AI tools and automation

@angus.sewell

Use AI to analyse retention data from paired video tests (trial reels), find which phrases make viewers leave, rewrite the better version, and store the patterns as lessons you get coached on in Notebook LM.

129k plays · 27 Oct 2026

What they said

Premise: Hormozi has spent about $4M on content in 4 years (stated as "four million in the past four years", "mostly spent on the testing side"). The on-screen Google search and articles show "$4M content investment, across roughly 40 months" and "Test. Measure. Learn. Repeat."
Therefore "you can vibe code his whole system".
Step 1: post on TikTok "of all places" and capture your analytics through the workflow on screen. The workflow nodes shown: Post tests, Saved browser login, Capture Studio analytics, Latest metrics, Snapshot / story.
Step 2: with two tests and the retention data saved for both, have the AI analyse both videos and map out which phrases make the most people leave (prompt shown on screen: use video IDs, compare TikTok videos 8956 and 8957, read both full transcripts and on-screen text, etc.).
Step 3: the AI tells you how to make a better version based on both videos' analytics (on-screen example: retention at 10 s, saves per 1,000 views, "Optional edit test for 8956: keep 0-24s, change 24-27s, cut 29-31s", and a note that the cut is a hypothesis, the archive tests did not find a reliable worst passage, only a re-cut test against the original would show whether it works).
Step 4: post the two versions as trial reels; the winner goes to the main page.
Step 5 (level up): have the AI analyse every bit of data you have ever collected, find patterns in your content analytics, and save them as definitive lessons in a table (columns like Saved lesson, Evidence, Stronger / More support / Counter, Next experiment).
Step 6: have it coach you via something like Notebook LM on those lessons so you improve over time. The Notebook LM screen shown: "Why the designer, McDonald's and Kalshi videos did what they did".
Frame: "You make the content and the AI just does the analysis." Not AI-generated content.
Frame data seen on screen: charts "20 prior comparable videos", "retention curve", "Typical total retention", analysis per 0-3 s, 4-5 s etc. with "usual loss" figures, with early-data caveats ("Early data, descriptive comparison", "Better / Usual / Worse").

Try this

Creators: Post two versions of the same video as paired tests, then compare retention at 3 seconds, 10 seconds and the end. Post the winner to your main page.
Use AI to read the data: Give it both transcripts and the retention curves, and ask which lines make people leave and what to cut or change. Treat the answer as a hypothesis and re-test it.
Keep a lessons table: Record each lesson, the evidence, whether it is supported or contradicted, and the next experiment. Review it monthly.
Coaching: Load the table into a notebook tool and ask it questions, such as why certain posts worked.
Be careful: Small samples are noisy, and automated logins or scraping may break platform terms. Your content stays yours; the AI only analyses.

Watch the original

Watch the full reel on Instagram by @angus.sewell. Follow them for more.

Video, voice and ideas belong to @angus.sewell. Shown here as a short preview with our own summary, for commentary and education. Want it removed? Tell us.

00:00