How You Can Grow Your YouTube Channel with Claude
You will connect Claude Code to your YouTube channel, run one prompt that sorts every public video by impressions, click-through rate and retention, and get a 3D dashboard showing which topics to repeat and which videos need new packaging.
What you need
Claude Code (or Codex)
A YouTube API key from Google Cloud
YouTube OAuth access to your channel
A channel with some published videos
Step by step
Learn the three metrics you are diagnosing
Sort every problem into one of three questions: - Impressions: is YouTube showing the video? If not, change the topic. - CTR: do people click? If not, fix the packaging (thumbnail plus title). - Retention: do people stay? Watch the first thirty seconds, where most viewers drop.
Get your YouTube keys
Create a YouTube API key in Google Cloud, then set up YouTube OAuth for your channel. Together they give Claude Code (or Codex, or any AI system you choose) remote access to your YouTube account and analytics.
Google Cloudhttps://console.cloud.google.com
Test the connection
Open Claude Code and ask for your latest videos. If it lists them correctly, the connection works.
Test promptHey, can you check my YouTube channel using the keys and tell me what are the three last videos that I had?
Leave out unlisted and private videos
Check the answer: the channel data includes unlisted and private videos. If you have any, say clearly in your next prompt that you only want public videos analyzed.
Send the analysis prompt
Pick a strong model at high effort for the main session (I use Fable 5 on high) and ask for Opus 4.8 subagents so it runs faster. Then paste the prompt below in one go.
Analysis promptHey Claude, I want you to get all of the public videos from my YouTube channel. Then use a workflow to analyze all of the videos, their data from the moment they were released, detect which ones had a very good conversion rate, which ones had a very good amount of impressions, and categorize the videos in a 3D graph, showing me the ones that were bad because there were not enough impressions, the ones where people did click on the video, and the videos 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. And remember, this should be done in 3D.
Let the dynamic workflow run
Wait while Claude fetches the data. The dynamic workflow gives each video its own subagent (agent insight batch one, two, three, and so on), with an insights task and then a synthesis task. In Codex, ask for subagents instead.
Read the five categories on the dashboard
Open the HTML page Claude builds. Each video lands in one category: - Winners - Clicked, then left (low retention) - Stayed but few clicked (low CTR or impressions) - Seen, no click, no hold - Invisible (almost no impressions)
Repeat the topics that won
Look at the cluster up and to the right of the 3D graph. Those are winners you can reproduce. Pick topics that are still relevant, such as a follow-up on something that is coming soon.
Repackage low-CTR videos
Filter for videos with good impressions but low click-through rate. Redo them with a different hook, title or thumbnail. My example: 400 views at 1.6% CTR, where multiplying CTR by four would mean at least 2,000 views.
Watch out for
Private and unlisted videos are included by default. Tell Claude to analyze public videos only.
Low impressions are hard to fix with AI: they depend on the topic, so change the topic rather than the packaging.
The written insights are long. Use the visual dashboard to decide, and skim the text only to confirm it is relevant.
Use AI for the analysis, not to replace you on camera: people need to see you are an actual human.
