A system for generating YouTube video descriptions relied on static text extraction from video scripts, resulting in unstructured, SEO-deficient descriptions that lacked keywords, structured content, and algorithm signals. This approach failed to optimize visibility for both horizontal videos and Shorts formats.

Issue

The descriptions generated for YouTube videos were limited to a hook question and source URL, lacking essential SEO elements. Specifically:

Horizontal videos: No keywords, structured sections (e.g., “What You’ll Learn”), or calls-to-action.
Shorts: Descriptions contained only a full video link with no technical context or algorithm categorization.

This resulted in poor search rankings and reduced visibility in YouTube’s recommendation algorithm.

Culprit

The system used a simple text extraction approach that:

Captured only the first paragraph of the script, typically a hook question.
Had no integration with language models (LLMs) for SEO optimization.
For Shorts, replaced descriptions with just a full video link without additional context.
Relied on hardcoded string parsing rather than content-aware generation.

This approach failed to incorporate keywords, structured content, or algorithm-friendly metadata.

Solution

Two-Tier LLM Description System

A new system was implemented to generate SEO-optimized descriptions using LLMs, tailored for both horizontal videos (400-600 characters) and Shorts (100-150 characters).

1. Prompt Engineering

Structured templates were created with SEO best practices:

Keywords placed in the first two sentences.
Use of technical terms relevant to the target audience.
Avoidance of clickbait or overselling.
Inclusion of structured content like bullets and hashtags.

2. Integration with LLM Tools

New LLM tools were added to handle description generation:

One tool for generating descriptions for horizontal videos.
Another tool for generating descriptions for Shorts, optimized for algorithm visibility.

3. Orchestrator Integration

The system was updated to replace static text extraction with LLM-based generation, ensuring:

Descriptions for horizontal videos include keywords, structured sections, and calls-to-action.
Descriptions for Shorts provide algorithm context and technical terms.

4. Shorts Optimization

Descriptions for Shorts were enhanced to include both a conversion link and algorithm-friendly context, ensuring better categorization by YouTube’s recommendation algorithm.

Key Improvements:

Horizontal videos now feature structured descriptions with keywords, bullets, and hashtags.
Shorts include algorithm-focused descriptions with technical terms and relevant context.
Performance optimizations to reuse existing LLM client configurations.