How does the YouTube Shorts algorithm actually work?

Short answer: YouTube shows your Short to a small group of strangers, watches what they do in the first few seconds, and expands or kills distribution based on that. The whole game is the swipe decision and what happens right after it.

Most algorithm advice treats Shorts like small long-form videos. They aren't. Shorts run on a separate ranking lane with different signals, different math, and one brutal difference: on long-form, the viewer chose your video from a thumbnail; on Shorts, your video interrupted their scrolling. Everything the algorithm measures flows from that interruption — did they stay, or did they swipe?

Once you see it that way, the entire system makes sense.

A separate lane from long-form

The first thing to internalize: the Shorts feed is its own recommendation system, largely independent of your long-form performance. A Short does not drag your channel's long-form videos up or down, and — this is the part that surprises people — your subscriber count barely counts toward how far a single Short travels.

Each Short is judged on its own merits, by strangers, in the feed. That is bad news if you were counting on your 200,000 subscribers to carry a weak Short. It is very good news if you have 12 subscribers and a strong one. The Shorts feed is the closest thing YouTube has to a meritocracy: no reputation, no head start, just the video against the swipe.

This also means Shorts and long-form success barely transfer. Channels that dominate one often flounder in the other, because they are different sports that happen to share a stadium.

The first gate: viewed vs swiped away

When your Short appears in someone's feed, the algorithm's first question is the simplest: did they stop, or did they swipe past?

YouTube calls this "viewed vs. swiped away," and you can track it in YouTube Studio next to the "shown in feed" metric. It is the top-of-funnel gate — the first test every Short must pass — and it replaces the thumbnail click-through rate that governs long-form. There is no thumbnail to click. There is only the opening frame and the first second of motion, competing against a thumb already in motion.

The benchmarks floating around creator communities are unforgiving: if well under two-thirds of viewers stop to watch, the Short tends to flop; in the 70–90% range, it tends to break out. Treat exact numbers as directional — YouTube doesn't publish official thresholds and they shift — but the shape is right. The swipe decision is binary, instant, and it filters most Shorts out before anything else gets measured.

This is why the first frame matters more than anything you were taught about thumbnails. The hook isn't a title. It's the first thing moving on screen.

The second gate: how much they actually watch

Pass the swipe test and the algorithm asks its second question: of the people who stayed, how much did they watch?

Two metrics here — average view duration and average percentage viewed — and the percentage matters more than the absolute seconds. A 20-second Short watched to the end beats a 3-minute Short abandoned at second eight. This is why shorter Shorts often outperform longer ones: it is easier to hold someone for 18 seconds than for 90, and the algorithm grades on completion, not endurance.

Rough benchmarks from creator testing: sub-30-second Shorts want around two-thirds average completion to widen distribution; 30-to-60-second Shorts want about half; longer Shorts need roughly 40–45%. Again, directional, not gospel. But the principle is exact: retention rate beats watch time. A short video fully watched is worth more than a long video half-watched, every time.

Note the quiet implication: making your Short longer to "say more" usually makes it perform worse. Say it in 25 seconds or don't say it in Shorts.

Loops and shares: the accelerants

Two signals can supercharge a Short beyond what retention alone earns: loops and shares.

Because Shorts auto-play on a loop, a viewer who watches again — even partially — registers as a rewatch, and percentage viewed can climb past 100%. A Short looping at 120% or 150% is screaming that people can't stop watching it, and the algorithm hears that scream clearly. This is why the best Shorts are engineered to loop: the ending flows back into the beginning so seamlessly that the second watch starts before the viewer decides to leave. A perfect loop is a retention cheat code.

Shares carry heavy weight for a different reason: they predict reach outside the algorithm. When someone sends your Short to a friend, they're doing distribution work YouTube can't do itself, and the system rewards the video that earned it. Among engagement signals, shares sit at the top of the Shorts hierarchy.

Likes and comments, honestly? Weak correlation with Shorts distribution. They're nice community signals, but chasing them as a growth lever is chasing the wrong number. A Short with modest likes and a high loop rate will outrun a Short with great likes and a high swipe-away rate. Optimize for the behavior that predicts more watching, not the behavior that feels like applause.

Satisfaction: the survey you never see

Here is the signal most creators don't know exists: YouTube surveys a sample of viewers about what they watched and feeds the answers back into ranking. The platform's framing is "satisfaction, not raw watch time" — and on Shorts it gets enforced harder than on long-form, because a swipe-away is such a clean negative signal.

This is the algorithm's defense against gaming. A Short can rack up watch time through bait-and-switch tactics — a shocking first frame, a payoff that never comes — and still lose its reach, because the people who watched it told YouTube they didn't enjoy it. Designing for the metric while betraying the viewer is a strategy with a short half-life. The survey always catches up.

The practical takeaway is almost moral: make the Short the opening promises. Retention earned honestly and retention earned by trickery look identical in the first hour and very different in the second week.

What doesn't matter (as much as you think)

Almost as useful as knowing the signals is knowing what the algorithm mostly ignores on Shorts:

• Subscriber count. Your subs don't carry a Short. Strangers decide.
• Thumbnails. There is no click, so there is no click-through rate. The first frame is the thumbnail.
• Posting time and frequency. The feed is not chronological; consistency helps you learn faster, but no upload slot is magic.
• Hashtags and descriptions. Marginal at best for Shorts distribution. Write them, don't obsess over them.
• Likes as a lever. As covered above: community signal, not a distribution lever.

Every hour spent optimizing these is an hour not spent on the two things that actually move: the first second and the reason to stay. Most Shorts advice is long-form advice wearing a costume. Strip the costume off.

The search afterlife

One structural advantage Shorts has over TikTok and Reels is worth building for deliberately: YouTube is a search engine, and Shorts get indexed by it.

On TikTok and Reels, distribution is almost entirely the initial algorithmic window — once the spike passes, the video is effectively dead. A YouTube Short, by contrast, can keep earning views from search for months after the feed forgets it. A Short that answers a real question — how to fix an error message, how long to cook something — becomes a durable asset, not just a lottery ticket.

This doesn't change what the algorithm measures. But it changes what you should make: mix some evergreen, searchable Shorts in with the feed-chasing ones. The viral swings get the attention; the searchable ones pay rent. Over a year, a library of twenty searchable Shorts can quietly outperform one viral hit that everyone forgot by Friday.

What to actually do with all this

If the whole system had to compress into a checklist, it would be this:

• Win the first 1–3 seconds. Movement, a face, a question, a payoff promised — something that beats a thumb already scrolling. YouTube measures first-frame engagement separately, so treat the opening as its own product.
• Build for completion, then for loops. Shorter is usually stronger. End where you began so the loop feels continuous.
• Make something shareable, not just watchable. The share is the strongest vote a viewer can cast. Ask what would make someone send this to exactly one friend.
• Keep the promise of the opening. Satisfaction surveys punish bait. The algorithm has a longer memory than your analytics dashboard suggests.
• Judge each Short alone. A flop doesn't poison the channel; a hit doesn't guarantee the next one. The lane resets every upload, which is terrifying and liberating in equal measure.

And one structural note worth knowing: YouTube keeps Shorts discoverable through search long after the feed spike fades, unlike TikTok and Reels where distribution is almost entirely the initial window. A Short that answers a real question can keep earning views for months. Build a few of those — the library compounds while the feed forgets.

So — how does the Shorts algorithm actually work? It tests your video on strangers, measures whether they stayed, how long, whether they watched again, and whether they were glad they did. Everything else is commentary. Make the first second undeniable and the last second loop back to it, and you have understood the entire machine.