Under the Hood: How Our AEO Scanner Benchmarks AI Search Readiness

**TL;DR / Executive Summary**

- Standard SEO tools measure backlinks and keyword density, completely failing to diagnose why AI answer engines like ChatGPT and Perplexity ignore or cite specific web pages.

- Our **AEO Scanner Tool** provides full transparency into the exact engineering metrics used to evaluate modern web pages: **Bi-Encoder Cosine Similarity**, **Stage 2 Cross-Encoder Rerank Score**, **Query-Passage Alignment**, and **Chunk Extraction Efficiency**.

- **AEO Impact:** Understanding these four metrics allows content teams to audit and optimize content for neural retrieval pipelines *before* publishing.

Why Standard Technical SEO Audits Fall Short for AI Search

Traditional SEO audits analyze meta tags, page load speed (LCP/CLS), and backlink profiles. While these metrics remain important for traditional crawler indexing, **they provide zero visibility into how Large Language Models (LLMs) evaluate web content during Retrieval-Augmented Generation (RAG)**.

An AI search engine does not rank pages based on domain authority alone. It retrieves discrete, semantically relevant text chunks from across the web and filters them through a multi-stage neural pipeline.

To give digital marketers, content strategists, and technical SEOs total visibility into this process, our **AEO Scanner Tool** evaluates web pages across **4 Core Engineering Metrics**.

The 4 Core Metrics of the AEO Scanner

                             [ AEO Readiness Index ]
                                       │
       ┌───────────────────────┬───────┴───────────────┬───────────────────────┐
       ▼                       ▼                       ▼                       ▼
(1) Bi-Encoder Cosine   (2) Stage 2 Cross-Encoder  (3) Query-Passage      (4) Chunk Extraction
    Similarity (0-100%)     Rerank Score (0-100%)      Alignment Score        Efficiency Index

Metric 1: Bi-Encoder Cosine Similarity (0–100%)

What It Measures

Bi-Encoder Cosine Similarity measures the broad, high-dimensional vector proximity between a target search prompt $E(Q)$ and your web page chunk vector $E(P)$.

The Engineering Formula

$$\text{Cosine Similarity} = \frac{E(Q) \cdot E(P)}{\|E(Q)\| \|E(P)\|}$$

Benchmark Scale

• **85% – 100% (High Match):** Your content vector sits directly in the target candidate cluster.
• **50% – 84% (Moderate Match):** Broad conceptual overlap, but lacks specific terminology.
• **Below 50% (Low Match):** Your chunk will not make the Stage 1 vector retrieval candidate pool.

Metric 2: Stage 2 Cross-Encoder Re-rank Score (0–100%)

What It Measures

Once a page passes Stage 1 vector retrieval, our scanner runs candidate chunks through a **Joint-Attention Cross-Encoder Simulation**. This model evaluates exact word-for-word interaction, heading co-occurrence, and passage concentration.

How the Re-rank Score is Calculated in Code

As implemented in our core scanning action ([`aeo-scan.ts`](file:///c:/Users/jason/OneDrive/Documents/Work/Unparallel/unparallel-site/src/app/actions/aeo-scan.ts)):

function computeCrossEncoderRerank(query: string, heading: string, text: string, biEncoderScore: number): number {
  let rerankScore = biEncoderScore;
  const qTerms = query.toLowerCase().split(/\s+/).filter(w => w.length > 2);

  // 1. Heading Intent Alignment Boost (+12)
  const headingMatch = qTerms.filter(t => heading.toLowerCase().includes(t)).length;
  if (headingMatch > 0) {
    rerankScore += Math.round((headingMatch / qTerms.length) * 12);
  }

  // 2. Term Co-occurrence in single sentence (+15)
  const sentences = text.toLowerCase().split(/[.!?]+/);
  let maxSentenceMatch = 0;
  sentences.forEach(s => {
    const matchCount = qTerms.filter(t => s.includes(t)).length;
    if (matchCount > maxSentenceMatch) maxSentenceMatch = matchCount;
  });

  if (maxSentenceMatch >= Math.min(3, qTerms.length)) {
    rerankScore += 15;
  } else if (maxSentenceMatch > 1) {
    rerankScore += 8;
  }

  // 3. Passage Length Concentration (+8 or -10)
  const wordCount = text.split(/\s+/).filter(Boolean).length;
  if (wordCount >= 60 && wordCount <= 280) {
    rerankScore += 8;  // Density reward
  } else if (wordCount > 400) {
    rerankScore -= 10; // Fluff penalty
  }

  return Math.min(100, Math.max(0, Math.round(rerankScore)));
}

Metric 3: Query-Passage Alignment Score

What It Measures

Query-Passage Alignment checks whether a single paragraph block provides an **immediate, standalone answer** to the implied intent of its parent heading.

What Ruins Alignment:

• **Delayed Answers:** Preamble filler placed before the primary answer.
• **Fragmented Terms:** Key query entities split across distant paragraphs.

Metric 4: Chunk Extraction Efficiency Index

What It Measures

Chunk Extraction Efficiency evaluates the **Signal-to-Noise Ratio (SNR)** of your text chunks. It flags chunks that contain excessive boilerplate text, unformatted lists, or inline ad code that pollute LLM context windows.

[ Clean Paragraph Chunk ]  ===> 90% Signal / 10% Noise ===> High Extraction Score
[ Bloated DOM Chunk ]      ===> 30% Signal / 70% Noise ===> Low Extraction Score

The Composite AEO Readiness Index

The AEO Scanner combines these four metrics into a single **Composite AEO Readiness Index (0–100 Score)**:

$$\text{Composite AEO Index} = 0.30(\text{Cosine Similarity}) + 0.40(\text{Cross-Encoder Rerank}) + 0.15(\text{Alignment}) + 0.15(\text{Efficiency})$$

| Score Range | AEO Status | Action Required |

|---|---|---|

| **85 – 100** | **AI Search Ready** | Fully optimized for top RAG candidate selection & LLM citations. |

| **65 – 84** | **Moderate Visibility** | Good vector similarity, but requires tighter heading alignment. |

| **Below 65** | **Low Citation Probability** | High risk of being filtered out during Stage 2 Cross-Encoder re-ranking. |

How to Action Your AEO Scanner Audit Results

1. **Fix Paragraph Lengths:** Ensure primary answer paragraphs stay within **60 to 280 words**.

2. **Align Headings to Queries:** Rewrite vague headings (e.g. *"Overview"*) into explicit, intent-driven questions (e.g. *"How Does Bi-Encoder Search Work?"*).

3. **Eliminate Introductory Filler:** Remove preamble sentences to boost your Signal-to-Noise Ratio.

Benchmark Your Website's AEO Readiness

Stop guessing how AI search engines evaluate your content.

**Get Full Engineering Visibility Today:**

Audit your URL using our AEO Scanner to receive a complete breakdown of your Cosine Similarity, Cross-Encoder Rerank Scores, and Chunk Extraction Efficiency.

**[Run Your Free AEO Audit Now](#)**