Method and limits
Sixteen dimensions.Seven categories.One rule: show the work.
This page says how a persona is put together, how it changes over time, how it uses your material, and what it is not. Every number here describes the product as it runs today, and the page changes when the product does, not before.
16 dimensions in 7 categories, each from a named framework.
What the profile holds
- 16 dimensions in 7 categories, each recorded in its framework’s own terms: a score, a type or a written record.
- 30 personality facets under the Big Five. Replies draw on facets, not only the five headline scores.
- A life story with formative experiences, turning points, beliefs with strength, wounds with triggers, proud moments.
Not every trait shapes every reply. Which ones fire depends on the moment, and the trace shows which did.
Core personality
Big Five with 30 facets
How guarded, exploratory, disciplined or reactive an answer is, shaped by facets as well as the five headline scores.
How they think
Cognitive biases, need for cognition, locus of control, time orientation
What the persona actually hears in a question, and which shortcuts it takes to an answer.
Why they act
Schwartz values, Self-Determination Theory, moral foundations
What a proposal has to satisfy before it is acceptable, and what offends on principle.
How they feel
Emotional architecture
Baseline mood, triggers, regulation, and how much of a shift shows in the reply versus stays in the monologue.
How they relate
Attachment, social identity
How trust is extended or withheld, and how the persona reads the person asking.
Who they are
Narrative identity: life story, beliefs, wounds, proud moments
Which questions land on an old wound, and which beliefs a persona will defend or, under enough argued challenge, revise.
Dynamic layers
Communication style, Hofstede’s cultural dimensions, context
Vocabulary, directness, formality, humour and how the setting changes the register. Hofstede’s dimensions are recorded in the profile but do not shape replies yet.
Reasoned in stages, not generated in one shot. Every stage leaves a record.
Nothing is fine-tuned and nothing is a single prompt. A reply is reasoned in stages: the persona reads who is asking and in what frame, processes what it actually heard through its own biases, recalls what it knows and remembers, registers how the question lands and whether it touches a belief or a wound, and only then answers. Keeping the stages apart is the point. A persona that decides how it feels before it decides what to say behaves differently from one that generates both at once.
How the stages are wired, which models do what and what each stage costs are internal decisions that change as models improve, and every model call is metered so a change shows up. What does not change is the record: every stage writes to the trace, and the trace is yours to read.
What the trace holds, for every reply
- 1Verdict · why it answered as it did, in one paragraph
- 2Situation · who it thinks is asking, and in what frame
- 3What it heard · the question as its biases received it
- 4What it drew on · the memories and the sources that surfaced
- 5How it moved · emotion and trust, with this turn’s change
- 6What was touched · a belief validated or challenged, a wound, a pride
- 7What it withheld · the thought it had and chose not to say
All of it is visible in the Inner World panel of every conversation, and it is what the report cites.
The persona at week six is not the persona at week one.
- Six memory types: episodic, semantic, emotional, procedural, prospective, associative. One memory forms per reply, and what comes back is weighted by relevance, emotional weight and recency.
- Emotional state belongs to the conversation: it is still there when you come back, and it settles back toward the persona’s baseline over hours. A new conversation starts from the baseline.
- Trust starts at 50 for everyone and never decays on its own. It moves on what you say, and the move is stored with the turn.
- Consolidation. Over time, old and low-weight memories cluster and fade while the important ones stay, so the persona at week six is shaped by what mattered.
- Beliefs evolve. Repeated, argued challenges accumulate as evidence. Past a personality-dependent threshold the belief is rewritten and a memory records the change.
Knowledge is a separate substrate
Material is not memory. Sources are attached to personas with a declared fluency and retrieved alongside memory when a question calls for them. The grounding mode decides how the persona treats a fact it cannot find: open personas improvise like a well-read person, grounded personas flag the assumption in their inner world, strict personas say they do not know, in character.
Feeding a persona an experience writes learned facts into a source of its own, so what it lived through becomes something it knows.
The limits, on the label.
- Not a prediction of real people. No public concordance study exists yet. Until one does, the honest description is a rehearsal, not a replacement, for talking to real customers. That sentence is in the product and in the terms, and every exported report ends with the line that personas are AI simulations, not real customers.
- Not a model of anyone real. Personas are fictional or composite. The terms forbid re-identifying or targeting real individuals, and forbid presenting simulated output as responses from real participants.
- Not a fine-tuned model. The behaviour comes from stored psychology, memory, grounding and orchestration over frontier models. A change in those models changes the product, and the metering exists to see it.
- Not finished. A full-screen Live view that replays how each reply was reasoned is built and awaiting deployment. Teams, billing and PDF export are not built. This page changes when the product does, not before.
Straight answers.
Do the personas predict what real people will say?
We do not claim that yet. No public concordance study exists, and until one does the honest description is a rehearsal, not a replacement, for talking to real customers. The first concordance study, run against published human research, is the next milestone, and its results will be published whether they flatter the product or not.
Which models power it?
Frontier models from Anthropic for the reasoning and the report compilation, and OpenAI for the embeddings behind memory and knowledge retrieval; both are named as processors in the privacy policy. Which model runs which part is an internal decision that changes as models improve, and every model call is metered so a change shows up. Nothing is fine-tuned; the behaviour comes from the stored psychology, the memory and the grounding.
Is my research material used to train models?
No. Conversation content and persona profiles are sent to Anthropic and OpenAI to generate replies and embeddings. Perplica does not permit its API traffic to be used to train their models. The full policy is in the privacy policy on this site.
Can I model a real, named person?
No. Personas are simulations of fictional or composite people built from research material. The terms forbid attempting to re-identify or target real individuals, and forbid presenting simulated output as responses from real participants.
What does “16 dimensions” mean?
Sixteen psychological dimensions in seven categories, each drawn from a named framework: the Big Five with 30 facets, cognitive biases, need for cognition, locus of control, time orientation, Schwartz values, Self-Determination Theory, moral foundations, emotional architecture, emotional intelligence, attachment, social identity, narrative identity, communication style, Hofstede’s dimensions and context. The method page lists what each changes in a reply. Hofstede’s dimensions are recorded in the profile but do not shape replies yet.
How is pushback measured?
Trust starts at 50 for every persona and moves on what you say; it never decays on its own. We measure the pushback rate across trust levels and across conversations, so whether low-trust personas challenge more is a number we look at, not a promise we make.
Private beta · by application
Bring one real research question.
Access opens by application. Tell us the question you are trying to answer and who has to be convinced by the result.
A rehearsal, not a replacement, for talking to real customers. Stated in the product and in the terms. Every exported report ends with the line that personas are AI simulations, not real customers.