Aura™ GEO Audit Methodology

1. System Overview & Architecture

1.1 What is Aura™?

Aura™ is a production-grade Generative Engine Optimization (GEO) audit system that analyzes a business's visibility in AI-powered search results. Unlike traditional SEO tools that measure Google rankings, Aura™ measures how likely your business is to be cited by large language models (LLMs) like ChatGPT, Google AI Overviews, Perplexity, and Gemini.

1.2 Core Innovation: Modular Architecture

Traditional GEO tools use a single AI prompt to generate entire reports. This "monolithic" approach suffers from:

• Low accuracy (~70-85%) due to task complexity
• Slow generation (20-30 seconds, sequential processing)
• No differentiation between business types
• Difficult to debug or improve individual components

Aura™ v3.0 implements a modular prompt decomposition architecture based on the "Least-to-Most Prompting" research (Zhou et al., 2023). The audit task is decomposed into 10 independent subtasks, each with:

• Specialized prompts: Task-specific instructions and constraints
• Schema validation: Zod runtime validators ensure type safety
• Parallel execution: 3 concurrent API calls (2.5x faster)
• Error isolation: Single section failure doesn't break entire report

User Input → Archetype Detection → 
  ┌─────────────────────────────────────────────┐
  │     Modular Prompt Engine (10 Sections)     │
  ├─────────────────────────────────────────────┤
  │  [Technical Health] [Brand Authority]       │
  │  [Sentiment]  [Social Media]  [Content]     │
  │  [AI Gaps]  [Roadmap]  [Summary]            │
  │  [GAS Score]  [AVS Score]                   │
  │                                             │
  │  Parallel Execution (3 concurrent)          │
  │  ↓ Zod Validation (per section)            │
  └─────────────────────────────────────────────┘
            ↓
  ┌─────────────────────────────────────────────┐
  │        Reconciliation Layer                  │
  │  • Normalize percentages (sentiment = 100%) │
  │  • Validate correlations (AVS ≈ GAS + 3)   │
  │  • Enforce business rules                   │
  └─────────────────────────────────────────────┘
            ↓
  ┌─────────────────────────────────────────────┐
  │    Local Context Adjustment (if applicable) │
  │  • Detect hyper-local market               │
  │  • Apply score boosts (+8-10 points)       │
  │  • Add asterisk notation                   │
  └─────────────────────────────────────────────┘
            ↓
  ┌─────────────────────────────────────────────┐
  │       Baseline Comparison (if exists)       │
  │  • Compare to previous scans               │
  │  • Calculate deltas                        │
  │  • Show consistency rating                 │
  └─────────────────────────────────────────────┘
            ↓
         Report Output + Survey
        

1.3 Key Metrics Generated

2. Modular Prompt Decomposition

2.1 The 10 Specialized Sections

2.2 Parallel Execution Strategy

Sections are processed in batches of 3 (concurrency limit) to:

• Avoid API rate limits (Gemini: 60 requests/minute)
• Maximize throughput (3 concurrent vs 1 sequential)
• Reduce total latency from ~20s to ~14s (30% improvement)

const CONCURRENCY_LIMIT = 3;

for (let i = 0; i < SECTION_ORDER.length; i += CONCURRENCY_LIMIT) {
  const batch = SECTION_ORDER.slice(i, i + CONCURRENCY_LIMIT);
  const promises = batch.map(sectionId => generateSection(sectionId, context));
  await Promise.all(promises);  // Wait for batch to complete
}

// Result: 10 sections in 4 batches (⌈10/3⌉ = 4)
// Total time: 4 × 3.5s = 14s (vs sequential: 10 × 3.5s = 35s)
        

3. Business Archetype Detection

3.1 Why Archetypes Matter

Through empirical testing of 11 businesses, we discovered that business archetype is a stronger predictor of GAS than business scale.

3.2 The 5 Business Archetypes

3.3 Detection Algorithm

Business archetype is inferred from domain, business name, and industry keywords using a decision tree classifier:

if (domain in ['zillow.com', 'amazon.com', 'homes.com']) 
  → data-platform

else if (keywords.includes('national') && keywords.includes('brokerage'))
  → national-brokerage

else if (keywords.includes('luxury') && city_population > 100K)
  → regional-brokerage

else if (keywords.includes('discount') || keywords.includes('percent'))
  → disruptor

else if (city_population < 50K || keywords.includes('local only'))
  → small-local

else
  → local-established (default)
        

Validation: 11/11 businesses classified correctly (100% accuracy)

4. Empirical Validation & Testing

4.1 Test Sample Selection

We tested Aura™ on 11 real businesses representing diverse scales and industries:

4.2 Statistical Validation

5. Scoring Formulas & Correlations

5.1 Primary Formula: AVS from GAS

5.2 Archetype-Specific GAS Formulas

7. Local Context Adjustments

7.1 The Hyper-Local Problem

7.2 Adjustment Formula

if (isHyperLocalMarket && authority >= 70 && authority < 80) {
  adjustedAuthority = authority + 10;  // 72 → 82*
  adjustedGAS = gas + 8;               // 60 → 68*
  adjustedAVS = avs + 10;              // 58 → 68*
  
  notation = "* Adjusted for local market context";
  relativeRanking = "Likely #1-2 in [city]";
}
        

Detection Criteria: City population < 50K OR keywords include "small town", "parish", "neighborhood" OR estimated competitors < 15

8. Continuous Learning System

8.1 Baseline Tracking

Aura™ stores baseline reports for known businesses and compares new scans to detect:

• Accuracy drift: Are scores changing systematically?
• Consistency: Same business → similar scores
• Improvements: Did our formula refinements help?

8.2 User Survey Integration

After each audit, users are invited to complete a 3-question survey:

Target Completion Rate: 60% (3 questions optimizes completion vs data quality)

9. Accuracy Metrics & Statistical Validation

9.1 Component-Level Accuracy

9.2 Overall System Accuracy