YouTube SEO: How to Stop Your Videos from Flapping in the Wind

Most creators think a "bad" video is the reason they aren't getting views, but usually, the algorithm just doesn't know who to show the video to. If your metadata is vague, you're essentially gambling with your reach. The real bottleneck isn't always the content quality—it's the gap between your video's intent and the search terms users actually type into the search bar.

To fix this, you need a real-world AI workflow to optimize your titles and descriptions before you hit publish. I've been experimenting with prompt engineering to turn raw transcripts into high-CTR metadata.

Here is a practical tutorial on how to use an LLM agent to optimize your YouTube reach:

1. Feed the Context: Don't just ask for a "catchy title." Paste your entire video script or a detailed outline into the AI.
2. Define the Persona: Tell the AI to act as a YouTube growth expert who specializes in A/B testing titles.
3. Apply the Prompt: Use a specific framework to generate options.

Act as a YouTube SEO expert. I will provide a video transcript. 
Your goal is to generate 5 high-CTR titles and a 200-word description.
- Titles must be under 60 characters.
- Use "curiosity gaps" without being clickbait.
- Include the primary keyword [Insert Keyword] naturally in the first sentence of the description.
- Identify 10 long-tail keywords for the tags section.

4. Refine and Validate: Take the top 3 suggestions and check them against YouTube's auto-suggest search bar to see if people are actually searching for those phrases.

This approach transforms the process from guessing to a data-driven deployment. Instead of spending hours staring at a blank screen, you're using a complete guide for metadata that actually aligns with how the recommendation engine works. It's a massive time-saver for anyone trying to scale a channel from scratch.