3 days free on every plan · Try any plan before you pay.

Start free trialStart

BlogShort-video data

How does the YouTube Shorts algorithm work? YouTube's words, our data

What YouTube itself says about how Shorts are recommended, quoted and linked, and what 15,007 of our Shorts show about seed audiences and the first day.

By Abondy ResearchPublished 10 min read

In short: YouTube says its recommendations try to show each person videos they'll value, using signals such as what people watch, likes, dislikes, survey answers and "the reputation and the quality of a channel." For Shorts, its product lead has said the feed is swiped rather than chosen, and that the system can find "a seed audience" for a new Short; depending on how that goes, the Short "may get a lot more traffic or it may kind of taper off." Our data fits that picture. Of 2,341 Shorts posted October 24 to November 12, 2026, not one of the 1,702 still under 100 views at 72 hours had passed 1,000 by our last reading. Shorts that did pass 1,000 were usually there within a day.

YouTube hasn't published a list of Shorts ranking factors. Everything below is either YouTube's own words, linked and dated, or our own data, labelled as such. Where YouTube hasn't said something, we say so rather than guess.

Does YouTube Shorts use an algorithm?

Yes. YouTube's How YouTube Works page on recommendations, read November 22, 2026, says "the core job of recommendations is to meet user needs by helping people find the videos that they want to watch and that will give them value."

The most detailed account of Shorts in particular comes from YouTube's Creator Insider channel. In YouTube Shorts Algorithm — Explained! (August 23, 2023, transcript read November 22, 2026), Todd Sherman, who described himself as "the Product Lead for YouTube Shorts," put it this way: "The audience is the algorithm." The video is from 2023, so some details, like the 60-second focus he mentioned, have changed since.

How is the Shorts algorithm different from long-form YouTube?

The difference YouTube describes is how people arrive at a video. Sherman said that on long-form, "people are choosing which video by tapping on it on their phone or clicking on it on the web." On Shorts, "people are swiping through a feed and they're discovering things as they go," and "that leads us to need to measure in different ways."

He named one example: "how diverse the feed is is a little bit different when you have hundreds of videos versus maybe 10 or 20 videos." The goal is the same in both, he said: "trying to get videos to people that they value."

What signals does YouTube say it uses?

YouTube has described the signals for its recommendations as a whole, not for Shorts on their own. In a 2021 post on the YouTube Official Blog, On YouTube's recommendation system, YouTube's VP of Engineering Cristos Goodrow wrote that the system is "learning every day from over 80 billion pieces of information we call signals." The ones he named were "clicks, watchtime, survey responses, sharing, likes, and dislikes."

Two details in that post matter for Shorts:

  • Satisfaction, not just time. YouTube measures "valued watchtime" through surveys that ask people to rate a video from one to five stars, and trains a model "to predict potential survey responses for everyone."
  • No social graph. "Unlike other platforms, we don't connect viewers to content through their social network."

The current How YouTube Works page adds that YouTube looks at "viewing behaviour, likes, dislikes, subscriptions and feedback, including from satisfaction surveys," and that "we also consider the reputation and the quality of a channel to determine how, when and to whom the content can be surfaced."

For Shorts, YouTube Analytics shows one metric built for the swipe. Its content performance page defines "Stayed to watch" as "The percentage of times viewers stayed to watch past the initial seconds of a Short." In 2023 Sherman described the same metric, then called viewed versus swiped away, as a way to see "which ones are grabbing the audience versus people maybe just deciding to swipe past them."

What does "seed audience" mean on Shorts?

It's Sherman's explanation for why some Shorts take off and others stall:

"There are parts of the algorithm that try and find people find creators and audience and sometimes those algorithms will go and effectively find a seed audience find a set of people that may enjoy your video and depending on how that goes it may get a lot more traffic or it may kind of taper off."

He added that "especially creators starting out ... a lot of those early views they're very exploratory as we're trying to proactively help you find the audience that enjoys your content."

Our own readings show what that looks like from the outside. We keep a Short's stored view readings for 30 days, so for Shorts posted October 24 to November 12, 2026, we still have all of them. We store a reading when the count goes up, at most one an hour in a Short's first two days and one every six hours after, so "views 24 hours after posting" means the last stored reading at or before that hour. Here is where 2,341 of them stood 24 hours after posting, against where they ended:

Views 24 hours after postingShortsEnded under 100Ended at 100 to 999Ended past 1,000
Under 101,44399.3%0.6%0.1%
10 to 9930785.0%11.1%3.9%
100 to 9992810%82.6%17.4%
1,000 or more3100%0%100%

At 72 hours the split was sharper still. Of the 1,702 Shorts under 100 views at that point, none had passed 1,000 by our last reading, and only 0.5% even reached 100. We read 90% of them again more than a week after posting. Of the 267 between 100 and 999, one Short later passed 1,000.

For the 373 Shorts in this group that ended past 1,000 views:

  • The first reading that showed 1,000 or more came a median of 8.8 hours after posting.
  • 83.1% showed 1,000 or more within 24 hours, and 90% within about 33 hours.
  • At 24 hours the median one already had 91.7% of its final views, and at 72 hours 99.6%.

So in our data, a Short's fate was mostly set in its first day. One limit: we store at most one reading an hour in those first days, so "8.8 hours" is when a stored reading first showed 1,000, not the exact moment it got there. The full timing curve for every app is in when do views stop coming in.

Is it the video or the channel?

YouTube says channel reputation and quality is one of the things it weighs. In our data, the channel told us more than the video did. Taking the spread in (log) views across our read Shorts:

Grouped byGroupsShortsShare of the spread that sat between groups
Channel (10 or more read Shorts)3767,72540.5%
Upload, the same source video (20 or more read Shorts)2089,90113.3%

Put plainly, the same video posted on many channels landed in very different places, while one channel's Shorts tended to land near each other. Among those channels, the share of Shorts past 1,000 views ran from 3.4% at the 10th percentile to 71.4% at the 90th. Among uploads it ran from 8.7% to 61.4%. That's a pattern on small accounts, not proof of how YouTube scores a channel, and do followers matter for views covers the subscriber side of it.

YouTube's October 1, 2026 Shorts update works at the channel level too: it said channels "primarily aggregating or re-uploading other creators' videos without significant changes will likely see less distribution within the Shorts feed." What that means for brands is in YouTube's Shorts reused content policy.

Do likes and comments push a Short further?

YouTube's 2021 post says "people are more likely to be satisfied by videos that they share or like," and that the system "uses this information to try to predict the likelihood that you will share or like further videos." It doesn't say likes on one video push that video to more people.

In our Shorts with 100 or more views, the biggest ones had the lowest like rates:

ViewsShortsLikes per 1,000 viewsComments per 1,000 views
100 to 9991,2687.310.95
1,000 to 9,9995,1265.650.52
10,000 or more1224.770.19

One reading: a Short that reaches more people reaches more of the casual swipers, who like less. Either way, a Short in our data didn't need a high like rate to travel.

Do posting time, hashtags and thumbnails matter?

Sherman's answers in the 2023 video:

  • Time of day: "No, it's not something that we think about optimizing for." He said freshness can matter for news, which is "more about quickly responding."
  • Hashtags: "If the question is are they required: no, of course not." He said they "could be" meaningful, "but it's hard to generalize across every vertical for every creator."
  • Thumbnails: "most thumbnails are never seen because people are in the feed and they're swiping."

Our own posting-time data did show a pattern, with Shorts posted 8 p.m. to midnight New York time passing 1,000 views more often. That's an association in our sample, covered in the best time to post YouTube Shorts, not a sign the feed checks the clock.

Is the YouTube Shorts algorithm like TikTok's?

YouTube hasn't published a comparison. What it has said is that it doesn't recommend through a social network, and that Shorts are measured differently from long-form. Our data shows the two apps judged the same file differently. For 6,313 versions posted on both Shorts and TikTok and read on both, the rank correlation of their views was 0.074, where 0 means no link. Whether a version beat the middle of the pack on one app matched the other 51.8% of the time, about a coin flip. More on how the apps differ is in TikTok vs YouTube Shorts vs Instagram Reels.

What is the 7-second rule on YouTube?

We didn't find it on any YouTube page. It's a rule of thumb from outside YouTube. The nearest official thing is the Stayed to watch metric, which counts viewers who got "past the initial seconds of a Short." YouTube doesn't say how many seconds that is.

Should you delete and repost a Short that didn't take off?

Sherman's answer was direct: "No, don't. I would not advise that." He linked it to spam detection and said: "Take everything you learn from that and make a better version of it." Given our first-day data, a Short under 100 views after three days is very unlikely to recover, so the next version is the better use of time. How often to post those versions on one channel is in how many YouTube Shorts to post a day.

What this means for a brand

  • Judge a Short after a day, and a campaign after many Shorts. In our data the first day decided almost every Short, and most Shorts stayed small.
  • Expect channels to differ a lot. The same video did very differently from one channel to the next.
  • Make the opening count. YouTube's own Shorts metric is about whether viewers stay past the first seconds.
  • Don't chase likes as the goal. Our biggest Shorts had the lowest like rates.

Where Abondy fits

Abondy spreads one video over many channels, so each version gets its own first day. You upload a video, Flagship 1.2 makes versions of it, and real people post them on their own YouTube Shorts, TikTok, Instagram Reels and Facebook accounts, and your dashboard shows each post's views as they come in. A post costs 16¢ to 20¢ on a plan, plans start at $20 a month (pricing), and a post under 100 views 24 hours after its link is handed in isn't charged. The monthly views report shows what did best.

How we counted

  • Sample: YouTube Shorts posted August 29 to November 12, 2026 (New York time), pulled November 22, 2026. 15,286 counted; 279 (1.8%) never returned a view count and are left out, leaving 15,007. One brand's campaign makes up 54.0% of them and Abondy's own campaigns 23.2%; most are gaming clips, largely Roblox, on small accounts.
  • First-day tables: the 2,341 read Shorts posted October 24 to November 12, whose stored readings are all still kept (we keep YouTube readings for 30 days). This group differs from the rest: its largest campaign is 39.2% of it, Abondy's own campaigns 35.7%, gaming 66.1%, and only 15.9% passed 1,000 views.
  • Spread: variance of log(1 + views) inside groups against the total, for channels with 10 or more read Shorts and uploads with 20 or more.
  • Views: each Short's public count when we last read it. Flagged, rejected and deleted posts, posts whose views were confirmed as bought, and test accounts are left out.
  • Associations, not proof: nothing here shows what YouTube's system does inside. It shows what happened to our Shorts.

Questions people ask

How does the YouTube Shorts algorithm work?

YouTube says it tries to show each viewer Shorts they'll value, judged by signals like watching, likes, dislikes and surveys. Its Shorts lead described the system sometimes finding a seed audience for a Short, then widening or tapering its reach depending on how that audience responds.

Do Shorts get tested on a small audience first?

YouTube's Shorts product lead said the system can "find a seed audience" for a video. In our data, Shorts under 10 views after 24 hours ended under 100 views 99.3% of the time.

How long does it take a Short to take off?

In our recent Shorts that passed 1,000 views, the median one was past 1,000 by 8.8 hours after posting, and 83.1% were there within 24 hours.

Do subscribers matter for Shorts?

Subscriptions are one of the signals YouTube lists for recommendations. In our data the channel said more about a Short's views than the video did; the breakdown by subscriber count is in do followers matter for views.

Does the Shorts algorithm favour new uploads?

YouTube hasn't said so on any page we read. In our data, the median Short that passed 1,000 views had 99.6% of its final views within three days.

If you'd rather test one video on many channels at once, see the plans.

Read next