This is a great resource by Ayo, here is the video broken down so you can take the learnings and apply them accordingly.
Pre-Scoring Content: Before a video even plays, TikTok’s servers assign it "recommend scores". These scores calculate the exact mathematical probability that you will skip, finish, like, comment, or follow the creator.
On-Device Monitoring: The app runs over 25 AI models directly on your phone to continuously monitor your real-time state and environment.
Context Gathering: Every time the app requests a new batch of videos, it sends over 58 variables. This includes your battery percentage, whether your phone is charging, screen brightness, mute status, and gyroscope data to detect if you are walking, lying down, or in a car.
Micro-Behavior Tracking: The app collects keystrokes, including messages you type but ultimately delete or decide not to send.
Attention Recovery: A predictive model constantly monitors your behaviour to guess if you are about to skip the remaining loaded videos. If it thinks you will, it requests an "emergency batch" of videos to immediately recapture your attention.
Lost Focus Protocol: If you keep skipping videos despite the emergency batch, the system flags you as having "lost focus" and serves a Live video into your feed as a last-resort hook.
Drowsiness Detection: A specialised model detects if you have fallen asleep by monitoring how long the screen goes untouched while you are lying down.
Predictive Pre-Warming: The algorithm anticipates exactly when you are likely to open the app next and launches it in the background so content is ready instantly.
The TikTok Algorithm Cheatsheet
Recommend Scores - Evaluates a video's likelihood of engagement (skip, finish, like, comment) before serving it to your feed.
Physical State Adaptation - Uses gyroscope data to serve a different feed when you are lying in bed versus walking or on a lunch break.
Device Variables - Tracks 58+ real-time phone states (battery, brightness, charging, mute) to contextualise your scrolling session.
Unsent Text Tracking - Captures drafted text and discarded messages to further build your profile.
Emergency Batches - Predicts an incoming scrolling rut and rapidly fetches new videos to stop you from closing the app.
Live Video Push - Pushes a Live stream as a pattern interrupter when you are aggressively skipping videos.
Drowsy Detection - Identifies sleep states based on physical orientation and lack of screen interaction.
App Pre-Warming - Learns your daily habits to load the app in the background right before you typically click the icon.
