Video content for AI search: my automated pipeline with Claude Code, SE Ranking MCP and YouTube API

Ahrefs data from September 2026 puts YouTube at 22.9% of AI Overview citations. Otterly.ai found in March 2026 that 94% of YouTube citations go to long-form videos. These are just some of the many many studies that came out this year that underline how much YouTube is retrieved in AI search.

If you want your brand in AI answers, and AI answers in your sector heavily cite YouTube, long-form video deserves a place in your content plan and organic strategy. The catch is that a language model reads your video as text, through the transcript, title, description and chapters. Those fields are everything the model knows about your video, so they really need the same care you’d give a landing page.

How the AI pipeline works

Everything starts from one slash command in Claude Code, with four inputs:

/video-pipeline

Topic: ...
Short form: IG reel / YT short / FB reel / TikTok reel
Long form duration: XX minutes
Notes: evergreen content, YouTube channel name, channel link, market and language, tone, angle, ...

I fill in the topic or main keyword, the short-form platform, the target length of the long-form video, and optional notes on tone, angle or brand, and that’s all I type.

Claude Code acts as the orchestrator, it pulls SEO data through the SE Ranking MCP, benchmarks competitors through the YouTube API, and chains seven sub-skills in order. Each skill does one job and hands its output to the next one.

The 7 steps

Research

Step 1 is keyword research: Claude queries SE Ranking MCP for related keywords, question keywords and long-tail variations, then groups them by search intent.

Step 2 checks what Google shows for the topic today on the SERP: People Also Ask, featured snippets, the AI Overview and the video pack.

Step 3 benchmarks the competition on YouTube: the YouTube API returns the top 10 videos by views, and Claude analyses their hook patterns and where their structure leaves gaps.

Creation

Step 4 writes the long-form script: the word budget is calibrated on the target duration, so a ten-minute video gets a ten-minute script, and every claim goes through fact checking with web search.

Step 5 adapts the long-form into one short-form version for the chosen platform, with the hook landing in the first three seconds.

Publishing

Step 6 produces the title, description, tags, chapters and internal links, for the long and short form. Then it runs a check to see whether the priority keywords and findings from the research phase made it into the script and metadata, and if all the AI retrievability rules were applied correctly.

Post-publication

Step 7 audits the transcript for errors and inconsistencies once the video is filmed and uploaded. 

Why the transcript audit is a must

YouTube’s auto-generated captions can mangle brand names, entity names, industry jargon, punctuation, grammar and so on. Especially if the language is not English!

If AI reads your video through the transcript, a misspelled brand name is a mention that may never get connected to you, as you may know, in terms of entity alignment, definitely not the best.

Step 7 checks the transcript, with the original script and the brand context as benchmark, and flags grammar errors, spelling, punctuation, proper names and industry jargon, so you can fix everything in YouTube Studio and have a spotless transcript.

What the /video-pipeline produces

Every run ends with two files, mostly because two different people use them.

  • video-package-[topic-slug].md is the research package: it collects all the data from the analysis phase and run in one markdown file, for whoever owns SEO.
  • video-script-[topic-slug].docx is the editorial document: it holds only scripts, descriptions and metadata, for the content or production team, so they can review, film and publish without scrolling past keyword tables and JSON files.

What changed, and what I learned

I created this AI workflow because I saw a pain point and optimization opportunity in my team’s everyday work. Which by the way is where some of the best AI automations come from! A job that took my team one day now takes about two hours: around 30 to 50 minutes for the pipeline to complete (depending on the topic), and an hour on top to do all the necessary checks and adjustments from specialists and industry experts.

Some insights I got from building and using this pipeline:

  • Every skill works better with a fixed output template, because runs stay consistent and are easier to review. That’s the T in my SCRIPT framework, and it applies to skills as much as prompts.
  • Someone who knows the topic still has to sign off before anything gets filmed. The pipeline researches and drafts, and the expert check stays mandatory.
  • Garbage in, garbage out. A vague topic or thin notes produce a generic script, and the more context I put in the input, the better the output.
  • The first versions of each skill weren’t good. Running the pipeline on real topics, reviewing the output and feeding corrections back into the skills is what made it usable. Long live the feedback runs!
  • Finally, it only runs locally because the YouTube API key lives in a script on my machine. So sharing it with a team takes extra setup.

Final note

I really want to highlight the fact that the more personalized the pipeline is for your own brand or your client’s one, the better the output. I’m talking about giving the model real examples of already published gold standard videos, writing rules, structure rules, etc.

If you want to recreate it, give this article (and the deck as well if you want) to your AI chatbot of choice along with this prompt:

## Situation
I've attached an article describing a 7-step AI pipeline that plans and scripts YouTube videos for SEO and AI search: keyword research, SERP and AI Overview analysis, YouTube competitor benchmark, long-form script, short-form adaptation, metadata with a coverage check, and a post-publication transcript audit.

## Character
Act as an AI workflow engineer with solid knowledge of SEO, YouTube and prompt design.

## Request
Help me build my own version of this pipeline, adapted to my brand, market and tools.

## Instructions
1. Before writing anything, ask me about: my brand and YouTube channel, target market and language, the SEO data tool I use, whether I have a YouTube Data API key, the AI tool I'll run the pipeline in, and 2-3 published videos I consider gold standard.
2. Based on my answers, tell me which of the 7 steps I can run as they are, which need adapting, and which to skip for now.
3. Build one step at a time, starting with step 1. For each step, write the prompt or skill and its fixed output template, then wait for me to test it before moving on.
4. Keep a human review mandatory before filming, and flag anything I need to verify.

## Proof
Use the article as the reference for each step's scope and outputs. Use my gold-standard videos as the benchmark for tone, structure and hook style.

## Template
For each step, give me:
- Step name and goal
- Inputs it needs
- The full prompt or SKILL.md
- The output template it must follow

Resources!

Share it!