Document Weight: How Heavy Is Your Page for AI to Lift?

Document Weight measures four dimensions of how heavy your HTML document is for AI crawlers to process:

1. Total HTML size: The raw byte count of the entire HTML response. This overlaps with Response Efficiency but contributes differently to the sub-score. 2. DOM node count: The number of HTML elements in the document. More nodes means more parsing work for crawlers. 3. Inline CSS + JS size: The combined byte count of all inline style and script blocks (excluding JSON-LD). 4. Large blob count: The number of individual inline style or script blocks that exceed 25KB.

The distinction from Response Efficiency is granularity. Response Efficiency looks at the total payload size. Document Weight breaks down where that weight comes from. A 300KB page could have 300KB of content (fine) or 30KB of content with 270KB of inline CSS/JS (bad). Document Weight catches the second case by penalizing inline bloat and DOM complexity separately.

The scorer calculates four sub-scores:

Total HTML (0-4 points): - Under 100KB: 4 points - 100-250KB: 3 points - 250-500KB: 2 points - 500-800KB: 1 point - Over 800KB: 0 points

DOM nodes (0-2 points): The scorer counts opening HTML tags using a regex pattern. This approximates DOM complexity. - Under 600 nodes: 2 points - 600-1200 nodes: 1 point - Over 1200 nodes: 0 points

Inline CSS + JS (0-2 points): The scorer extracts all <style> blocks and all <script> blocks (excluding those with type="application/ld+json") and sums their byte counts. - Under 20KB combined: 2 points - 20-80KB combined: 1 point - Over 80KB combined: 0 points

Large blobs (0-2 points): Any single inline style or script block over 25KB is a "large blob." These are individual payloads that significantly inflate the document. - 0 large blobs: 2 points - 1 large blob: 1 point - 2+ large blobs: 0 points

Maximum score: 10 (4+2+2+2).

Step 1: Externalize CSS and JavaScript

This is the single most impactful fix for Document Weight. Move all inline styles and scripts to external files. This reduces three sub-scores simultaneously: total HTML size, inline CSS/JS, and large blob count.

```html <!-- Heavy: 80KB inline --> <style>/* 45KB Tailwind utilities */</style> <script>/* 35KB app initialization */</script>

<!-- Light: 0KB inline, loaded externally --> <link rel="stylesheet" href="/styles.css"> <script src="/app.js" defer></script> ```

Step 2: Reduce DOM complexity

Pages with deeply nested component structures (common in React/Vue apps) generate excessive DOM nodes. Flatten unnecessary wrapper divs:

```html <!-- Unnecessary nesting: 4 nodes for one text --> <div class="wrapper"> <div class="container"> <div class="inner"> <p>Actual content</p> </div> </div> </div>

<!-- Flat: 1 node --> <p class="content">Actual content</p> ```

Step 3: Paginate long content

Pages with 50+ FAQ items, 100+ product listings, or very long articles generate thousands of DOM nodes. Consider paginating, using "load more" patterns, or splitting into multiple pages.

Step 4: Audit framework overhead

Server-side rendered frameworks often inject large state blobs. Check for __NEXT_DATA__, __NUXT__, or similar patterns. Reduce the data passed through server props to the minimum needed for initial render.

Step 5: Audit third-party embeds

Chat widgets, analytics dashboards, and social media embeds often inject large inline scripts. Move these to external loading patterns or lazy-load them after the main content renders.

Document Weight carries 1% weight in the Technical Foundation pillar. Combined with Response Efficiency (1%) and Critical Path Efficiency (1%), the Page Speed trio contributes 3% to the total score. These three criteria share an overlap group, meaning the fix plan consolidates their recommendations.

Most content-focused sites with external CSS/JS score 7-10/10. The failures concentrate in:

• SPA frameworks that inline large state objects: these inject 50-200KB of JSON into the HTML
• CSS-in-JS libraries that embed framework-generated styles: can add 40-100KB of inline CSS
• E-commerce pages with hundreds of product cards: DOM counts of 2000+ are common
• Long-form pages with embedded media and complex layouts: DOM complexity scales with content length

The most efficient fix path is externalizing inline CSS and JS. A single change (moving styles to an external file) can improve all three sub-scores: total HTML drops, inline CSS/JS drops, and large blob count drops.

AI crawlers parse HTML to extract content. The content-to-markup ratio determines how efficiently a crawler can find the useful parts of your page. A page where 90% of the HTML is actual content parses faster and more completely than a page where 90% is framework overhead.

GPTBot processes HTML by stripping markup and extracting text content. Pages with high DOM complexity and large inline payloads require more stripping work. While GPTBot handles this efficiently for individual pages, when crawling thousands of pages on a site, the overhead compounds.

ClaudeBot evaluates content density as a quality signal. Pages with a high content-to-markup ratio indicate content-first architecture, which correlates with higher content quality. Pages where the actual text content is a small fraction of the total HTML suggest framework-first architecture with content as an afterthought.

Perplexity's real-time processing is most sensitive to document weight because it processes multiple pages concurrently under time constraints. Lighter documents get processed more completely, meaning Perplexity is more likely to find and cite the specific passage that answers the user's query.

Google's crawling infrastructure handles document weight efficiently but still applies crawl budget optimization. Lighter pages consume less crawl budget, allowing Google to crawl more pages on your site and keep its index fresher.