From 14K to 249K Indexed Videos: Diagnosing Video Indexing at Scale

The symptom

The client, a scientific video publisher, had a catalogue of approximately 30,000 video pages spanning 14 languages and multiple product lines: journal articles, educational content, and lab manuals. Despite proper VideoObject structured data implementation, Google was indexing roughly 1,000 videos per product type on average. Across the entire corpus, fewer than 14,000 pages had their video content recognised in Google's Video indexing report.

Video results drive higher engagement and click-through rates for scientific content queries, particularly in educational contexts. Missing out on video indexing meant missing the rich result carousel entirely, ceding visibility to competitors whose video content was indexed.

The internal team had already verified the structured data. JSON-LD was valid. Thumbnails were accessible. Video files existed. Yet the indexing needle barely moved.

The diagnosis

The first step was ruling out the obvious suspects. I ran a systematic audit across indexed and non-indexed pages to identify correlations:

What wasn't causing the problem:

• Structured data: JSON-LD was valid and complete across all pages. Both indexed and non-indexed pages had identical VideoObject markup. No correlation.
• Content age: Old journal articles from 2008 were indexing. New 2024 content wasn't. And vice versa. No age correlation.
• Traffic and engagement: High-traffic pages weren't preferentially indexed. Low-click educational pages were sometimes indexed while popular content was ignored.
• Product type: The issue affected all collections equally: journal videos, educational content, lab manuals.

The breakthrough came from examining Google's rendering of non-indexed pages versus indexed ones. Using URL Inspection's rendered HTML output, I compared page structure across both sets.

The pattern emerged: on pages where Google indexed the video, the video player was prominently positioned and the surrounding text content was minimal. On non-indexed pages, the video player was buried beneath substantial text content (abstracts, descriptions, methodology sections) competing for visual prominence.

Server log analysis revealed a secondary issue: inconsistent delivery of video sample files. Different CDN edge servers were returning conflicting headers for the same video assets, potentially confusing Google's video processing pipeline.

The constraints

This wasn't a simple structured data fix:

• 30,000 pages across 14 languages: Any solution needed to scale automatically. Manual intervention was impossible.
• Established page templates: The existing layout served user needs. Researchers wanted methodology details, abstracts, and supporting text. Simply removing content wasn't acceptable.
• Multiple product lines: Journal articles, educational videos, and lab manuals each had different template requirements and stakeholder expectations.
• No guarantee of outcome: Google's video indexing criteria aren't documented. We were working from observed patterns, not confirmed specifications.

The approach

Based on the diagnostic findings, I proposed a two-phase strategy focusing on video prominence and technical delivery:

Phase 1: Video sample delivery (weeks 1–4)

The inconsistent CDN headers were causing Google's video crawler to encounter different responses for the same video sample. We standardised server configuration across all edge locations:

• Unified Content-Type headers for video samples
• Consistent Accept-Ranges support for byte-range requests
• Proper Cache-Control directives for video assets

Phase 2: Page restructuring for video prominence (weeks 4–12)

Rather than removing valuable content, we restructured the page hierarchy:

1. Video player moved above the fold: The player became the dominant visual element on initial page load
2. Text content collapsed by default: Abstracts, methodology, and detailed descriptions moved into expandable sections, available on interaction rather than competing for initial render prominence
3. Reduced above-fold text density: Supporting information that previously surrounded the player was relocated below or made collapsible

The template changes rolled out progressively: English pages first, then international variants language by language to monitor impact.

Tech stack: Cloudflare CDN, server-side template modifications, Google Search Console Video indexing report

The result

Within six months of the final rollout:

• Video indexing: From approximately 14,000 to 249,000 indexed videos (18x increase)
• Coverage: Video indexing now covers the majority of the 30,000-page video catalogue
• Non-indexed pages: Reduced from the majority to approximately 75,000 pages, many of which are legitimately non-video content or recent additions awaiting crawl

The growth followed a predictable pattern: initial gains appeared within weeks of the first template changes, with sustained improvement as Google re-crawled the expanded page set over subsequent months.

Key takeaways

1. Valid structured data isn't sufficient for video indexing. Google evaluates page structure and video prominence independently of JSON-LD validity. A technically correct VideoObject implementation won't index if the video isn't visually prominent on the page.

2. Diagnostic elimination saves time. By systematically ruling out structured data, content age, traffic levels, and product type, we avoided weeks of wasted effort on fixes that wouldn't have moved the needle.

3. Page restructuring can satisfy both users and crawlers. Collapsible content sections preserved all the information researchers needed while making videos prominent for Google's evaluation. The solution wasn't removal; it was hierarchy.