YouTube outlier score: definition, formula, ranges

A YouTube outlier score is a video's views divided by the median views of its channel's other mature videos in the same format. A score of 5x means five times that channel's typical result; Vidfora calls 2x or higher an outlier.

Everything else on this page is a consequence of that one sentence: which videos belong in the baseline, what happens when there are too few of them, and how much confidence a given number deserves.

The formula, written out

The short version is a division:

multiplier = views / median(views of the channel's other mature
                            same-format videos)

"Mature" means at least 30 days old. "Same format" means Shorts are compared with Shorts and long-form with long-form, never mixed; Vidfora classifies a video as a Short when its duration is 60 seconds or less. "Other" means the video being scored is left out of the median it is measured against.

The full version adds one adjustment, because a video that is six days old has not finished accumulating views:

expectedFraction(d) = d >= 30 ? 1 : clamp((d / 30) ^ 0.55, 0.1, 1)
baseline   = median(mature same-format peers, each normalized
                    to the 30-day scale) * expectedFraction(age)
multiplier = views / max(baseline, 1)

Three rules govern the peer group. When at least five mature same-format peers remain after excluding the video itself, the median comes from those. When there are fewer, Vidfora falls back to all same-format peers, with each peer's views normalized to the 30-day scale so that young peers do not deflate the baseline. A channel with only one video in a format has no peers at all and scores 1x, which is the honest answer rather than an invented breakout.

Scores are rounded to one decimal place and calculated against the dataset's snapshot time rather than the clock at the moment you load the page, so two people reading the same page see the same number.

Why the median and not the mean

A mean is pulled around by the very videos you are trying to detect. Take a hypothetical channel with six mature long-form uploads: 8,000, 9,000, 10,000, 11,000, 12,000 and 200,000 views. The mean of those six is about 41,700. The median is 10,500.

Now score the 12,000-view upload, excluding it from its own baseline. The remaining five are 8,000, 9,000, 10,000, 11,000 and 200,000. Their median is 10,000, so the video scores 1.2x: an ordinary upload, correctly described. Their mean is 47,600, so against a channel average the same video scores 0.3x, as though it had failed badly.

That is the failure mode of an average baseline. One breakout in the history makes every subsequent ordinary video look like a collapse, and it makes the next genuine overperformer look unremarkable. The median asks a different question. Half the channel's videos did better than this number and half did worse, so what does beating it actually mean?

Why the video is excluded from its own baseline

A breakout that sits inside its own baseline conceals part of its own size, and the smaller the channel, the worse the distortion.

Take a hypothetical channel with two mature long-form uploads: 20,000 views and 140,000 views. Include both in the baseline and the median is 80,000, so the big video scores 1.75x and never crosses the outlier threshold at all. Exclude the video being scored and its baseline is the other video's 20,000 views, so it scores 7x, which is what actually happened.

With thirty peers the effect shrinks to almost nothing, and with three it decides the answer. Excluding the video costs nothing when the sample is large and prevents a specific class of wrong answer when it is small, so Vidfora does it in every case.

Why Shorts are pooled separately

Shorts and long-form videos are distributed by different surfaces. A channel's Shorts median can sit an order of magnitude above its long-form median, or well below it, and neither says anything about the other.

If both were pooled, a channel whose Shorts routinely outdraw its long-form videos would show every Short as an outlier and no long-form video as one, and a channel with the opposite profile would show the reverse. In both cases the multiplier would be measuring which surface a video was served on rather than how well the idea performed against its own audience. One baseline per channel per format keeps the comparison about the video.

Why institutional channels sit outside the default rankings

Some channels overperform for reasons that do not transfer. Corporate newsrooms have wire access and breaking-news spikes. Television brands, studios and record labels distribute licensed footage and franchises. Megastar artist channels ride global fame. Their outliers are real numbers and they teach an average creator nothing, and left in an unfiltered ranking they crowd out everything you could learn from.

Vidfora suppresses a curated, deliberately narrow list of those channels from its default corpus-wide rankings. This is suppression, not deletion: the channels stay in the dataset and still appear on channel pages, in explicit niche selections and in searches. An independent creator working in the News niche is a legitimate model and stays in the default rankings.

A worked example

These numbers are hypothetical and chosen to be easy to follow.

A channel's mature long-form uploads have a median of 20,000 views. One of its videos, published eight months ago, has 140,000 views.

multiplier = 140,000 / 20,000 = 7x

That video reached seven times the channel's typical long-form result. Now consider a second video on the same channel, published six days ago, with 30,000 views. Six days is well short of maturity, so the baseline is scaled down:

expectedFraction(6) = (6 / 30) ^ 0.55 = 0.41
baseline            = 20,000 * 0.41   = 8,250
multiplier          = 30,000 / 8,250  = 3.6x

Without that adjustment the young video would have been divided by the full 20,000 and scored 1.5x, which would have described a strong early result as a below-average one. The curve is a single average shape applied to every topic, though, so it fits a search-driven tutorial and a news reaction differently. Treat a score on anything younger than a week as provisional.

What the ranges mean

General guidance for reading a multiplier, not a measurement of any dataset:

Read the multiplier next to the absolute view count, always. A 30x result on a 2,000-view baseline is 60,000 views, which may be a smaller opportunity than a 2x result on a large, relevant audience. Neither number should quietly veto the other.

Scores move, so date them

An outlier score is a dated observation, not a permanent property of a video.

Three things shift underneath it. Views keep climbing, on the scored video and on every peer. The video ages past 30 days and stops receiving the accumulation adjustment. The channel publishes more uploads, which moves the median, sometimes decisively if the channel changes topic or format.

So a video that read 8x in March can read 5x in September without anything going wrong. The channel got better, or the early number was flattered by an age adjustment that has since expired. When you record research, record the multiplier, the view count, the channel median and the date together. A multiplier on its own cannot be rechecked later.

How this differs from a channel-average score

Tools that score a video against a channel average inherit the problem described above: their baseline is dragged upward by the channel's own hits, so ordinary videos look like failures and the next real breakout looks ordinary. Tools that score against subscriber count measure something else entirely, since public subscriber numbers are rounded, include long-lapsed viewers, and say nothing about how one format performs.

None of these are wrong in principle; they answer different questions. What matters is that a multiplier is only comparable with another multiplier built the same way. Before putting two tools' numbers in the same column of a spreadsheet, check five things: median or mean, how many peers, Shorts pooled or separate, self included or excluded, and whether young videos get an age adjustment.

What the score cannot tell you

It is not a YouTube metric and not a prediction. It says nothing about watch time, click-through rate, retention or revenue, none of which are public for another creator's channel.

It cannot explain itself. A 12x video might owe its result to a collaboration, a news cycle, a paid placement, an external link or a thumbnail refresh, and public numbers usually cannot separate those from the idea.

It depends on the sample it was given. A channel that pivoted six months ago has a median describing a channel that no longer exists. Deleted, unlisted and members-only videos are invisible to any outside observer, so every baseline computed from public data is an approximation.

And it says nothing about feasibility. Relative performance cannot tell you whether you have the access, the equipment or the credibility to deliver the same kind of value. That judgment stays yours.

Used with those limits in mind, the score does one job well: it points your attention at the uploads where something unusual happened, on channels of any size. To see the current highest-scoring videos and the channel medians behind them, browse the public outlier leaderboards. For the step-by-step method of finding and reading outliers, by hand or with a tool, read how to find YouTube outliers.