YouTube demonstrated full-video A/B testing for YouTube Studio in a Creator Insider video published on September 25, 2026, showing a creator uploading alternative cuts of the same upload and receiving watch-time retention curves for each one.

In Short

YouTube showed new tools that let a creator upload two or three different edits of the same video, then show them to different viewers to see which one keeps people watching longest. This matters to anyone who pays creators or buys ads around their videos, because the video and thumbnail you see may no longer be the same ones the next person sees. None of it has a release date yet: YouTube's written materials describe video testing as coming soon, even though the demo talks about it in the present tense.

A five-minute class on Studio

Creator Insider, the channel YouTube uses to explain product changes to creators, published "Video A/B Testing & Dynamic Thumbnails - Made On YouTube" two days after the company's annual product event on September 23. The video runs just under five minutes and is staged as a lesson. "Welcome to Studio School," the presenter says at the opening; the description identifies him as Julian Shapiro Barnum of the @CelebritySubstitute channel.

The chapters follow a production workflow: a new Research tab at 0:25, draft feedback at 1:11, dynamic thumbnails at 2:05, full-video A/B testing at 2:43, back-catalogue work in Ask Studio at 3:31 and Ask Studio on mobile at 4:20. The description calls the six features "coming to YouTube Studio," a phrase that matters later.

At capture on September 27, the video had 4,672 views, 114 likes and 25 comments, and the channel listed 919,000 subscribers, up from 892,000 in July, when Creator Insider presented custom thumbnails for Shorts. The watch page labels audio tracks for some languages as automatically generated, so a briefing on YouTube's newest AI tools is itself distributed through auto dubbing.

PPC Land covered the wider Made on YouTube slate on September 24. The Creator Insider video adds the interface, the pitch and several details the written materials leave out.

The whole video goes into the test

The centrepiece is video A/B testing. "If you love the title and thumbnail AB testing from last year, hold on to your tube socks, cuz now you can do it with the whole dang video," the presenter says. The worked example is a familiar one for anyone who edits video. "Cut A starts with a dramatic reaction and cut B starts with an unhinged question." A creator uploads the cuts, and YouTube runs them before audiences and "helps you select the video cut that holds audience attention best, even giving you a breakdown of watchtime retention curves."

The demo gives no number of cuts. A September 23 post on the YouTube Official Blog by Aparna Pappu, a YouTube vice president, supplies it: creators will be able to test up to three video cuts to find the best-performing hook, a feature the post places under coming soon.

The video does not say how long a test runs, how traffic is divided between cuts, or what happens to the views, comments and likes a losing cut collects. Nor does it say whether YouTube applies the winning cut automatically. The phrase "helps you select" points to a decision left with the creator, which would depart from the way YouTube's existing tests end.

Against the title and thumbnail test

That existing system is documented. When YouTube extended A/B testing from thumbnails to titles in December 2025, creators with advanced features enabled could test up to three titles, thumbnails or combinations on long-form videos, through Studio on desktop. Variants were shown equally for up to two weeks. The system then picked the option with the highest watch time per impression and applied it. Editing a title or thumbnail mid-test cancelled the test, and when no winner emerged, the first uploaded combination became the default.

The metric changes with the object. Watch time per impression measures whether a click turns into viewing, which is a test of the promise a title and thumbnail make. A retention curve shows where inside a video viewers leave, which is a test of whether the video keeps that promise. Broadcasters have started to write the same logic into commissioning: the BBC's brief for YouTube documentaries asks that the opening of a film quickly validate the promise made by its title and thumbnail.

The timing precedent is on record. In April 2025, a YouTube Studio product management director hinted at title testing in an interview with the company's creator liaison. The feature was presented on stage at Made on YouTube on September 16, 2025 and reached creators on December 9, 84 days later.

Shorts get a test before their thumbnails do

When YouTube began rolling out custom Shorts thumbnails on July 25, 2026, access was limited to Partner Program channels, confined to desktop and excluded from A/B testing.

Thumbnails play a small part in Shorts distribution anyway. In January 2025, YouTube's Shorts product lead put the share of Shorts views originating in the Shorts feed at 99.9%, and the feed displays no thumbnail. For almost every Shorts view, the opening seconds do the work a thumbnail does on long-form. A test on the cut reaches that surface; a test on the thumbnail could not.

The sources disagree on scope, however. The narration says creators can upload cuts of "shorts or long form." The video description says "Upload different video cuts for Shorts to test hooks," which can be read as limiting the feature to Shorts. Pappu's post names no format. None of the materials reviewed reconciles the two statements.

One upload, several thumbnails

Dynamic thumbnails are the second test-related feature, and the presenter frames them against years of practice. "For years, we've had to pick just one thumbnail for every single viewer on Earth. But here's the thing. Different audiences respond to thumbnails differently." In the new arrangement, "YouTube can now serve tailored thumbnails chosen by you to different audiences." The description calls these "creator-selected thumbnails to different audience segments." According to Pappu's post, YouTube will recommend the best of three options to different segments of an audience.

An A/B test ends in one winner that every viewer then sees. Dynamic thumbnails keep several images live at once, and a thumbnail that performs worse overall can still be the one shown to a given group. How those groups are formed goes unexplained. Neither the video nor the description names a basis such as geography, language, device or watch history.

YouTube has segmented thumbnails before, on a stated basis: a September 2025 pilot of multi-language thumbnailsdisplayed localised text according to a viewer's selected language. With dynamic thumbnails, as presented, the allocation rule sits with YouTube's systems.

Feedback on a draft nobody has seen

The draft feedback chapter shifts review to before publication. "In the past, you could only get feedback on videos after publishing from your viewers, at which point there's not a whole lot you can change," the presenter says. "Now you can upload as many draft edits of a video as you like and get feedback on how to improve performance, retain viewers." On screen, the tool tells him "how I can tighten up the first bit in this video." From there, Studio "can generate thumbnails directly in the drafts."

The description summarises the feature as feedback "on pacing and retention." Pappu's post lists pacing, structure and storytelling. The forum briefing lists title, structure, script and pacing, arriving soon on iPhone and Android. None of the three says what the advice is benchmarked against: the channel's own history, comparable videos, or something else.

Generated thumbnails are not new. In July 2026, Ask Studio gained thumbnail generation globally on desktop, drawing on a video's transcript and the channel's previously posted content. The change is placement, inside unpublished drafts. In May 2026, YouTube moved its generative AI labels to more visible positions, with the requirement keyed to realistic content and meaningful alteration. Where a machine-generated thumbnail falls against that threshold is addressed neither in the July video nor in this one.

Insights, Research and the outlier hunt

The first chapter covers analytics. "Analytics in YouTube Studio are getting an upgrade to insights where creators can find new intelligent analytics to help them understand their data," the presenter says. Inside Insights sits a new destination, Research, which lets a creator "discover a collection of top performing content beyond your channel, tailored specifically to your audience and niche." The demonstration types "music theory" into a search bar and returns content the channel's viewers have already watched. "It even surfaces outliers and summarizes patterns across tons of videos to help shape your strategy."

Studio already reports some of this. Help Center pages list nine Audience tab reports, including one on what a channel's audience watches elsewhere over seven days and another on channels the audience also watches over 28 days. Research, as shown, adds topic search, outlier detection and written summaries on top of that audience data.

It also extends a line of ideation tools that began when YouTube put generative AI into the Inspiration tab at Made on YouTube 2024. When YouTube closed its trending page in July 2025, which showed identical lists to every viewer in a country, the Inspiration tab stayed. Research, as demonstrated, is tailored to one channel's audience and visible only to that channel.

The demo does not define an outlier, specify the time window the data covers, or say whether outliers are measured relative to channel size.

An agent in the back catalogue

The back-catalogue chapter makes the strongest claim about Ask Studio, YouTube's conversational assistant inside Studio. "In the past, it was just for answering questions, but now Ask Studio has evolved into a truly agentic partner," the presenter says. The word carries weight in advertising, where agentic AI describes software that receives a goal, breaks it into steps and acts without a human approving each one.

In the example, a cooking channel has weak early thumbnails, and Ask Studio "can now parse through your back catalog identifying the videos that need a little TLC and suggest new thumbnails." Then: "If you like, it can even go off and run AB testing on all of them." The description calls this "Back Catalog Optimization" and speaks of "automated tests." Pappu's post says the assistant can run in the background.

The assistant's trajectory is short and steep. It debuted at Made on YouTube 2025 as conversational analytics, began producing thumbnail artwork in July 2026 and is now pitched as running tests across a library. In April 2025, the same Studio product director described an ideal in which creators set metadata once and the platform optimises it from then on. The back-catalogue agent moves in that direction, with the creator's consent as the stated gate.

Scale brings its own effects. YouTube's German-language FAQ of September 2 states that changing a title or thumbnail may cause its systems to reassess a video, which it attributes to altered viewer interaction rather than to the edit. An agent refreshing thumbnails across dozens of older uploads would prompt that reassessment many times over. The demo does not say how concurrent tests are handled, or whether the rule that an edit cancels a running test applies to changes the agent makes itself.

Some channels may have no access at all. Under YouTube's current rules, setting an entire channel as made for kidsremoves Ask Studio along with other features. The Creator Insider video does not mention the restriction.

Ask Studio on the phone

The final chapter is the shortest: "Ask Studio is coming to mobile." The pitch is continuity, with creators able to check "performance and audience sentiment literally anytime or explore ideas on your mobile to pick up later on your desktop and vice versa." The description promises syncing between the two.

Studio's desktop and mobile apps have diverged for some time: generative thumbnails arrived in July on desktop only, while in June an Account Status page reached the Studio mobile app before desktop. Which Ask Studio functions, the back-catalogue agent included, will work on a phone is not specified.

Now on camera, soon in writing

The narration uses the present tense throughout, from "you can now get video feedback and thumbnails" to "Ask Studio can now parse through your back catalog." The written materials do not match. The video's own description calls the features "coming to YouTube Studio." Pappu's post files video cut testing under coming soon. The forum briefing describes the three-thumbnail upload as arriving soon and places Studio upgrades on a rollout running through 2027. The analytics change is described in the video itself as "getting an upgrade," not as complete. This article treats the features as presented and demonstrated, not as available.

No material reviewed gives eligibility criteria, markets or a date for any of the six features. The title and thumbnail test required advanced features; whether cut testing and dynamic thumbnails carry the same condition, or a Partner Program gate like custom Shorts thumbnails, is not stated.

What the comment section said

The 25 comments visible at capture were mostly thanks, but the critical ones drew the most likes. The top comment, with eight likes, read: "Now we need to hire a whole team for YouTube." Two others were blunt: "Video A/B testing is going to be a mess," with five likes, and "A/B testing for videos is a bad update," with four. Another compared it to selling "three versions of the same movie."

Why the marketing community is watching

For brands and agencies, the demonstration changes the unit of purchase. A brand deal buys an integration inside a creator's video, and a hook test by definition changes how that video opens. The materials do not say whether a video declared as containing paid promotion can be entered into a cut test, whether every cut must carry the integration, or whether the disclosure label follows each cut. On September 3, YouTube set out a refreshed disclosure label and automated detection of undeclared branded content in newly uploaded videos. Whether an alternative cut counts as a new upload for that purpose is unanswered.

Paid amplification raises a parallel question. Advertisers link a creator video by URL in Google Ads' Data Manager and wait for the creator to accept before running it as an ad. If a video is under test, which cut runs? The September 23 briefing described a one-click code that lets a brand view a video's performance; nothing in the Creator Insider demo describes sharing per-cut retention curves with a partner.

Counting adds a further layer. Since August 24, YouTube has counted a public view from the first frame across all formats. A hook test optimises precisely the opening seconds in which that view now registers. Views are a reporting figure, not a billing unit, but creator deals are routinely negotiated against them.

The pieces also stack. YouTube is still testing dynamic brand segments, which let a creator insert a sponsor segment into up to 20 long-form videos at once and target it by region. Add dynamic thumbnails and cut testing, and the thumbnail, the opening and the sponsor message of a single video can each change after publication, and differ between audiences. A brand checking a partner's upload on the day it goes live may be looking at one of several variants.

The demonstration closes with "classes dismissed." Dates, eligibility and the treatment of sponsored videos under test remain open.

Timeline

Summary

Who: YouTube, through its Creator Insider channel, with a demonstration presented by Julian Shapiro Barnum of @CelebritySubstitute, supplemented by a September 23 post on the YouTube Official Blog by Aparna Pappu, a YouTube vice president. The features affect creators, and indirectly the brands, agencies and media buyers that sponsor, boost and evaluate creator videos.

What: A demonstration of six YouTube Studio features: an Insights upgrade with a Research destination for outlier videos; feedback on unpublished draft edits with generated thumbnails; dynamic thumbnails that serve up to three creator-selected images to different audience segments; full-video A/B testing of up to three cuts with watch-time retention curves; an Ask Studio agent that refreshes and tests back-catalogue thumbnails; and Ask Studio on mobile.

When: The video was published on September 25, 2026, two days after Made on YouTube on September 23. The narration describes the features as available now, while the description and YouTube's written materials describe them as coming or coming soon, with no dates given.

Where: YouTube Studio on desktop and, for Ask Studio, mobile. No markets or eligibility criteria are specified. The narration includes Shorts and long-form in video testing; the description refers to Shorts.

Why: YouTube is extending testing from the packaging of a video to the video itself, and moving Ask Studio from answering questions to acting on a channel's library. For marketers, a single sponsored upload may now exist as several cuts and several thumbnails, and the materials do not yet say how disclosure, paid amplification or reporting apply to them.