The Kale Algorithm
Filter your video feed to only content you'll be glad you watched
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 82%
George Mack audited roughly his last hundred watched YouTube videos and found ~80% were regrettable. Instead of building a complex recommender, he looked for the single variable dividing videos he enjoyed in hindsight from ones he didn't, and found it was duration: videos over ~30 minutes were almost always worth it, videos under 30 minutes almost always weren't (the monkey brain justifies a 15-minute hit-piece but not a 2.5-hour deep dive). He installed Tampermonkey and used Claude/ChatGPT to write a script that permanently removes any video under 30 minutes from his feed. His regret rate inverted: from ~80% regretted to ~80% he's glad he watched, while keeping the randomness and optionality of the algorithm.
Origin
Custom-built by George Mack in 2024 after years of debating with Chris Williamson whether platforms would ever let users control their own algorithm; named the 'Kale Algorithm' as the opposite of the addictive 'cocaine algorithm' of short-form.
Core principles
- 01You can't out-discipline an algorithm engineered to hijack you
- 02Prevent the temptation appearing, don't fight it once it's on screen
- 03One clean filter beats a complex recommender you can't control
- 04Longer-form content self-selects for higher signal
How to run it
- 1
Stare into the abyss
Open YouTube.com, go to History, and scroll your recent watches honestly. Notice how much of it you regret.
Pro tip Use ~100 videos so the pattern is statistically obvious, not cherry-picked.
- 2
Split enjoyed vs regretted
Tag each video as one you were glad you watched in hindsight or one you weren't.
Watch out Judge by hindsight value, not how fun it felt in the moment.
- 3
Find the one dividing variable
Look for the simplest single attribute that separates the two piles. For George it was video length crossing ~30 minutes.
Pro tip Resist building a complex multi-factor rule; the power is in one clean filter.
- 4
Install a userscript engine
Add the Tampermonkey Chrome extension so you can run custom scripts on the YouTube page.
- 5
Generate the filter
Ask an LLM (Claude or ChatGPT) to write a script that hides any video under your threshold, and paste it into Tampermonkey.
Pro tip Share the script with friends; it's reusable once written.
- 6
Leave it running
The filter stays on permanently, so short-form junk never renders and you only see long-form worth your time.
Watch out You still keep the algorithm's discovery, so don't also disable recommendations unless you want a blank feed.
In the wild
Before the filter, George's feed pushed Logan Paul NFT exposes and crypto-drama clips he'd click reflexively. After installing the under-30-minute filter, his feed became stand-up comedians, long documentaries, and long-form comedy podcasts he genuinely enjoyed and remembered.
→ Regret rate on YouTube time flipped from ~80% regretted to ~80% glad he watched.
Common mistakes
Building a complex recommender
The instinct is to engineer an elaborate scoring system; the leverage is finding the single variable and filtering on it.
Relying on willpower at the thumbnail
Assuming you'll resist a tempting thumbnail fails; the fix is preventing the item appearing at all, not out-disciplining it.
Is it for you?
Best for
Desktop YouTube heavy-users who use it as an alternative to TV and want the upside of discovery without the junk.
Not ideal for
People whose weakness is short-form scrolling on mobile, where a duration filter doesn't apply.
From the transcript
“I built a script that removes any videos under 30 minutes, and it's now the full KL algorithm. And I've gone from about 80% of…”
“any video that seemed to be over 30 minutes long for the most part, I enjoyed in hindsight, and any video under 30 minutes long…”
“it's not necessarily about discipline, it's about preventing that coming on in the first place”
From the episode
Christmas Special - Life Hacks, Biggest Lessons & Best Resolutions - #881