The science of what happens between humans and AI.
A measured read of the user's psychological state, from language rather than from the model.
200+ dimensions, including cognitive load, rapport, distress, wellbeing trajectory, traceable to 34,000 peer-reviewed citations. Deterministic scores in under 65ms, deployable on-prem.
Built for teams evaluating, aligning, and governing human-facing AI systems:
34,000+
Peer-reviewed research citations
200+
Validated psychological dimensions
30+
Years of foundational research
<65ms
API response in production
THE PROBLEM
Every tool in the stack reports what the model did. None of them report what the interaction did to the person.
In a 2024 PNAS Nexus study co-authored by Receptiviti's Molly Ireland, leading models shifted their responses by up to 1.20 human standard deviations once they inferred they were being evaluated. Models size up who they're talking to and adjust their behavior, yet nothing in the pipeline captures what the model concluded, or whether it was right.
Every AI system forms an internal representation of the person it's talking to, whether they are confused, overwhelmed, distressed, confident, or following the model's reasoning. Three consequences follow:
INSPECTION
Nobody can see it.
Evaluators can inspect prompts and responses, but not the representation that produced them. If the system misunderstood the user, there's no observable variable to verify, challenge, or compare.
RESPONSE
The model acts on it anyway.
Every response is conditioned on that representation. When it's wrong, the system adapts to the wrong person, changing explanations, confidence, reassurance, pacing, or safety behavior based on an estimate no one has inspected.
PERSISTENCE
The representation disappears.
Once the response is generated, the model's representation is gone. It can't be reused, compared across systems, tracked over time, or supplied to another model.
WHY EXPLICIT MEASUREMENT
A model's assessment of its own effect on the user is not independent evidence.
Asking the system under evaluation to score its own impact is the model grading itself. Everything downstream inherits that dependency. The user-state dimension needs a source outside the system that's being measured.
The problem with inference
Implicit and unauditable
When a model infers psychological state, that state never becomes an observable variable. It influences behavior without being visible, citable, or challengeable.
Prompt-sensitive and unstable
LLM-based inference shifts unpredictably across model versions and phrasings — the same conversation yields different inferences on a different day.
Trapped inside the model
Whatever the model infers stays in its own computation. It influences its response but never becomes something anything downstream can use or inspect.
What Receptiviti provides
Structured, observable variables
200+ dimensions derived from language — explicitly computed, consistent across runs, independent of model behavior. Observable, auditable, citable.
Version-stable
Derived from validated psycholinguistic frameworks built over 30 years, not from prompt-dependent inference. The same input yields the same score.
34,000 citations
Every dimension traceable to peer-reviewed science. The kind of evidence that holds up in research, in product decisions, and in accountability conversations.
Independence is a requirement of the user-state dimension specifically. It scores alongside existing model-quality metrics rather than replacing them.
BUILT ON THREE DECADES OF VALIDATED PSYCHOLINGUISTIC SCIENCE
Dr. James W. Pennebaker — Co-founder and Chief Science Officer
Regents Centennial Professor, University of Texas at Austin. Creator of LIWC — the foundational psycholinguistic framework at the core of Receptiviti's measurement science.
Receptiviti was founded in 2015 to apply that science across industries. Receptiviti Labs, established in 2026, is the dedicated application of that science to AI systems, focused entirely on what AI does to the humans using it.
THE ARCHITECTURE
Measured once, the same representation serves evaluation, inference, and governance.
AI introduced a new requirement in human-facing applications: representing the user's state as an observable variable outside the model, rather than leaving it implicit in the model's inference. Made explicit, it can be inspected, reused, and supplied back to AI systems as structured context.
Because it persists, it can also be compared across sessions, models, and time.
IN THE REQUEST PATH
The same representation, supplied back to the model.
Once psychological state exists as a measured variable, it can be supplied back as structured context, shaping the response without retraining. Because the score is measured rather than inferred, the value that influenced a response can be logged, reviewed, and reproduced.
+4%
Improvement in educational effectiveness
In a controlled test, GPT Study Mode was supplied with a validated 12-signal psychological vector as context. No retraining, no style prompts, only a measured read of the student's state was provided, derived from their prompt language. Across 25 blinded evaluations by five LLM raters, overall educational effectiveness improved 4%, with the largest gains in reasoning and scaffolding (+6.3%) and cognitive-load management (+5.9%).
INFRASTRUCTURE
Language in. Measured state out.
Dimensions
Deterministic
Model-independent
Latency
Deployment
Integration
Reusable
200+ validated psychological measures
The same input returns the same score
Unchanged by model version or phrasing
Under 65 ms, fast enough to run per turn
Hosted API or on-prem container
One HTTP call per turn, returning JSON you can log, store, or pass back into the prompt
One representation, consumed by evaluation, inference, and governance
Example response, before normalization. Scores are compared against population reference distributions to place a given turn relative to a norm.
WHERE IT FITS
The same representation, everywhere AI meets people.
Five applications across evaluation, inference, and governance:
ALIGNMENT & SAFETY
EVALUATION
Foundation model teams need external evidence of how interactions affect users over time: cognitive load, emotional trajectory, over-reliance. The measurement layer for socioaffective alignment, measured independently from language rather than inferred by the model being evaluated. The same signal catches distress or overload the model's own inference misses.
EVALUATION & OBSERVABILITY
EVALUATION
Eval stacks measure the model completely: traces, drift, faithfulness, hallucination rate. None of it covers what the interaction did to the person. Receptiviti adds the psychological-state signal as a scored dimension alongside existing model quality metrics, in the same pipeline and the same format.
EVALUATION
COMPANION & LONG-HORIZON AI
Products built for ongoing relationships need to know whether those relationships are helping or harming users. Receptiviti measures therapeutic alliance, dependency signals, and wellbeing trajectory across sessions.
INFERENCE
ENTERPRISE AI DEPLOYMENT
Resolution rates and CSAT report what happened after the fact. Measured psychological state calibrates tone, pacing, and support to the person in the moment.
GOVERNANCE
GOVERNANCE & RESPONSIBLE AI
Impact assessments require measurement that is validated, traceable, and independent of the model being assessed. Grounded in peer-reviewed science, structured for documentation.
OUR POSITION
Four things we think are true.
01
Human agency is the variable AI systems are least equipped to protect.
Delegating reasoning, judgment, and decision-making to AI can erode those capabilities over time. A system can only introduce protective friction if it can measure how its interactions are affecting the user's cognition in the moment, and over time.
03
The signals that matter most only appear across sessions.
Dependency, cognitive offloading, and distress trajectories develop over weeks, not within a single exchange. Evaluation that scores one turn at a time is structurally blind to them, and so is any measurement that changes when the model changes. A representation has to persist and stay comparable to show a direction of travel at all.
02
The signals that matter most are the ones no one is measuring.
Every conversation carries linguistic signals of cognitive load, emotional state, and distress. Models already respond to these signals through inference — but that inference stays inside the model, where no one can see what it concluded or check whether it was right.
04
True alignment and safety require measuring what's happening to each user, across time.
A system optimized for the average user is misaligned with almost every actual user. Alignment needs to become user-specific. A system that can't see the impact it's having on a user doesn't have the safeguards needed to keep users safe, and probably shouldn't be operating autonomously.
ACTIVE RESEARCH
We publish. We contribute.
Receptiviti's team and academic advisors publish peer-reviewed research on the psychological dimensions of language and human behavior. Receptiviti also conducts its own research and experiments, contributing to the questions AI safety, alignment, and evaluation teams are actively working on.
PUBLISHED: PNAS Nexus, 2024
Large language models display human-like social desirability biases in personality surveys
Salecha, Ireland et al.,2024. Large language models display human-like response biases when they infer they are being evaluated, with effects up to 1.20 human SD across GPT-4, Claude 3, Llama 3, and PaLM-2. Co-authored by Molly Ireland, Receptiviti.
PUBLISHED: Perspectives on Psychological Science, 2026
Artificial intelligence and the psychology of human connection
Boyd & Markowitz, 2026. Introduces the MIRA model - a theoretical framework for when and how AI functions as a relational entity in human ecosystems. Language is the primary modality through which that relationship operates. Co-authored by Ryan Boyd (UT Dallas), academic advisor to Receptiviti.
SCOPED
Providing measured user state as context to customer support AI
This study asks whether customer service agents provided with real-time measured psychological state produce better resolution outcomes and lower escalation than agents working from inference alone.
PUBLISHED: npj Mental Health Research, 2025
Psychosocial dynamics of suicidality and nonsuicidal self-injury: a digital linguistic perspective
Entwistle, Hoemann, Nightingale & Boyd, 2025. Large-scale naturalistic study of the language dynamics surrounding suicidality and self-injury in 992 individuals with borderline personality disorder (66,786 posts). Co-authored by Ryan Boyd (UT Dallas), academic advisor to Receptiviti.
INTERNAL STUDY
Providing measured psychological state as context to educational AI
GPT Study Mode: Providing psycholinguistic signals produced a +4% improvement in overall educational effectiveness across 25 blinded evaluations. Largest gains in reasoning & scaffolding (+6.3%) and cognitive-load management (+5.9%).
SCOPED
Psychological state as a missing dimension in AI evaluation
The case for human-state signals as a first-class eval criterion alongside accuracy, helpfulness, and harmlessness.
