Cover: AI-generated editorial composition by TMRW. Announcements are from the YouTube Blog; rollout details credit TechCrunch.
At Made on YouTube on 23 September, YouTube announced its biggest batch of AI creator tools yet: feedback on drafts, thumbnails that change by audience, a Gemini editing assistant, and a way to test three versions of the same video. The last one could change how creators edit. Most of this arrives next year, so what creators need now is a plan, not a download.
What was announced, and when you get it
- Video A/B testing. Upload up to three cuts of a Short or video. YouTube shows each to part of the audience, reports which one holds attention, and lets you lock in the winner. It launches to select creators in 2027.
- Conversational editing. A Gemini assistant in Shorts and the YouTube Create app that trims pauses, reorders clips, adds text and syncs music to the beat when you ask. You can drop back into the timeline at any point. It is due in early 2027.
- Ask Studio feedback. Share a draft and get advice on pacing, structure and storytelling, based on your channel's history and comments. It is expanding to YouTube's mobile apps.
- Dynamic thumbnails. Give YouTube three thumbnails and it picks which one each audience segment sees. With your permission, Ask Studio can also run thumbnail tests on old videos in the background.
- AI comment moderation. Opt-in, in testing. It learns what you usually remove.
- Voice in likeness detection. YouTube's deepfake detection will start matching speaking voices alongside faces later this year.
YouTube says creators have run more than 40 million title and thumbnail tests since that feature launched in 2024. It also says hundreds of thousands of channels a day used Gemini tools in Shorts and YouTube Create in August, and that 72% of US video posters aged 14 to 44 used AI to create or edit in the past year. That last figure comes from YouTube's own survey.
Why testing cuts is the big one
Title and thumbnail tests change the packaging. Testing three cuts changes the video itself. The question stops being "which image gets the click" and becomes "which first 30 seconds keeps people watching." Until now, creators answered that with instinct and retention graphs, one upload at a time.
The catch is labor. Three cuts means three edits. That is where the conversational editor fits: if an assistant can produce a tighter-hook version while you sleep, testing alternatives becomes affordable for a one-person channel. YouTube has timed the two for the same year for a reason.
What gets lost
Dynamic thumbnails mean your video no longer has one face. Different viewers will see different images, which is good for clicks and bad for a recognizable channel identity if the three options drift apart. Keep them consistent in type, color and framing.
AI feedback built on your channel's history has a quieter bias. It tells you what worked for your past audience. That is useful when you want more of the same, and misleading when you are trying something new. TechCrunch noted that some creators worry one AI mistake could cost them credibility. With feedback tools, the bigger risk is a slow slide toward sameness.
What to do before 2027
Use the title and thumbnail testing you already have, and write down what wins. That record will be your baseline when video tests arrive. Start saving alternate openings when you edit, even if you publish only one. When three-cut testing reaches your channel, you will have hooks ready to test instead of starting cold.
Our advice: trust the tests more than the feedback. A test measures what your viewers actually did. Feedback is a model's guess about what they will do. When YouTube hands you both, let the test settle arguments.