing). Rules: Second-person present-tense framing required for vmPFC self-referential processing; first-person past-tense for embedded "person like me" social proof arc. Face or emotional-state imagery must match the target resolution emotion (relief, clarity, confidence) — mismatched affect in testimonials suppresses MNS simulation. All cohort framing ("others who received this reading...") must use verified real user data — fabricated social proof is deception under DMCC and ASA CAP Code. Content labeled predictive or interpretive, never factual-scientific. Outputs: Parasocial bond sequence (narrative arc + social proof placement + embodied metaphor copy); share-intent signal (post-session share prompt acceptance rate, target ≥15% of completers); cohort split result (high-DMN engagement vs. baseline by content format tested).

Recipe 3: Daily-Cadence Retention (post-purchase)

Goal: build a voluntary daily engagement habit that the user values, without compulsion design.

Stack:

1. Reward anticipation (#10): Design a daily reveal or unlock that creates genuine wanting — a named card, a daily insight, a progress update. VTA dopamine onset ~200ms before reveal; the anticipation, not just the content, is the engagement driver. Cap the chain; provide explicit completion signals.
2. Memory consolidation (#9): Time the daily cue to early evening (consolidation window onset) or morning (post-sleep memory freshness). Measure Day-7 and Day-30 retention as a function of notification timing cohort.
3. Attention & salience (#1): The daily notification must use top-down salience (user-relevant, personalized, named) rather than bottom-up salience (loud, high-contrast interruption). Bottom-up salience for a recurring cue trains the user to dismiss it.
4. Predictive processing (#12): Maintain strong format consistency across daily units. Prediction satisfaction — the cue arriving as expected, in expected form — is itself rewarding. Reserve genuine novelty for special events.

Ethical-bound check: Wanting loop must have a ceiling. Consolidation-window timing must not interrupt sleep. Notifications must be easy to disable (DMCC reversibility test).

Fail signal: Streak completion rate high but re-engagement intent (next-session survey) is low — user is mechanically completing a streak, not experiencing genuine wanting; wanting loop has decoupled from liking.

Inputs: Content type for daily reveal (named card, daily insight, progress update, personalized reading); notification timing options (early evening 7–9pm consolidation window OR morning within 30min post-wake); audience baseline retention (Day-7 and Day-30 cohort rates from prior releases); persona arousal profile (high-anticipation-seeking vs. routine-preference dominant). Rules: Notification must land in the evening 7–9pm consolidation onset window OR within 30 minutes of wake — consolidation replay begins during these windows; notifications outside them fragment encoding and increase negative product association risk. Format consistency ≥90% across daily units — prediction satisfaction from format conformity is itself rewarding; reserve genuine novelty for designated special events only. Wanting-loop cap: ≤7 consecutive daily unlocks before an explicit rest or completion signal; unbounded chains exceed the misuse boundary for reward anticipation (#10). Notifications must be trivially disable-able (DMCC reversibility test). Outputs: Cohort split by notification timing window (evening vs. morning vs. control) with Day-7 and Day-30 retention rates; wanting-loop cap event count (number of times the rest signal was triggered per user per month, target ≥1 to confirm cap is active); ethical pass/fail flag (satiation signal documented in design spec, rate-cap verified before ship).

Recipe 4: Conversion Landing Page, Mixed Audience (pre-purchase)

Goal: maximize conversion across a mixed BAS (promotion-seeking) and BIS (prevention-vigilant) audience without a single-tone funnel.

Stack:

1. Attention & salience (#1): Above-the-fold uses top-down salience cues (problem statement that matches user prior, personal pronoun "you"). No bottom-up salience noise at entry.
2. Approach-avoidance / BIS/BAS (#5): Headline A/B: promotion frame ("Unlock daily clarity") vs prevention frame ("Never miss an important day again"). BAS users convert on approach; BIS users convert on prevention. Test or personalize by referral source signal.
3. Neuroaesthetics (#7): Visual design uses peak-shift on key differentiating visual element; symmetry in layout; color palette empirically associated with target emotional register (calm, warmth, or energy — product-appropriate). Aesthetic reward at first glance reduces the cognitive cost of reading on.
4. Mirror systems (#6): Testimonials use real face + real emotional expression matching the resolution state (relief, clarity, confidence). MNS simulation must match the emotion, not just any positive face.

Ethical-bound check: BIS prevention framing must not manufacture threat. Testimonials must be real. Aesthetic quality must be matched by functional delivery.

Fail signal: Bounce concentrated in one regulatory-focus segment — BIS vs BAS mis-match; check copy tone against BIS/BAS segmentation data.

Inputs: Audience BIS/BAS split estimate (referral-source signal, prior copy-test data, or assumed 50/50 if unknown); social proof artifacts (real verified testimonials with face imagery and documented resolution emotion); notification timing options (above-the-fold entry cue — not applicable here, but post-visit retargeting timing if used); persona arousal profile (calm-landing vs. high-energy entry). Rules: Headline A/B mandatory: promotion frame ("Unlock daily clarity") for BAS segment; prevention frame ("Never miss an important day again") for BIS segment — do not ship single-tone without segmentation signal. Testimonial face imagery must match the target resolution emotion (relief, clarity, confidence), not a generic positive affect — MNS simulation fires on emotion congruence, not valence alone. BIS prevention framing must not manufacture threat or false urgency (DMCC false-urgency prohibition). Aesthetic quality must be matched by functional delivery — aesthetic reward without functional payoff amplifies prediction error on first real product contact. Outputs: A/B copy variant results (BAS-frame vs. BIS-frame conversion rate by segment); cohort split by referral-source BIS/BAS proxy; ethical pass/fail flag (testimonials verified, no manufactured urgency, aesthetic-to-delivery ratio documented).

Recipe 5: Trust Repair After Error (post-purchase)

Goal: restore trust after a product error or service failure without manipulating the user into false forgiveness.

Stack:

1. Social bonding (#3): Acknowledge the failure with a named human voice, not a system message. Oxytocin affiliative response requires social presence; automated impersonation of warmth makes trust repair harder. Real person acknowledgment.
2. Interoception (#8): Invite the user to describe their experience before offering a resolution. Somatic marker theory: the user's decision to continue is encoded in body state; helping them articulate and feel heard changes the somatic marker from threat to acknowledgment.
3. Predictive processing (#12): Provide an explicit account of what failed and what changed. The violation was a prediction error; closing it requires a new, more reliable prior — not just an apology, but a systemic explanation that earns a revised trust prior.
4. Approach-avoidance / BIS (#5): BIS-dominant users in error contexts are running prevention-mode appraisals; frame resolution in prevention terms ("We've put a safeguard in place so this cannot recur") not only gain terms ("Here's what you get now").

Ethical-bound check: Do not use warmth signals to gloss over a genuine product failure without fixing the underlying issue. The repair must be real.

Fail signal: NPS recovery less than 50% of pre-incident baseline within 30 days — warmth signals are not backed by operational repair; user's somatic marker has not shifted.

Inputs: Touchpoint sequence available for repair (named human outreach, system message, in-app banner, email); social proof artifacts (real support agent identity, documented operational fix); audience baseline retention pre-incident (NPS and Day-30 cohort rate); persona arousal profile (post-incident elevated-BIS state assumed for all users). Rules: First contact must be a named human voice, not a system message — oxytocin affiliative response requires social presence; automated warmth impersonation accelerates trust destruction, not repair. Response within 200ms in chat or equivalent synchronous channel (eye-contact analogue for digital trust); email within 4h of incident detection. Reciprocity ratio ≥1:1: the repair gesture must match or exceed the severity of the failure in tangible terms (credit, fix, explanation), not in warmth language alone. Resolution framed in BIS prevention terms ("safeguard in place so this cannot recur") for the post-incident audience — all users in error context are running prevention-mode appraisals. Outputs: Touchpoint repair script (named human + explanation of failure + systemic fix statement + resolution offer); trust-decay timing measurement (NPS delta at 7 days and 30 days post-incident, target ≥50% recovery of pre-incident baseline by Day-30); ethical pass/fail flag (warmth backed by real operational fix verified, no warmth-without-repair pattern).

Recipe 6: DMCC Compliance Audit

Goal: confirm any recipe applying neuroscience primitives passes the DMCC Act 2024 harm test before shipping.

Stack:

1. Apply the Ethical Bounds harm test (three gates: user endorsement on reflection, easy reversal, no exploitation of pre-conscious mechanisms against user interests).
2. Check against Online Choice Architecture dark-pattern list: confirm-shaming? pre-ticked defaults? drip pricing? false urgency? forced continuity?
3. Vulnerable-user check: is the target audience in a wellness, anxiety, or financially sensitive context? If yes, apply the stricter manipulation-table column throughout.
4. Biometric/neuro capture check: does the design, research plan, or analytics pipeline capture GSR, HRV, eye-tracking, or facial EMG? If yes, confirm UK GDPR Article 9 lawful basis is documented before deployment.
5. Anticipation-loop cap check: does any wanting mechanic (#10) have explicit satiation signals and a rate-cap? Document the cap in the design spec.
6. Signal honesty check: are all arousal (#2), warmth (#3), urgency (#6 from behavioral-economics), and social-contagion (#6) signals verifiable and accurate?

Ethical-bound check: this recipe IS the harm-test. Fail signal: any "yes" on the dark-pattern list, any vulnerable-user trigger without stricter-side controls, any biometric capture without Article 9 basis.

Inputs: Design spec or recipe output under audit (UX flow, copy variants, notification timing plan, anticipation-loop design); content type and audience context (wellness/anxiety/spiritual = vulnerable-user flag triggered); biometric/neuro-signal capture plan if any (GSR, HRV, eye-tracking, facial EMG); notification timing options as specified in the design. Rules: All six audit gates must be run sequentially — harm test → dark-pattern list → vulnerable-user check → biometric lawful-basis check → anticipation-loop cap check → signal honesty check. A single "fail" on any gate blocks ship. Wanting-loop cap (≤7 consecutive unlocks) must be explicitly documented in the design spec, not assumed. For EU-market products using AI-driven affect inference, confirm the Article 50 transparency notice is live now (in force 2 August 2026) and Annex III high-risk readiness is on track for the deferred 2 December 2027 deadline; Article 5 prohibition check mandatory for vulnerable-audience products. Outputs: Audit pass/fail flag per gate (6 gates, each documented with evidence); ethical fail flag (binary: ship-blocked or ship-cleared); remediation list if any gate fails (specific design change required, owner, and re-audit trigger).

Recipe 7: Attention-Aware AI Assistant UX (app-builder)

Goal: design an AI assistant or conversational product (chatbot, copilot, AI companion) whose output cadence, response framing, and notification behavior earn and conserve user attention — without exploiting pre-conscious mechanisms.

Stack:

1. Predictive processing (#12): Establish a consistent response format prior early. Users build a generative model of how the assistant responds; violating that model (unexpected length shifts, sudden tone changes, unexplained refusals) incurs prediction-error cost that degrades trust. Reserve format novelty for high-value reveals only.
2. Attention & salience (#1): Top-down salience dominates in AI UX — relevance to the user's stated task, not motion or contrast. Avoid decorative animations, status spinners with no informational value, or unsolicited agent proactivity that charges attentional cost without earning it.
3. Arousal calibration (#2): Match response pacing to the cognitive-load state of the task. For high-stakes or complex tasks, reduce sentence length and information density per turn; for exploratory or creative tasks, the user's arousal optimum is higher — match it. Do not sustain high information density across the entire session arc.
4. Memory consolidation (#9): If the product includes proactive reminders or scheduled summaries, time them to early evening or morning (consolidation onset windows). Assistant check-ins sent at midnight interrupt NREM replay and create negative product association.
5. Reward anticipation (#10): For AI products with a reveal arc (agent completing a long task, generating a final output, progressive report building), preserve the anticipation phase — progress signaling before the reveal earns the VTA dopamine onset. Do not drop the final output silently; surface the completion as a named event.

Ethical-bound check: AI assistants must not use proactive nudges, tone modulation, or pacing manipulation to manufacture dependency or increase session frequency beyond the user's own goals. Any product capturing voice arousal or facial expression for adaptive response owes an EU AI Act Article 50 notice now, and must be Annex III high-risk compliant by 2 December 2027. Proactivity requires consent — users must be able to silence or reconfigure assistant-initiated contact with one step (DMCC reversibility test).

Fail signal: Session-length metrics rise but task-completion satisfaction drops — the assistant is holding attention without delivering value; attentional debt is accumulating. Qualitative signal: "the assistant feels pushy" or "I feel like I can't stop."

Inputs: Task cognitive-load profile (high-stakes decision vs. exploratory vs. creative); response format prior established in onboarding; proactive notification plan (timing, cadence, opt-out path); any affect-inference capability in the assistant pipeline (voice, facial, text tone). Rules: Consistent response format for ≥90% of turns — prior satisfaction is inherently rewarding (#12); reserve structural novelty for explicitly flagged "new feature" or "important update" moments. Proactive nudges must be trivially disable-able (DMCC reversibility test). If assistant captures voice or facial signal for affect inference: Article 50 user notice required before capture (live since 2 August 2026); Annex III high-risk obligations apply from 2 December 2027. No manufactured urgency in assistant-initiated messages — DMCC false-urgency prohibition applies. Outputs: Response format spec (consistent structure template + novelty trigger list); notification timing policy (evening/morning consolidation windows, opt-out path documented); ethical pass/fail flag (affect-inference AI Act readiness checked; proactivity reversibility verified before ship).

Knowledge Base & Operational Guides

The references and playbooks below form the operational layer on top of the 12 primitives and 6 composition recipes above. The primitives describe neural mechanisms. The operational layer describes how to select and compose frameworks, what instrumentation and vendors to use, how to read observed signals, how to pass regulatory gates, and how to run a study or act on an observation.

Workflow

1. Identify the neural surface you are designing for (attention, arousal, social trust, narrative, regulatory orientation, aesthetics, interoception, memory, anticipation, embodied metaphor, prediction).
2. Use the Decision Checklist to identify which primitives are relevant.
3. Open the per-primitive playbook in assets/templates/consumer-neuroscience/ for the full definition, misuse boundary, and worked example.
4. Apply the Ethical Bounds harm test and the DMCC compliance check to each technique before implementation.
5. For compound design problems, use the Composition Recipes as starting stacks.
6. Check the Anti-Patterns table to confirm you are not inadvertently shipping an exploitative pattern.

ASCII Flow

Engagement or perception problem
  -> Identify neural surface: attention, arousal, trust, narrative, memory, reward, embodiment
  -> Confirm signal or proxy is available
     +-- no signal -> treat as hypothesis, not neuroscience claim
     +-- signal exists -> select primitive and playbook
  -> Apply ethical and regulatory gates
  -> Compose design pattern and measurement plan
  -> Ship only with user benefit, evidence, and monitoring

Navigation

• Per-primitive playbooks: assets/templates/consumer-neuroscience/ (one file per primitive)
• Composition guide: assets/templates/consumer-neuroscience/README.md
• Formal theory map: references/formal-theory-map.md
• Patterns, scenarios, and traps: references/patterns-scenarios-traps.md
• Domain-agnostic primitives overview: references/primitives-overview.md
• Sources: data/sources.json
• references/frameworks-meta.md — meta-frameworks (SOR, Consumer Decision Journey, Predictive Coding, Reactance) for primitive composition
• references/instrumentation-vendor-landscape.md — vendor and tool reference (August 2026), modality decision tree
• references/biomarker-signal-dictionary.md — signal → primitive → design-action cookbook
• references/ethics-operational-checklist.md — DMCC + EU AI Act + GDPR operational gates
• assets/playbooks/study-design.md — end-to-end study-design playbook
• assets/playbooks/signal-to-design-cookbook.md — observation → action recipes

Related Skills

• marketing-cro — conversion rate optimization; applies attention (#1), neuroaesthetics (#7), and social bonding (#3) at the page and funnel level
• marketing-content-strategy — narrative and copy; applies narrative transportation (#4), embodied cognition (#11), and mirror systems (#6) in content design
• marketing-paid-advertising — ad creative and landing pages; applies salience (#1), BIS/BAS framing (#5), and arousal calibration (#2)
• software-ui-ux-design — interface design; applies neuroaesthetics (#7), embodied cognition (#11), predictive processing (#12), and cognitive consistency
• product-management — onboarding and feature design; applies reward anticipation (#10), memory consolidation (#9), and retention loop design
• startup-business-models — pricing and packaging; applies interoception (#8) and social bonding (#3) in trust-based purchase design

Fact-Checking

• Primary sources are cited in each per-primitive playbook and in data/sources.json.
• Canonical references: Treisman 1980 + Itti & Koch 2001 (attention and salience), Yerkes & Dodson 1908 + McEwen 2007 (arousal physiology), Zak 2012 + Carter 2014 (social bonding and oxytocin), Green & Brock 2000 + Buckner 2008 (narrative transportation and DMN), Higgins 1997 + Carver & White 1994 (BIS/BAS and regulatory focus), Rizzolatti & Craighero 2004 + Hatfield 1993 (mirror systems and emotional contagion), Ramachandran & Hirstein 1999 + Chatterjee 2014 (neuroaesthetics), Damasio 1996 + Craig 2009 (interoception and somatic markers), Hebb 1949 + Walker 2017 (memory consolidation and sleep), Knutson 2001 + Berridge 2007 (reward anticipation and incentive salience), Lakoff & Johnson 1999 + Barsalou 2008 (embodied cognition), Friston 2010 + Clark 2013 + Constant et al. + Sprevak 2024 (predictive processing and active inference).
• 2025 source anchors: Bigne 2025 P&M (neurophysiological tools); Frontiers in Neuroergonomics 11 July 2025 (neuro-insights systematic review, DOI: 10.3389/fnrgo.2025.1542847); Bansal 2025 IJCS (neuromarketing and marketing mix); F1000Research 14:1132 (noninvasive neuromarketing methods); Frontiers in Human Neuroscience 2024 (xAI/fMRI brand perception); Sprevak 2024/2025 (predictive processing review).
• Numeric specifics cited where canonical: VTA dopamine anticipatory onset ~200ms before cue (Schultz 1998), oxytocin plasma half-life ~3–5 min (Ott et al. 2013), working-memory chunk limit ~4 items (Cowan 2001 — cross-referenced with behavioral-economics #14). Effect sizes vary by domain and population — always measure on your own users before treating published parameters as design constants.
• CORRECTION for #2 (arousal physiology) — BAAS, Nature Communications 2025: Neural affective arousal (cortical-subcortical signature: prefrontal, periaqueductal gray, thalamo-amygdala-insula) is statistically separable from autonomic arousal (GSR, HRV) and from wakefulness; validated across 24 studies, n=868 (Declerck-independent; Nature Communications, DOI: 10.1038/s41467-025-61706-0). Do not treat GSR/HRV as full proxies for subjective affective arousal — they capture physiological activation but miss the subjective-experience component. Where fMRI/EEG is unavailable, acknowledge this dissociation as a measurement limitation.
• CORRECTION for #3 (social bonding) — oxytocin–trust replication caution: The Kosfeld 2005 / Zak 2012 narrative that oxytocin universally increases trust has not replicated under registered conditions (Declerck et al. 2020, Nature Human Behaviour, >95% power: no main effect; DOI: 10.1038/s41562-020-0878-x). A 2025 preregistered high-powered study finds a selective ~15–17% trust increase in low-trust-disposition individuals only (bioRxiv 2025, preprint; DOI: 10.1101/2025.10.01.679711; note: preprint, not yet peer-reviewed). Apply #3 as "oxytocin system is implicated in affiliative bonding and trust formation" — not as "oxytocin = trust lever". Do not claim that oxytocin-adjacent design patterns universally increase trust; effects are context-dependent and moderated by individual trust disposition.
• EEG metric reliability for ad testing (J. Advertising 2024, DOI: 10.1080/00913367.2024.2418109): ISC (intersubject correlation) is the highest-reliability EEG metric for video ad testing; n≈11–15 achieves r=0.7 reliability. Alpha-asymmetry reliability does not improve with additional viewings and should be treated with caution. Meta-analytic corroboration: ISC–attention r=0.65 across 14 studies (BMC Psychology 2025, DOI: 10.1186/s40359-025-02879-7). Prefer ISC over alpha-asymmetry as primary EEG metric for ad/content evaluation.
• UPDATE for #10 (reward anticipation) — neuroforecasting property (Genevsky, Tong & Knutson, PNAS Nexus 2025; DOI: 10.1093/pnasnexus/pgaf029): NAcc activity during choice forecasts aggregate internet-market outcomes regardless of lab-sample demographic representativeness; MPFC predicts individual but does not generalise to aggregate in commercial market contexts. Minimum viable lab sample ~20–25 subjects. 2026 domain-extension (Srirangarajan et al., PNAS Nexus 2026; DOI: 10.1093/pnasnexus/pgag012; n=34): in conservation/social-media domain, group MPFC activity (not NAcc) forecast aggregate engagement on Instagram out-of-sample; NAcc and MPFC both predicted individual liking and donations. The two findings together establish a domain-dependent pattern — NAcc generalises to aggregate in commercial markets (Genevsky 2025); MPFC generalises to aggregate in social-media/conservation domain (Srirangarajan 2026). Practical implication: identify which region to prioritise for aggregate forecasting based on domain; small-N study remains sufficient.
• CORRECTION (2026-08-14) — EU AI Act high-risk date was 2 August 2026 in prior versions; it is now 2 December 2027. The AI Digital Omnibus (OJ 24 July 2026, in force 27 July 2026) deferred standalone Annex III high-risk obligations by 16 months, and Annex I to 2 August 2028. Article 50 transparency, Article 5 prohibitions, and GPAI obligations were not delayed — Article 50 applies from 2 August 2026 as originally scheduled (four-month grace to 2 December 2026 for the Art. 50(2) watermarking duty only). Any prior instruction to "be high-risk ready by August 2026" is superseded; the live obligation is the transparency notice.
• CORRECTION (2026-08-14) — US neural-data laws are narrower than prior versions implied. They do not generally cover GSR, HRV, eye-tracking, facial coding, or voice affect. Montana SB 163 expressly excludes "downstream physical effects of neural activity" (pupil dilation, motor activity, breathing rate); California SB 1223 excludes data "inferred from nonneural information"; Colorado reaches only identification-purpose data. Montana SB 163 (effective 1 Oct 2025) was missing entirely and is the strictest US regime for true neural data (per-purpose, per-recipient express consent). Source: FPF, "The Neural Data Goldilocks Problem"; Cooley neural-data patchwork survey (Feb 2026).
• EEG preference-measure consistency caveat: a 174-paper systematic review (Brain Informatics) reports frontal alpha asymmetry as the most-cited time-frequency preference signal and LPP as the most reliable ERP component, but finds limited consistency across papers, with each measure showing mixed results against actual preference and purchase behaviour. Combined with the J. Advertising 2024 reliability finding (alpha-asymmetry reliability does not improve with repeated viewings), this is a reason to prefer ISC and to treat any single-metric EEG preference claim as weak evidence.
• CORRECTION (2026-07-11) — Vermont H.814 overstated in prior versions of this skill: the enacted Act 101 has no consent requirement and no private right of action (both stripped in the Senate); its rights statement is largely declaratory with enforcement resting solely with the Attorney General. It is not a consent gate. Vermont's binding neural-data framework is S.71, effective 1 January 2028; see the corrected US Regulatory Context section above and references/ethics-operational-checklist.md.
• Do not conflate primitive #11 (embodied cognition) with discredited social-priming demonstrations (Bargh 1996 elderly-priming, money priming, cleanliness priming) — those largely failed multi-lab registered replication (Doyen et al. 2012; Many Labs 2 2018). Primitive #11 is conceptual-metaphor/grounded-simulation fluency, not covert cross-domain behavior steering; see the replication-boundary note in assets/templates/consumer-neuroscience/11-embodied-cognition.md and the Anti-Frameworks table in references/frameworks-meta.md.
• If web access is unavailable at runtime, mark any runtime-specific claim as unverified.

Learnings Loop

Before applying this skill on a non-trivial task, read learnings.consolidated.md in this directory (and learnings.md if present).

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.