Podcast: Download (Duration: 43:45 — 50.3MB)
To analyze your YouTube channel with Claude Code, connect Claude Code to the YouTube Data API and the YouTube Analytics API, have it pull a year of data, and ask what is working and what is failing.
When I ran this on my own channel, Claude found that every video with a backstory section between the hook and the first point lost 15 to 20 retention points.
The worst part is that I caused it. I trained Claude on every script I had ever written, including years of flops, so it learned to write the exact intro that was costing me viewers.
In this episode, my co-host Toni Herrbach and I walk through the whole process. Toni wanted to figure out why a client’s channel had been stuck under 10,000 subscribers for about 20 months, so I shared exactly what Claude told me about my channel and which advice I am ignoring on purpose.
Get My Free Mini Course On How To Start A Successful Ecommerce Store
If you are interested in starting an ecommerce business, I put together a comprehensive package of resources that will help you launch your own online store from complete scratch. Be sure to grab it before you leave!
Table of Contents
Key takeaways
- Connect two APIs. Claude Code needs both the YouTube Data API and the YouTube Analytics API, and setup takes about 30 minutes the first time.
- Cut the backstory from your intros. Videos with a backstory beat between the hook and the first point lost 15 to 20 retention points.
- Train your AI only on winning scripts. Feeding it every old script teaches it to repeat the habits of your flops.
- One topic word can sink a video. On my channel, putting “TikTok Shop” in a title consistently flopped.
- Refresh old hits before making new videos. A new title and thumbnail on a two-year-old winner can be worth more than a new upload.
- Views and email signups are different goals. A recent video that flopped on views got about 50 times the email signups of a typical video.
How to connect Claude Code to your YouTube channel
You connect Claude Code to YouTube by setting up two Google APIs: the YouTube Data API and the YouTube Analytics API. Google makes this more complicated than it needs to be, because you have to create credentials inside the Google Cloud console.
The easiest way through is to tell Claude Code that you want to connect to your YouTube channel and ask it to walk you through the process. Expect it to take about half an hour the first time, and a little longer if you have never touched the Google Cloud console.
What to do when you get stuck on the YouTube API setup
When you get stuck, ask Claude to explain the step differently until it makes sense. Toni does this constantly, asking Claude to break a step down further or to describe exactly which button to click.
When Claude tells you to click a menu option that does not exist, take a screenshot and paste it in. Telling Claude “I don’t see that option” with a screenshot usually gets you a corrected instruction right away.
What Claude Code can analyze on a YouTube channel
Claude Code can analyze click-through rates, retention curves, traffic sources, and video transcripts once it is connected to your channel. I had it grab a year of data and asked which content was working, which videos flopped, and why.
The analysis took about five to ten minutes. It came back with a list of specific recommendations, and some of them stung.
Why backstory intros cause a YouTube retention drop
Backstory intros cause a retention drop because viewers who clicked for a list or an answer leave when the video detours into context first. Claude called this the single highest-leverage fix on my channel: every video with a backstory beat between the hook and the first step lost 15 to 20 retention points.
A boring products video shows the problem clearly. If I promise ten boring products and then spend a minute explaining where I gathered the data, I lose people before product number one.
My recent “Temu is dead” video was a real example. After the hook, I spent time on the US policy backstory, and Claude pinpointed the exact stretch of the retention curve where viewers left.
It could do that because it also had the video transcripts. I matched the drop-off timestamps to what I was saying, and it was the backstory every time, so now I jump straight into the first point.
The AI script training mistake that taught Claude my bad habits
The mistake was training Claude on every YouTube script I had ever written. Those scripts included my first couple of years, when I had no idea what I was doing, and most of those videos were not good.
Claude studied all of them and wrote into my style guide that I like to open with a context and backstory section. I did not notice until the retention analysis flagged the problem and I asked Claude why my scripts kept doing it.
Its answer was simple: that is how I talked in the scripts I uploaded. I told it to cut that habit and jump straight to the topic, and the new scripts no longer do it.
How to choose winning scripts to train your AI
Train your AI only on scripts from videos that performed above average for your channel. A winner is any outlier that does exceptionally well relative to your subscriber count, so the bar is different for a small channel than for a big one.
If I trained Claude to sound like me today, I would upload only those outlier scripts. Your voice still comes through, and the AI stops copying the habits of your flops.
Which YouTube video topics Claude told me to double down on and drop
Claude told me to double down on boring and dumb product listicles and to drop TikTok Shop and AI search topics. It even set a target of at least three more boring product videos in the next 90 days.
It suggested titles too, including “9 Ecommerce Niches Where Amazon Can’t Compete” and “7 Boring Products That Print Money in 2026.” I like the Amazon niches angle because it ties directly to what I teach.
| Claude’s recommendation | What the data showed | What I am doing |
|---|---|---|
| Fix the intro retention cliff | Backstory intros lost 15 to 20 retention points | Doing it now on every script |
| Publish more boring product listicles | Boring and dumb product videos are my most reliable hits | About one a month |
| Drop TikTok Shop topics | Only my “don’t sell on TikTok Shop” video did well | Dropping TikTok Shop for good |
| Drop the AI search and AEO lane | Those videos top out around 21,000 views | Ignoring it, because that is where ecommerce is going |
| Refresh the top lifetime videos | About 60% of traffic comes from subscribers | Planned for old hits over two years old |
| Put a news angle on “Forget X” videos | News-anchored alternatives videos beat pure listicles | Making new versions of past hits |
Why boring product listicles still work on YouTube
Boring product listicles still work because everyone wants to know which unglamorous products make millions, while only a small slice of viewers wants a detailed business how-to. I have never whiffed on that format, so I use it when the channel needs a hit.
Toni called those videos my Prozac. I do not want to be known only as the boring products guy, because I want authority on current ecommerce news, so roughly one listicle in every four videos is a trade-off I can live with.
Claude also noticed that my “Forget X” alternatives videos do well when they are tied to current events. My Alibaba alternatives video has almost 3 million views, so Claude suggested a new version called “Forget Alibaba, 7 Suppliers Winning Under the New Tariffs.”
Why TikTok Shop titles flop on my YouTube channel
TikTok Shop titles flop on my channel because, according to Claude, the only TikTok Shop video that ever did well was the one telling people not to sell on TikTok Shop. My latest test proved the point.
I took my reliable boring products format and put TikTok Shop in the title, calling it “9 Unsexy Products Quietly Making $10M+ on TikTok Shop.” It was one of the worst performers I have put out in a while.
Months earlier, a realistic TikTok Shop strategy video reached almost 100,000 views and then got taken down for a policy violation, which put the first strike on my channel. I still do not know why, and when I had Claude review the script against YouTube’s guidelines, it found nothing beyond one case study revenue figure.
Toni and I suspect YouTube is suppressing videos that promote TikTok Shop, since Google is competing head to head with TikTok on shopping features. We cannot prove it, and other platforms have been known to limit reach on content about their competitors.
Why I am keeping AI search videos anyway
I am keeping AI search videos because they cover where ecommerce is heading, even though Claude says that lane caps out around 21,000 views. Claude concluded my audience is Amazon sellers more than SEO operators.
Those Amazon sellers are the people who will want their own Shopify stores and AI search traffic soon. These videos also cover what I am actually doing in my business right now, so I will take 20,000 views on them.
How to refresh old YouTube videos for a click-through rate lift
To refresh old YouTube videos, retitle and re-thumbnail your top lifetime videos that are two or more years old, then watch whether click-through rate improves.
Claude recommended this because about 60% of my traffic comes from subscribers, plus weak search traffic, which means the algorithm is not pushing my old catalog to new viewers.
A click-through lift on an old hit can be worth more than a brand-new upload. My Alibaba video has almost 3 million views with a pretty bad thumbnail, so a 20% lift in CTR there would make a huge difference.
How to use AI to judge YouTube thumbnails
Claude Code does not analyze thumbnails from the API data, so I judge them with a mix of AI tools. I make three thumbnails per video and upload them to Claude, Gemini, Grok, and ChatGPT for opinions, and I also run them through the vidIQ thumbnail grader.
Thumbnails are still hard to call. On my product checklist video, the thumbnail I picked flopped, so I switched to a more boring one that vidIQ liked, and click-through rate shot up within 10 to 15 minutes.
The vidIQ score is not gospel either. The Amazon video that got a million views in a day used a thumbnail vidIQ did not like, and Toni finds its scores more accurate for Shorts than for long-form videos.
I have also been experimenting with AI touch-ups on my face, because my older thumbnails looked washed out straight from the camera. One version that emphasized every wrinkle on my face did really well, even if it is not the most flattering picture of me.
Why YouTube views and email signups are a trade-off
YouTube views and email signups are an ego versus business trade-off, because the videos with the most views are rarely the ones that convert best. My recent product checklist video came with a free scorecard app and got low views, and it produced about 50 times the email signups of a typical video relative to its views.
You can try that tool yourself with the product niche scorecard. A true banger still wins on total volume, though, because one of my videos got a million views in a single day and brought in thousands of email subscribers.
So I want a mix of both. The plan is to publish the high-view videos and point their end screens at the video that converts best to email.
Feed email signup data into Claude
Toni suggested feeding my email signup data into Claude alongside the YouTube data, and it is a great idea. My signup links are already tagged by video, so I can match each video to the subscribers it produced.
That would let Claude find the sweet spot between high views and high conversions. Tracking all the way to course sales is harder, because my conversion window can run up to a year and the course sits on a different site.
What Claude Code said about my YouTube channel overall
Claude said my channel is healthy overall, with two fragile spots. One video drove about 30% of the views over the period it analyzed, and the channel depends too much on existing subscribers instead of new viewers finding it through the algorithm.
That concentration is normal, since about 20% of your videos tend to generate 80% of your views, the same way a blog’s top 10 to 20 posts drive most of its traffic.
The high-leverage moves were fixing the intro cliff, publishing more boring product listicles, refreshing the top five lifetime videos, and dropping TikTok Shop and AI search.
I am taking every one of those except dropping AI search. Toni also pitched a video idea during this episode about celebrities with boring ecommerce businesses, which combines a listicle with a current events angle.
If you want more on how we use Claude Code across the business, we covered it in episode 646 on getting back 30 hours a week with Claude Code.
Frequently asked questions
Can Claude Code access YouTube Analytics?
Claude Code can access YouTube Analytics once you connect it to both the YouTube Data API and the YouTube Analytics API through the Google Cloud console. It can then pull click-through rates, retention curves, traffic sources, and transcripts for your videos.
How long does it take to set up the YouTube API for Claude?
Setting up the YouTube API for Claude takes about 30 minutes the first time, or a little longer if you have never used the Google Cloud console. Asking Claude Code to walk you through each step makes the process much easier.
What is a good way to stop viewers leaving at the start of a YouTube video?
The best way to keep viewers past the start of a video is to jump from the hook straight into the first point. On my channel, a backstory section between the hook and the first point cost 15 to 20 retention points per video.
Should you train AI on all of your old YouTube scripts?
You should train AI only on scripts from videos that performed above average for your channel. Training on every old script, including flops, teaches the AI to copy the habits that lost viewers.
Is it worth updating thumbnails on old YouTube videos?
Updating thumbnails on old YouTube videos is often worth it, especially on top lifetime videos that are two or more years old. A higher click-through rate on a proven video can bring in more views than a brand-new upload.
Do YouTube videos about TikTok Shop get less reach?
YouTube videos about TikTok Shop performed poorly on my channel, and one was removed with a policy strike for reasons I never learned. I suspect YouTube limits reach on TikTok Shop content because it competes with TikTok on shopping, but I cannot prove it.
