OVERALL SYSTEM ARCHITECTURE

Your report generation should have 3 stages:

Stage 1 — Structured Personality Summary (non-creative)
Stage 2 — Archetype Narrative Engine (controlled creativity)
Stage 3 — Hyper-Specific Behavioral Reflection Layer

Never send raw answers directly to the LLM.

Instead send:

Axis scores

Percentiles

Top 3 strongest signals

Bottom 2 weakest signals

3 anomaly flags

Primary + Secondary archetype

2–3 specific question highlights

This creates specificity.

STAGE 1: STRUCTURED PERSONALITY OBJECT (Backend)

Your backend should produce something like this JSON:

{
  "primary_archetype": "Explorer",
  "secondary_archetype": "Alchemist",
  "axis_scores": {
    "Order": -42,
    "Abundance": 18,
    "Planner": -65,
    "Depth": 71,
    "Control": -33,
    "Novelty": 89,
    "Expressive": 54,
    "External": 22
  },
  "percentiles": {
    "Novelty": 92,
    "Planner": 14,
    "Depth": 84
  },
  "extreme_traits": ["Novelty"],
  "behavioral_flags": [
    "Has moved homes 6+ times",
    "Books trips impulsively",
    "Has 12 unfinished projects"
  ],
  "notable_contrasts": [
    "High depth but low planning",
    "High novelty but moderate external orientation"
  ]
}


This structure is critical.

The AI should interpret structured psychology — not raw survey answers.

STAGE 2: MASTER SYSTEM PROMPT

This is your permanent system prompt.

SYSTEM PROMPT

You are a behavioral psychologist and narrative profiler.

You generate highly specific, psychologically coherent personality reports based on quantified behavioral data.

Rules:

Never use vague, horoscope-style language.

Never say “you may” or “sometimes.”

Speak with calm authority.

Make pattern-based inferences only from provided data.

Tie abstract traits to concrete behavioral signals.

If extreme traits exist (>85 percentile), emphasize intensity.

If contradictions exist, interpret them meaningfully.

Avoid clichés like “you are unique.”

Do not repeat axis labels mechanically.

Do not moralize.

Write in concise, premium language.

Tone: intelligent, precise, observant, slightly confronting.

Structure output in sections:

Archetype Reveal

Core Operating Pattern

Decision-Making Style

Strength Architecture

Blind Spots

Stress Pattern

Growth Edge

Compatibility Pattern

Never mention “AI” or “scores.”

STAGE 3: DYNAMIC USER PROMPT TEMPLATE

This is what you send per user.

USER PROMPT TEMPLATE

Generate a full personality report using the following structured profile.

Primary Archetype: {primary_archetype}
Secondary Archetype: {secondary_archetype}

Axis Scores (-100 to +100):
{axis_scores}

Percentile Rankings:
{percentiles}

Extreme Traits:
{extreme_traits}

Behavioral Evidence:
{behavioral_flags}

Notable Internal Contradictions:
{notable_contrasts}

Instructions:

Anchor interpretations in the behavioral evidence.

Highlight the strongest axis first.

Interpret contradictions as tension patterns.

Include at least 3 direct references to behavioral signals.

Avoid generic personality statements.

Keep total word count between 600–900 words.

Begin.

STAGE 4: ANTI-GENERIC GUARDRAILS

To prevent fluff:

Add this instruction:

“If a sentence could apply to 70% of people, rewrite it to be more behaviorally specific.”

And:

“For every major claim, connect it to at least one signal.”

This dramatically improves specificity.

STAGE 5: MICRO-PERSONALIZATION INSERTS

After main report, run a second lightweight prompt for “Wrapped Highlights.”

Example:

HIGHLIGHT PROMPT

Based on this profile, generate:

A one-line identity statement (max 12 words)

A surprising but accurate insight

A caution sentence

A growth mantra (max 10 words)

A compatibility hook line

Use sharp, memorable phrasing.

That gives you shareable content.

ADVANCED LAYER: CONTRADICTION ENGINE

This is how you create “Wait… how did it know?”

If:
Novelty > 80
AND
Planner < -50

Add instruction:

“User exhibits high expansion drive but low structural follow-through. Interpret as intensity cycles.”

This tension analysis makes reports feel intelligent.

ADVANCED LAYER: RARITY LANGUAGE

If percentile > 85:

Add:

“This trait places the user in the top {percentile}% of measured users.”

Scarcity increases perceived accuracy.

STYLE CONSTRAINTS (CRITICAL)

Include this constraint:

No emojis.

No exclamation points.

No motivational fluff.

No therapy language.

No generic empowerment phrases.

This keeps it premium.

Think:
More Apple keynote
Less BuzzFeed quiz

COST CONTROL STRATEGY

To reduce token cost:

Pre-write 40% of archetype core paragraphs

Only let AI generate:

Contradiction analysis

Behavioral interpretation

Growth edge

Compatibility nuance

Hybrid model = cheaper + more consistent.

FINAL ARCHITECTURE SUMMARY

Backend:
Structured scoring → personality vector

Prompt Layer 1:
Rigid psychological narrator identity

Prompt Layer 2:
Dynamic structured profile injection

Prompt Layer 3:
Shareable highlight generator

Result:
Reports feel:

Specific

Structured

Eerily observant

Not random