Content Dna

lib/content-dna/fingerprint.ts
lib/content-dna/profile.ts
lib/content-dna/synthesizer.ts

DNA Dimensions

Every piece of content is fingerprinted across these dimensions:

interface ContentDNA {
  contentId: string;
  organizationId: string;
  platform: string;

  // HOOK DNA
  hook: {
    type: string;           // "question", "statistic", "bold_claim", "story_opener", "how_to", "list_number", "controversy", "curiosity_gap"
    length: number;         // Characters in first line
    usesNumber: boolean;    // "5 things..." "73% of..."
    usesEmoji: boolean;
    usesYou: boolean;       // Addresses reader directly
    emotionalTone: string;  // "surprise", "urgency", "curiosity", "empathy", "authority"
  };

  // CONTENT DNA
  content: {
    topic: string;
    subtopic: string;
    angle: string;          // "educational", "inspirational", "entertaining", "promotional", "behind_scenes"
    structure: string;      // "tip_list", "story_arc", "problem_solution", "before_after", "myth_bust", "how_to"
    captionLength: number;
    paragraphCount: number;
    usesLineBreaks: boolean;
    readabilityLevel: string; // "simple", "moderate", "complex"
    callToAction: string;   // "save", "comment", "click_link", "follow", "share", "none"
    ctaPlacement: string;   // "end", "middle", "beginning"
  };

  // VISUAL DNA
  visual: {
    type: string;           // "ai_photo", "product_shot", "text_overlay", "carousel", "reel", "infographic"
    dominantColor: string;
    brightness: string;     // "light", "medium", "dark"
    hasText: boolean;
    textAmount: string;     // "none", "minimal", "moderate", "heavy"
    hasProduct: boolean;
    hasPerson: boolean;
    layout: string;         // "centered", "left_aligned", "full_bleed", "split"
    slideCount?: number;    // For carousels
  };

  // TIMING DNA
  timing: {
    dayOfWeek: number;
    hourOfDay: number;
    isWeekend: boolean;
    seasonalContext: string; // "none", "holiday_adjacent", "seasonal_topic", "cultural_moment"
  };

  // DISTRIBUTION DNA
  distribution: {
    hashtagCount: number;
    hashtagMix: string;     // "mostly_niche", "mostly_discovery", "balanced", "branded_heavy"
    mentionsCount: number;
    hasLocation: boolean;
  };

  // PERFORMANCE (filled after 7 days)
  performance?: {
    engagementRate: number;
    reachRate: number;
    saveRate: number;
    shareRate: number;
    commentRate: number;
    clickRate: number;
    percentileRank: number;
  };
}

DNA Profile (Per-Client Winning Formula)

interface DNAProfile {
  organizationId: string;
  platform: string;
  lastUpdated: Date;
  sampleSize: number;        // How many posts analyzed

  // Winning combinations (sorted by performance)
  winningFormulas: Array<{
    name: string;             // Auto-generated: "Educational List + Stat Hook + Morning"
    dnaSignature: Partial<ContentDNA>;
    avgEngagementRate: number;
    avgPercentileRank: number;
    occurrences: number;      // How many posts match this formula
    consistency: number;      // 0-1: how consistently this formula performs well
    lastUsed: Date;
    fatigue: number;          // 0-1: how overused this formula is becoming
  }>;

  // Losing combinations (what to avoid)
  losingFormulas: Array<{
    name: string;
    dnaSignature: Partial<ContentDNA>;
    avgEngagementRate: number;
    avgPercentileRank: number;
    occurrences: number;
    reason: string;           // LLM-generated explanation of why this fails
  }>;

  // Dimension-level insights
  dimensionInsights: {
    bestHookTypes: Array<{ type: string; avgPerformance: number }>;
    bestContentAngles: Array<{ angle: string; avgPerformance: number }>;
    bestVisualTypes: Array<{ type: string; avgPerformance: number }>;
    bestTimingSlots: Array<{ day: number; hour: number; avgPerformance: number }>;
    bestCaptionLength: { min: number; max: number; optimal: number };
    bestHashtagCount: { min: number; max: number; optimal: number };
  };
}

Database

model ContentFingerprint {
  id              String   @id @default(uuid())
  contentId       String   @unique
  organizationId  String
  platform        Platform
  dna             Json     // Full ContentDNA object
  performanceData Json?    // Filled after 7 days
  createdAt       DateTime @default(now())
  updatedAt       DateTime @updatedAt

  @@index([organizationId, platform])
}

model DNAProfile {
  id              String   @id @default(uuid())
  organizationId  String
  platform        Platform
  winningFormulas Json
  losingFormulas  Json
  dimensionInsights Json
  sampleSize      Int
  lastUpdated     DateTime @default(now())

  @@unique([organizationId, platform])
}

Building the Profile

// Runs weekly after Self-Evaluation completes

async function updateDNAProfile(organizationId: string, platform: Platform): Promise<DNAProfile> {
  // 1. Get all fingerprinted content with performance data
  const fingerprints = await prisma.contentFingerprint.findMany({
    where: { organizationId, platform, performanceData: { not: null } },
    orderBy: { createdAt: "desc" },
    take: 200,
  });

  if (fingerprints.length < 20) return; // Not enough data to build a profile

  // 2. Cluster high performers (top 25%)
  const sorted = fingerprints.sort((a, b) =>
    b.performanceData.percentileRank - a.performanceData.percentileRank
  );
  const topQuartile = sorted.slice(0, Math.floor(sorted.length * 0.25));
  const bottomQuartile = sorted.slice(Math.floor(sorted.length * 0.75));

  // 3. Find common DNA patterns in top performers
  const winningPatterns = findCommonPatterns(topQuartile);
  const losingPatterns = findCommonPatterns(bottomQuartile);

  // 4. LLM analysis to name and explain patterns
  const analysis = await analyzePatterns(winningPatterns, losingPatterns, fingerprints);

  // 5. Calculate dimension-level insights
  const dimensionInsights = calculateDimensionInsights(fingerprints);

  // 6. Calculate fatigue scores (winning formulas used too often lose power)
  for (const formula of analysis.winningFormulas) {
    const recentUses = fingerprints.filter(f =>
      matchesFormula(f.dna, formula.dnaSignature) &&
      f.createdAt > subDays(new Date(), 14)
    ).length;
    formula.fatigue = Math.min(recentUses / 5, 1); // Fatigue at 1.0 if used 5+ times in 14 days
  }

  // 7. Store profile
  return prisma.dNAProfile.upsert({
    where: { organizationId_platform: { organizationId, platform } },
    update: {
      winningFormulas: analysis.winningFormulas,
      losingFormulas: analysis.losingFormulas,
      dimensionInsights,
      sampleSize: fingerprints.length,
      lastUpdated: new Date(),
    },
    create: {
      organizationId,
      platform,
      winningFormulas: analysis.winningFormulas,
      losingFormulas: analysis.losingFormulas,
      dimensionInsights,
      sampleSize: fingerprints.length,
    },
  });
}

Content Creator Integration

When Content Creator generates a post, it receives the DNA profile:

// Injected into Content Creator's system prompt:

const dnaContext = `
CONTENT DNA PROFILE FOR ${platform}:

YOUR WINNING FORMULAS (create content matching these patterns):
${profile.winningFormulas
  .filter(f => f.fatigue < 0.8)  // Skip fatigued formulas
  .slice(0, 3)
  .map((f, i) => `${i + 1}. "${f.name}" (avg ${f.avgPercentileRank}th percentile, consistency: ${(f.consistency * 100).toFixed(0)}%)
   DNA: ${JSON.stringify(f.dnaSignature)}`)
  .join("\n")}

AVOID THESE PATTERNS (they consistently underperform):
${profile.losingFormulas.slice(0, 3).map(f =>
  `- "${f.name}": ${f.reason}`
).join("\n")}

TOP-LEVEL INSIGHTS:
- Best hooks: ${profile.dimensionInsights.bestHookTypes.slice(0, 3).map(h => h.type).join(", ")}
- Best angles: ${profile.dimensionInsights.bestContentAngles.slice(0, 3).map(a => a.angle).join(", ")}
- Optimal caption length: ${profile.dimensionInsights.bestCaptionLength.optimal} chars
- Optimal hashtags: ${profile.dimensionInsights.bestHashtagCount.optimal}

FATIGUE WARNING:
${profile.winningFormulas.filter(f => f.fatigue > 0.7).map(f =>
  `- "${f.name}" has been used ${Math.round(f.fatigue * 5)} times recently — mix in variety`
).join("\n") || "No formulas are fatigued — proceed normally."}

INSTRUCTION: Engineer new content to match a winning formula while keeping the surface-level content fresh and original. Vary the specific topic and words, but match the underlying DNA structure.
`;

Fatigue Management

The system prevents formula burnout:

Week 1: "Stat hook + educational list + morning post" → 7.2% engagement ✅
Week 2: Same formula → 5.8% engagement (still good)
Week 3: Same formula → 3.4% engagement (fatigue kicking in)
Week 4: System flags fatigue → rotates to "Story hook + problem-solution + evening post"
Week 5: Original formula rested → 6.9% engagement again ✅

Rules

1. DNA profiles require minimum 20 posts with performance data. Don't build profiles from insufficient samples.
2. Fatigue is real. Even winning formulas lose power when overused. Rotate formulas actively.
3. Surface variation is essential. Same DNA ≠ same content. The hook TYPE should match (question), but the actual question must be different every time.
4. Update weekly. Audience preferences shift. Last week's winning formula might fatigue this week.
5. Platform-specific profiles. What works on Instagram doesn't work on LinkedIn. Separate DNA profiles per platform.
6. The human sees this as "getting smarter." Mission Control Weekly Pulse: "I've identified that your audience responds 3x better to numbered lists with a statistic in the hook. I'm creating more of those."