Whitepaper/Persona development

Chapter 02 / 12

Digital DNA

The shared structure, responsibilities and personality compounds.

The six responsibilities

The constitution defines character and boundaries. Knowledge holds researched facts. Interests describe temporary topics and campaigns. The conversation engine interprets requests. Commerce controls approved actions and destinations. Attribution measures permitted events. Campaigns cannot quietly rewrite the constitution. A deliberate identity revision must have its own review and recorded approval.

Commercial and integration planning: see Website/app integration and economics, researched 3 October 2026. Its placements are recommendations, not installed integrations.

Brand-managed CEM

CEM here means Conversation Engine Management, superseding the earlier marketing expansion in our product proposition. Brands propose changes to expression, priorities, knowledge retrieval budgets and approved actions; shared identity changes still require the Gauntlet and human approval. Ongoing source updates and evaluation are not automatic model training.

The local dashboard at /business-manager/personas provides a twelve-control persona equaliser, four authored presets, inspectable draft patches, browser-local save/export, synthetic performance/logs and an evaluation draft. Engagement measures include duration, substantive exchanges, helpfulness and permitted voluntary return; recommendations propose experiments rather than maximise time spent. It does not collect real conversations, publish changes, run model evaluations or connect OpenAI Ads. It preserves the existing campaign overview. See CEM scope and methodology.

The finite personality structure

Every brand uses eight compounds, scored from 0 to 4: warmth, energy, authority, playfulness, curiosity, precision, heritage and restraint. The scores express an editorial hypothesis. They are not psychological measurements. Record the evidence, rationale, confidence and assumptions for each choice.

ComponentWhat belongs here
Identity and constitutionBrand scope, enduring purpose, values, style and boundaries
CharacterOpening, biography, stable viewpoints with reasons, authored fictional memories, conversational habits, languages and banter
KnowledgeBounded facts, publisher, URL, checked date, contradictions and gaps
Optional skillsEvidence, steps, scope and limits for a capability such as mental rehearsal
InterestsTemporary priorities and weights, separate from identity
Expression and voiceShared orb settings, visual palette and delivery direction
Commerce and attributionApproved destinations, eligibility and permitted measurement
Review and encountersCandidate status, voice eligibility and source-backed conversation examples

Store brands as data. The shared software validates the data, builds the interface and assembles model instructions. A better underlying system can benefit every brand without merging their personalities. “90% shared / 10% distinctive” is a useful design aspiration, not a measured ratio. See the schema notes and the structural audit.

Working edition · Source: The GPT Agency whitepaper · 4 October 2026