YouTube Shorts Algorithm 2026: A Guide for Creators

YouTube said Shorts passed 200 billion daily views in 2025, and the surface now sits on the YouTube home page rather than off in a tab. That scale is real, though part of the jump came from a counting change rather than more watching.

YouTube's Shorts feed does not work like TikTok's For You page, and it does not pay like it either. Both halves matter if you are producing AI-generated Shorts at volume.

#YouTube names four performance numbers in its own ranking docs

There is less mystery here than the "algorithm hack" genre implies. YouTube's help documentation says Shorts in the Shorts feed are ranked on performance and viewer personalization, and it names the performance inputs directly: the percentage of viewers who chose to view, average view duration, average percentage viewed, likes, and post-watch survey results.

The same page carries a sentence worth taping to your monitor. "Our system has no opinion about what type of Shorts you make. It also doesn't favor any particular format." Tutorials are not structurally favored over comedy. A tutorial tends to win because tutorials hold attention, and watch time is what the ranker actually reads.

Two of those inputs are duration ratios rather than totals, which is why a 20-second Short watched to the end can outrank a 90-second one that loses half its audience at the 30-second mark.

#Replays inflate your view count but not your paycheck

On March 31, 2025 YouTube changed what a Shorts view means. A view is now counted each time a Short starts to play or replay, with no minimum watch time. The old, stricter metric survives in YouTube Analytics under a new name, engaged views, and that is the one attached to money.

The split matters more than it looks. Partner Program eligibility and Shorts ad revenue sharing are both calculated on engaged views, so a Short whose public counter reads 400,000 can carry a fraction of that in the number YouTube pays against. YouTube extended the same principle in April 2026, when views from image posts in the Shorts feed started counting toward the total view count while staying excluded from eligibility and revenue share.

Practical version: use the public number for social proof and the engaged-views column for every decision about what to make next.

#Engagement signals changed shape in June 2026

The dislike button is going away. In a June 25, 2026 post, YouTube said it is retiring dislikes on the Shorts experience in favor of "Not Interested" and "Don't recommend this channel," which give the recommendation system a more specific instruction than a thumbs-down ever did. The thumbs-up is being replaced by a heart at the same time, alongside a clear-screen mode, double-speed playback, and an adjustable Shorts timer.

Advice built around "avoid dislikes" is therefore obsolete. What replaced it is harsher. A viewer who taps "Don't recommend this channel" removes your whole catalog from their feed, not one video, so a bait hook that oversells the payoff now costs more than a downvote ever did.

Comments and shares still register. A subscribe click taken from inside the Shorts player remains the highest-value action a viewer can take, so ask for one specific thing at the end instead of a generic follow.

#Shorts run to 3 minutes, and the rules tighten after 60 seconds

Since October 15, 2024 you can upload Shorts up to three minutes long through the YouTube app and YouTube Studio, provided the aspect ratio is square or taller. Anything wider, 16:9 included, gets treated as a regular video, which is the deliberate escape hatch when you do not want a clip classified as a Short. Standard vertical video at 1080x1920 stays the safe default.

Length brings two constraints that catch people out. Most songs in the Shorts Audio Library are limited to 90 seconds inside a three-minute Short, and some tracks are capped at 60 or 30 seconds. Separately, a Short longer than one minute that picks up an active copyright claim is blocked globally, not monetized, and not recommended, where a claim on a shorter Short is handled less severely.

The three-minute ceiling is real. Crossing 60 seconds is what changes your risk profile on music.

#Shorts and TikTok reward different behavior

The TikTok duration figures come from TikTok's own Content Posting API documentation, which also caps uploads at 4GB and accepts MP4, WebM, and MOV.

#Getting into the Partner Program through Shorts takes 10 million views

The Shorts route into the YouTube Partner Program is 1,000 subscribers plus 10 million valid public Shorts views in the last 90 days. Watch hours accumulated in the Shorts feed do not count toward the 4,000 public watch hours on the long-form route, so a Shorts-only channel cannot back into monetization the traditional way.

Once you are in, payment runs through a pool. Revenue from ads between Shorts is collected monthly into a Creator Pool, allocated to each monetizing creator by their share of engaged views in a given country, and you keep 45% of your allocation. Using music does not reduce that rate, which removed the old incentive to avoid licensed tracks. YouTube pays the same 45% on the Premium revenue allocated to Shorts.

Ten million views in a rolling 90-day window works out to roughly 111,000 a day. That is a publishing-cadence problem before it is a creative one.

#Templated AI output is a monetization risk, not a ranking one

On July 15, 2025 YouTube renamed its "repetitious content" policy to "inauthentic content" and spelled out what it covers: mass-produced or repetitive content, anything that looks built from a template with little variation between videos, and AI-generated content made with generic templates that adds no original insight or perspective. Image slideshows and scrolling text with no narrative sit on the same list.

Nothing in that policy touches ranking. The Shorts feed will still recommend an AI-made Short that holds attention. The policy governs whether the channel can be monetized at all, which is a slower and more expensive failure to discover.

The workable line is that AI handles production while you supply the angle. A generated script built from your own data, your customers' questions, or a position you actually hold reads as authored. Fifty variations of one listicle skeleton do not, and that is precisely the pattern the policy names.

#Structure a Short so it finishes

Completion is what two of the named inputs are derived from, so structure toward the end of the video rather than the middle.

1. Hook, 0 to 3 seconds. A pattern interrupt that earns the next three seconds.
2. Value, 3 to 40 seconds. Dense enough that skipping ahead loses something.
3. Payoff, 40 to 50 seconds. Deliver the thing the hook promised.
4. Ask, final 5 seconds. One specific action.

Most content types land best between 30 and 45 seconds. Shorts under 15 seconds give the ranker very little duration signal to work with, and past 60 seconds you need genuinely sustained material to keep the average-percentage-viewed number healthy, on top of the music and copyright constraints that kick in at that mark.

#Subtitles work as a topic signal and as sound-off insurance

YouTube classifies Shorts by topic using spoken words, on-screen text, visual analysis, and your title and description. Burned-in captions feed two of those four inputs at once, which is why keyword-bearing narration beats keyword-stuffed tags.

The sound-off argument is separate, and it gets overstated on YouTube specifically. The best-sourced figure belongs to Meta, whose own test of Facebook video ads found captions lifted view time by 12%. Treat it as directional for Shorts rather than a YouTube number, since YouTube has never published an equivalent.

Auto-generated subtitles with word-level timing cover both jobs in one pass. Keep them inside the safe area. Orange reserves 12% of the frame width on the right for platform action buttons and 18% at the bottom, because that is where the Shorts player draws its own interface.

#Shorts resurface months later, which changes what a bad week means

Unlike a TikTok that lives or dies inside 48 hours, a Short can be re-recommended long after upload. A video that pulled 500 views in week one can find 50,000 viewers a quarter later if the system detects sustained interest in the topic.

That asymmetry rewards cadence over perfectionism. Three to five Shorts a week keeps enough surface area in the pool for the system to find the one that works, and that volume is realistic when production is automated. Past two or three uploads a day you mostly compete against yourself.

It also makes evergreen framing worth more than trend-chasing here. A Short that explains something still makes sense in November. A Short built on a July audio meme does not, and that one gets a single shot at distribution. Which is why a faceless YouTube channel built on explainers compounds while a trend-reaction channel resets every week.