- Genial
- AI Automation Tutorials
- How You Can Grow Your YouTube Channel with Claude
How You Can Grow Your YouTube Channel with Claude
Most creators look at views and guess what to change next. Today I am going to show you how to use Claude to analyze all of your videos, understand what is working correctly, and spot what could use some repackaging. We will connect Claude Code to my own channel, classify every public video, and turn the results into a visual dashboard.
What you need
- A YouTube API key from the Google Cloud platform
- A YouTube OAuth connection, which gives full remote access to your YouTube account from any AI system you want: Claude Code, Codex, whichever you choose
- Claude Code (if you need help getting the keys, drop a comment on the video and I will send you a tutorial)
1. The three metrics that decide video performance (0:00)
Everything that matters comes after the moment you upload. There are three key elements. First, impressions: how many people have actually seen your video on their feed. This is where AI cannot do much, because impressions depend on the trending topic; if you did not reach as many people as you wanted, you have to change the topic. Second, CTR, the click-through rate: people see your video but do not click, which means you work on the packaging, the thumbnail plus the title. Third, retention: once people click, they have to stay, and the first thirty seconds are where people drop out.
So the three questions to optimize for are simple. Do people see my content? Do people click on my content? And do people stay to watch my content?
2. Connect Claude to your YouTube channel (1:33)
For the record, my channel is quite young, with a few videos that hit a couple of thousand views on a huge amount of impressions and a good click-through rate. What I want to figure out is what I did right and where I went wrong.
With the API key and OAuth in place, I start with a sanity check that the connection works:
Hey, can you check my YouTube channel using the keys and tell me what are the three last videos that I had?
Because my account is connected through those keys, Claude checks everything and lists my three latest videos. One thing to watch out for: the listing includes unlisted and private videos. If you have those, make sure to say you do not want them analyzed.
3. Build the analysis with one prompt (3:39)
Now we build the entire system in one single prompt. I am using Fable 5 on high settings for effort, and I am asking it to use Opus 4.8 for the subagents, which makes everything go a lot faster:
Get all of the public videos from my YouTube channel. Then use a workflow to analyze their data from the moment they were released. Detect which ones had a very good conversion rate and which ones had a very good amount of impressions, and categorize the videos in a 3D graph: the ones that were bad because there were not enough impressions, the ones where people clicked on the video, and the ones where people stayed. Then give me insights per video on what I should do with that topic. Please use a dynamic workflow and use Opus 4.8 agents. Remember, this should be done in 3D.
4. Why dynamic workflows improve the analysis (4:51)
The reason we use a dynamic workflow, which literally means Claude spins up a bunch of subagents, is that each video gets its own dedicated agent, and that helps the quality of your analysis. Whenever you run analysis on huge amounts of data, hundreds or thousands of documents, this is the feature of Claude Code you want. You can do the same in Codex by the way; there it is not called workflows, just subagents.
Claude breaks the job into stages. In the Claude application you can watch the workflow run: agents for insight batch one, batch two, batch three, and the workflow tasks themselves, first insights, then synthesis.
5. Explore the finished dashboard (6:07)
I asked for an HTML page to visualize the results, and Claude delivered that plus a full analysis of all my videos. The dashboard splits everything into five performance categories:
- Winners
- Clicked, then left (low retention)
- Stayed, but few clicked (low conversion or impressions)
- Seen, no click, no hold (the bad videos)
- Invisible (no impressions at all)
On the 3D graph, the winners cluster up and to the right, and it is immediately clear which ones are worth reproducing.
6. Find topics worth repeating (6:45)
In my case, the winners include building a CRM with Claude Code and my Gemini 3.6 Flash video. Claude Opus 5 was coming within a couple of weeks at the time, so that was an obvious candidate: a topic in my winning cluster with a fresh news hook. That is the whole point of the exercise, finding repeatable topics instead of guessing.
7. Fix videos with a low click-through rate (7:15)
Then I look at videos with good impressions but low CTR. Example: "Is Sonnet 5 Actually Worth It?" got four hundred views with a click-through rate of one point six percent. If I rework the packaging, a different hook, title, or thumbnail, and multiply that CTR by four, the same video should pull at least two thousand views. The insights Claude wrote per video are relevant, but Fable 5 and Claude in general are very verbose. Huge paragraphs you do not want to read are exactly why the visual dashboard is the better interface for understanding your data.
8. Apply the method to every social platform (8:28)
Do not stop at YouTube. The same approach works for Instagram, TikTok, LinkedIn, really anything, because AI is fantastic at recognizing patterns us humans cannot even see. That is how you implement it across your whole social media strategy. One recommendation: use AI for the analysis, not for the content creation. Right now it is extremely important to keep your face up front so people realize you are an actual human.
Pitfalls and tips
- The API fetch includes unlisted and private videos by default. Exclude them explicitly if you only want public data analyzed.
- Expect verbose written insights. Ask for a visual dashboard so the patterns are actually readable.
- Impressions problems are topic problems; CTR problems are packaging problems. Do not fix the wrong one.
Where to go next
- Give Claude eyes on your actual footage with Stop Wasting Hours: Analyze Any Video with AI.
- Make workflows like this cheaper to run with Claude Fable 5 Is Here to Stay - Maximize Your Usage.


