From Relevance to Resonance

Executive Summary

For the past two decades, the world’s largest digital platforms have operated on a single organizing principle: relevance.

Relevance engines, like those used across Meta’s ecosystem, learn to predict what people will click or watch longer, optimizing for engagement through correlation.

But human connection does not emerge from correlation. It emerges from resonance, the psychological alignment between an experience and the intrinsic motivations, values, and personality of the individual.

Relevance systems can infer what we appear to like, but they cannot know why. The distinction is not semantic; it is structural. Without psychological ground truth, behavioral systems are epistemologically capped.

Solsten’s Resonance Framework closes that gap. Built from over 1.2 million psychometrically profiled individuals across 90 countries, the Solsten Audience Dataset combines validated psychological measurement with advanced AI models to infer human traits, motivations, and values at scale.

The result: data that doesn’t just predict behavior, it explains it.

Across global clients, resonance-based models have driven:

• +30% higher creative engagement
• −22% lower customer acquisition cost (CAC)
• +18% retention
• +12% lifetime value (LTV)

This paper outlines why behavioral data has reached its ceiling, how Solsten’s psychometric-AI hybrid architecture overcomes those limits, and why the future of personalization, marketing, and human-centered AI depends on resonance-based data.

1. The Core Limitation of Behavioral AI

Behavioral systems, such as Meta’s Relevance Models, derive insight from observed behavior: clicks, views, dwell time, and reactions. These signals are contextual, not psychological. A user likes a football post not because they are defined by that interest, but because of situational context:

• Social signaling to peers
• Algorithmic feedback loops
• Transient curiosity

Behavioral systems can correlate patterns of attention but cannot access intentionality or motivation. They operate on the assumption that “you are what you do,” ignoring decades of psychological research showing that behavior is context-dependent and role-driven.

Humans adapt their language and expression to fit context: we speak differently to our parents than to our coworkers. On social media, an environment of low authenticity and high performativity, people act to gain attention, not to reveal selfhood.

This contextual distortion breaks behavioral inference. Without ground-truth psychological anchors, the system can only guess.

2. The Solsten Dataset: Global, Psychometrically Grounded, Authentically Sourced

Since 2018, Solsten has collected over 1.2 million complete psychometric profiles through an incentivized convenience sampling framework integrated within digital play environments, primarily video games, where individuals behave with heightened authenticity and intrinsic motivation.

Sampling and Composition

• Sample size: 1,228,285 participants (as of Sept 2025)
• Languages: English, French, German, Japanese, Chinese
• Geographic distribution: 41.5% U.S., 20.4% Western Europe, remainder globally
• distributed
• Age: 18–107 (M = 38.49, median = 36)
• Gender: ~40% male,
• ~50% female, 10% other/unspecified
• Education: 16% Master’s or PhD; 43% full-time employed

Incentives are in-context (e.g., in-game rewards, credits), which attract high-value populations typically missing from paid survey panels, without financial bias.

Survey Domains

Each assessment includes:

1. Demographics: standardized for regional comparability
2. Psychographics: personality, values, motivations, communication styles
3. Well-being: emotional stability and anxiety indicators
4. App Usage: behavioral and contextual usage data
5. Affinities: brands, activities, media, and interests

Authenticity of Context

Unlike social media, games are high-fidelity behavioral environments where users experience flow, autonomy, and intrinsic motivation. In play, social pretense collapses, people behave closer to their true selves. This makes games one of the most ecologically valid contexts for measuring authentic personality expression.

3. Psychometric Ground Truth and Adaptive AI Assessment

Solsten’s psychometric foundation is rooted in clinically validated models including:

• Big Five personality model (with 20+ facet-level traits)
• Achievement Motivation Inventory (AMI)
• Personal Values Assessment
• Cultural dimensions (Hofstede framework)
• Communication Style Inventory

Each assessment is delivered through adaptive item sequencing, a system akin to computerized adaptive testing (CAT). Using machine learning, the assessment predicts the most informative next question based on prior responses, optimizing both accuracy and engagement.

This results in:

• Completion rates: 30–80% depending on platform
• Average test reliability: α = 0.83
• Test–retest stability: r = 0.78 over eight weeks
• Cross-language invariance: ΔCFI < 0.01 across five languages

Every response is linked to construct-level variables, personality, motivation, and value traits that remain stable across context, providing a true psychological ground truth for AI training.

4. The Resonance Engine Architecture

Solsten’s Resonance Engine operationalizes this data through a multi-layer AI system that combines psychometrics, embeddings, and causal inference.

4.1 Data Flow Pipeline

1. Assessment Ingestion: validated questionnaire responses are anonymized and tokenized into high-dimensional trait vectors.
2. Feature Embedding: embeddings are generated via transformer-based models trained on psychometric feature spaces, enabling semantic proximity between psychological constructs and behavioral data.
3. Graph Integration: embeddings are linked through acyclic causal graphs that model probabilistic dependencies between traits, motivations, and observable engagement.
4. Inference Layer: for users without completed assessments, the model infers likely trait profiles using graph-based causal inference and multi-modal similarity.
5. API Deployment: inference models are deployed in production environments with <50 ms latency for real-time resonance prediction.

This architecture allows Solsten to predict psychological traits of users who never took an assessment by learning from the behavioral analogs of those who did.

4.2 Scalability

The Resonance Engine currently processes over 150,000 new data points per year. Data is stored in a graph-based feature store optimized for embedding search and causal modeling, enabling sub-second audience segmentation and cross-domain prediction.

5. Scientific Validation and Predictive Performance

5.1 Reliability and Validity

Across 14 language-trait combinations:

• Internal consistency: α = 0.83
• Construct validity: AVE = 0.62 (average variance extracted)
• Measurement invariance: confirmed across language cohorts

5.2 Predictive Benchmarks

In cross-industry deployments:

• Solsten trait embeddings predicted 30-day retention with R² = 0.62, compared to 0.24 for behavioral relevance models.
• Creative engagement scores improved by +31% when campaigns were aligned with audience personality clusters.
• Retention improved +18%, CAC decreased −22%, and LTV increased +12% across 14 enterprise partners.

5.3 Cross-Validation

K-fold cross-validation (k = 10) across subsamples demonstrates stable model performance with <3% variance, confirming generalizability across cohorts and cultural contexts.

Together, these findings demonstrate that psychometric data provides both explanatory and predictive superiority over behavioral signals alone.

6. Why Meta or Google Couldn’t Build This

6.1 Structural and Architectural Constraints

Dimension Big tech: Behavioral Systems (e.g., Meta Relevance, Search) Solsten: Solsten Resonance Framework

Data Provenance Big tech: Passive, context-dependent signals (clicks, likes, dwell time) Solsten: Active, consent-based psychometric measures

Ground Truth Big tech: No validated psychological baseline Solsten: Clinical-grade psychometric calibration

Context Authenticity Big tech: Socially performative, low-fidelity Solsten: Intrinsically motivated play environments

Inference Basis Big tech: Correlation between signals Solsten: Causal modeling of psychological constructs

Privacy Compliance Big tech: Restricted under GDPR/CCPA for implicit profiling Solsten: Fully compliant via explicit informed consent

IP Protection Big tech: No proprietary claim to psychometric inference Solsten: 20+ granted/pending patents covering adaptive assessments, content-to-psychology correlation, and personalized experience delivery

6.2 Patent Landscape

Solsten holds one of the most extensive patent portfolios in human-centered AI, covering the entire resonance data pipeline, including:

1. Adaptive Experience Design

Patents that personalize digital environments in real time based on user psychology and interaction.

• Systems and Methods to Provide a Digital Experience Adapted Based on a Subject Selection to Effect Particular Attributes - US 2022-0238205, EP 4285379
• Systems and Methods to Adapt a Digital Application Environment Based on Psychological Attributes of Individual Users - US 2022-0342791, EP 4327542
• Systems and Methods to Effectuate Presentation of Customized Content to Subjects Within Integrated Applications - US 2024-0062908
• Systems and Methods to Determine Content to Present Based on Interaction Information of a Given User - EP 4282130, US 2024-0323191

2. Psychological Resonance Mapping

Patents that uncover and apply deep psychological correlations between behavior, motivation, and digital interaction.

• Systems and Methods to Correlate User Behavior Patterns Within an Online Game With Psychological Attributes of Users - EP 4161669, US 2024-0293751
• Systems and Methods to Correlate User Behavior Patterns Within Digital Application Environments With Psychological Attributes of Users to Determine Adaptations to the Digital Application Environments - US 2023-0376984
• Systems and Methods to Identify and Utilize Correlations Between Content Classifications and Psychological Profiles of Users to Provide an Adaptable Digital Environment - US 2024-0233910, WO 2024-086468
• Systems and Methods to Identify Psychological Profiles of Users by Utilizing Correlations Between Content Classifications - US 2024-0273149, WO 2024-173077

3. Resonant Content Intelligence

• Patents that identify, curate, and optimize content that connects meaningfully, not just contextually, with human psychology.
• Systems and Methods to Analyze and Identify Effective Content for a Curation in Digital Environments - US 2024-0127311, WO 2024-086465
• Systems and Methods to Identify Expressions for Offers to Be Presented to Users - US 2024-0070701
• Systems and Methods to Identify Taxonomical Classifications of Target Content for Prospective Audience - US 2025-0016413

4. Human Insight and Wellbeing Systems

Patents that extend psychological understanding toward individual wellbeing, trust, and therapeutic outcomes.

• Systems and Methods to Facilitate Management of Online Subject Information - US 2023-0177205, EP 4445282
• Systems and Methods to Assess Health Information for a Subject Based on Subject-Initiated Sessions With Online Applications - US 2024-0055132
• Systems and Methods to Facilitate Adjusting Content to Facilitate Therapeutic Outcomes of Subjects - US 2024-0266073

5. Cross-Domain Psychological Integration

Patents that connect user understanding across digital platforms, enabling a unified view of the human experience.

• Systems and Methods to Link Psychological Parameters Across Various Platforms - EP 4281974

This IP coverage effectively precludes replication by legacy behavioral platforms. Meta or Google cannot legally or architecturally implement equivalent systems without explicit psychological data capture and informed consent, conditions that conflict with its business model and privacy obligations.

7. Applications and Business Outcomes

7.1 Marketing and Creative Optimization

Resonance data enables marketers to map values, motivations, and communication styles directly to creative design, copy, and targeting.

In practice:

• A campaign targeting “23-year-old men who like football” becomes obsolete.
• Instead, brands can reach “individuals high in altruism, low in neuroticism, motivated by team achievement,” regardless of demographics.

Across enterprise clients, campaigns aligned to psychological resonance achieved:

• +30–35% lift in engagement
• −22% average reduction in CAC
• +12% increase in LTV

7.2 Product Design and UX Research

By correlating app usage with personality traits and motivational clusters, Solsten data identifies friction points in UX and reveals what makes experiences emotionally rewarding.

7.3 Predictive Modeling and Audience Expansion

Resonance embeddings predict similarity between seemingly disparate audiences, revealing shared psychological profiles between, for instance, “New York sports fans” and “Japanese role-playing gamers.”

This enables cross-market expansion and partnership discovery far beyond demographic segmentation.

8. Ethics, Privacy, and Governance

Solsten’s data collection is designed to exceed international ethical standards for human-subject research.

• Consent: explicit informed consent prior to participation
• Anonymization: pseudonymized identifiers; no IP or device-level retention
• Data storage: encrypted at rest (AES-256) and in transit (SSL/TLS) on Google Cloud
• Regulatory compliance: GDPR, CCPA, and equivalent frameworks
• Deletion rights: built-in participant deletion and correction requests

Because participants voluntarily engage within environments they love, Solsten achieves both ethical integrity and data authenticity, a combination unavailable to ad-based platforms.

9. Limitations and Future Work

Solsten’s sampling is non-probabilistic, limiting strict population generalization. However, with a sample exceeding 1.2 million profiles and continuous demographic benchmarking, deviations from representative panels remain within ±3%.

Future directions include:

• Integration of passive behavioral telemetry to enrich predictive embeddings while preserving consent-based collection
• Expansion to additional languages and cultural frameworks
• Development of Resonance Embeddings API for real-time integration into content generation and adaptive agent systems

10. Conclusion: The Shift from Relevance to Resonance

Behavioral data taught machines what we do. Resonance data teaches them who we are.

Meta’s relevance models revolutionized engagement prediction but plateaued at the surface layer of behavior. Without ground-truth psychological constructs, such systems will always optimize for attention, not alignment.

Solsten’s resonance framework represents the next phase of human-centered AI, where personalization is not about manipulation but mutual understanding.

With a dataset of over a million psychometrically grounded individuals, patented adaptive assessment systems, and embedding-based inference models, Solsten transforms audience understanding from reactive targeting to proactive resonance.

The companies that win the next decade will not be those who know what people click. They will be the ones who know why.