YouTube tests 3 video versions at Made on YouTube

The update builds on a suite of AI-powered creator tools YouTube unveiled in 2025, whose initial dashboard covered A/B testing for video thumbnails and a chatbot for querying how content performed. This year's additions push into more complex work: an agent that scans a creator's back catalog for videos that have taken on a new life or started trending, then proposes thumbnail and title changes to ride that wave, and can assemble brand pitches from demographic and audience data pulled from the channel.

YouTube is also widening testing itself. A dynamic thumbnail tool lets creators upload up to three images per video and assigns the best one to different audience segments to try to boost watch time. Creators can additionally test three versions of a video — a different intro and hook, for example, or a different structure — with each version fed to a small audience segment.

The backdrop is that AI use by creators has become a reputational flashpoint, after Hank Green angered fans this summer by acknowledging he had relied on AI for video research. Hanif draws a line between using AI for efficiency and letting tools do the actual job of a creator — writing the script, shooting the scenes, creating the content. How the audience reacts may hinge on whether these behind-the-scenes tools stay on the efficiency side of that line, which is likely why YouTube declined to share data on whether the tools have improved metrics for the creators using them.