Kyle FernandezFounder of 1 Percent Digital

How YouTube’s AI Decides Who Sees Your Video

15 Nov 2025 · 1 min

Watch on YouTube

0:52

In this segment, the presenter explains the calibration period of YouTube’s recommendation system, focusing on the technical foundations of the AI that selects which videos are shown to users. The system consists of candidate generation and ranking models that draw on three broad categories of signals. The speaker details two of these categories: viewer signals—such as who clicked, the viewer’s country, watch history, interests, and similarity to other channels the viewer follows—and content signals—covering the video’s title, description, transcript, visual embeddings, tags, comments, emotional tone, and retention curve. These signals together inform how the AI calibrates its recommendations during the initial exposure phase.

Chapters

So let's talk about what the calibration period entails. Well, on a technical aspect, YouTube's AI systems, mainly the candidate generation and ranking models, rely on 3 categories of signals. The first signal is the viewer signals. Who clicked? From what country? What watch history or interest did they have? Another one under the view signals is do they watch similar channels? For instance, if you're watching me, I would hope that you probably watch Dan Coe or Think Media because those are people that I watch on a very regular basis. Number 2 is content signals. What is your video about? What's the title, description, transcript, visual embeddings, tags, and engagement comments? Another, what kind of emotion and retention curve does it produce?