I'm back in NYC today to do a keynote at the Cranium AI Summit & Headbangers Ball. It’s going to be an awesome afternoon.
In the news: On Wednesday, Google announced a new Gemini AI-powered ad product called Peak Points at its YouTube Brandcast event in New York. The AI identifies moments during a video when viewers are most engaged, then drops in an ad. In theory, ads will perform better because the messages appear during emotionally resonant or attention-rich segments. Of course, it could completely backfire. People may be so engaged in the programming that they ignore the ads. Test and learn.
Contextual targeting has been around for decades, and AI has been optimizing ad placement in programmatic media for years. What’s new here is the granularity. YouTube isn’t just using metadata, watch history, or audience segments; it’s using Gemini to analyze individual videos and spot the exact moments when viewers are leaning in. Not where you’re likely to be interested, but where you are interested.
If Peak Points works as promised, YouTube won’t just know what you’re watching — it will know how you’re watching it. That has broader implications than ad timing. Real-time attention data is the raw material for algorithmic recommendations, content development, talent discovery, and platform control.
For creators, this changes the rules of monetization. Midrolls will no longer be inserted at default intervals. Gemini will decide if your content is engaging enough to justify an ad. That subtly redefines the creative brief. You are no longer asked to retain viewers. You are now asked to spike their attention at the right timestamps.
For marketers, if you're running video, this is worth testing. You are no longer buying against content. You are buying against engagement. That is a different product, and it forces a different conversation about brand safety, storytelling, and outcomes.
If you thought attention was the currency of the internet, welcome to the new exchange: engagement spikes priced in CPM.
As always, your thoughts and comments are both welcome and encouraged.
