What is an outlier video, and how is its score worked out?

Short answer

An outlier video is one that got far more views than its creator usually gets. The usual measure divides its views by the median views of the creator’s recent videos and writes the result as a multiple: 30,000 views on a channel whose recent median is 2,000 is ×15. No network publishes an official formula, so the window, the baseline, the format and a minimum view count are choices, and each changes the answer.

Written by the Mellow team · Updated November 8, 2026, sources read November 8, 2026

The formula

An outlier score compares a video with its own creator, not with the whole network:

outlier score = the video’s views ÷ the creator’s usual views

The usual views are a baseline, most often the median views of the creator’s recent videos. ×1 means the video did what that creator’s videos usually do; ×15 means fifteen times as many views. So a small channel can post a ×20 video and a very large one a ×0.5 video in the same week.

YouTube applies the idea to your own channel: YouTube Studio’s Content tab compares each video’s views with the channel’s typical performance, and for retention, typical means what your last 10 videos of a similar length maintained (YouTube Help). An outlier score extends that comparison to any channel whose numbers you can read.

Why the median and not the average

One viral video drags an average up and barely moves a median. With invented numbers: a channel’s last 10 videos had 2,000 views each, except one with 100,000. The average is 11,800; the median is 2,000. A new video with 20,000 views is ×10 against the median and ×1.7 against the average, which makes the channel’s best recent result look ordinary.

A worked example

Invented round numbers, no real channel:

	Channel A	Channel B	Channel C
Median views of its last 10 videos	2,000	50,000	2
Views of the video	30,000	30,000	40
Outlier score	×15	×0.6	×20
Read as	A clear outlier	Below this channel’s usual	Noise, dropped by a 1,000-view floor

The same 30,000 views are a breakout on one channel and a weak week on another. Channel C scores highest and means least, which is what a minimum view floor is for.

The choices that change the answer

Choice	Example settings	The trade-off
Window: how recently the video was published	The last 3, 7 or 30 days	Short catches videos while they rise, but a two-day-old video is measured against older ones that had weeks to collect views. Long gives settled numbers and older finds.
Baseline: how many recent videos	10, 20 or 30	Fewer reflect the channel now and move with every upload; more are steadier and reach further back.
The video in its own baseline	In or out	Out keeps the baseline clean. It matters little for a median, a lot for an average.
Format	Shorts and long videos apart	Together, a Short on a long-video channel scores high for its format alone.
Network	One score per network	A ×5 on TikTok and a ×5 on YouTube rest on different definitions of a view.
Minimum views	A floor such as 1,000	It stops ×20 on 40 views from topping the list; too high a floor hides small channels’ real breakouts.

Shorts on YouTube. On a standard channel, a square or vertical video of up to three minutes uploaded since 15 October 2024 is a Short (YouTube Help). The Data API has no field that marks one; it gives the length, so a split by length is an approximation. search.list’s videoDuration=short means under four minutes, a different thing (Search: list).

How YouTube counts a view changed. From 31 March 2025 a Shorts view counted each time a Short started to play or replay. From 24 August 2026 a public view counts the moment any video begins to play, autoplay included (revision history; video resource). The documentation does not say whether earlier counts were recalculated, so a baseline reaching back past late August 2026 may mix two definitions.

Outlier is not the same as trending

YouTube’s Trending Charts weigh view count, how fast views grow, where they come from, the topic, the video’s age, and how the video performs against the channel’s other recent uploads (YouTube Help). An outlier score is that last signal alone. So a large channel’s ordinary upload can chart on raw views and speed while scoring about ×1, and a ×15 video from a channel with a 2,000-view median can stay far from any chart.

Working it out on YouTube, by hand or with an AI agent

The YouTube Data API v3 has everything the score needs. An AI agent that can call an API makes the same calls with your key; ask it to print the window, the number of videos, the median and the floor beside every score, or the ×N cannot be checked.

Find candidates. search.list with q, type=video, publishedAfter (your window) and order=viewCount, up to 50 results a page. Results carry IDs, titles and publish times, not view counts (search result).
Read their views. videos.list with part=statistics,contentDetails and the IDs returns statistics.viewCount and contentDetails.duration (Videos: list).
Build each baseline. channels.list with part=contentDetails gives relatedPlaylists.uploads (channel resource); playlistItems.list on that playlist lists the uploads, 50 a page (PlaylistItems: list); one more videos.list reads their views. Take the median of the last N.
Divide, apply your floor, keep Shorts apart.

Cost. By default a project gets 100 search.list calls a day in their own bucket, and 10,000 units a day for all other methods. videos.list, channels.list and playlistItems.list cost 1 unit a call, every further page costs again, and quotas reset at midnight Pacific Time (quota calculator). A baseline costs about three units per channel.

In a product. YouTube’s Developer Policies say an API client, meaning a website or application you develop, must not use API data to create new or derived metrics, and must delete or refresh other people’s statistics within 30 calendar days (Developer Policies). Since 1 June 2026, audited developers with an analytics use case who accept an amendment through the quota extension request may calculate listed kinds, such as custom channel scores and ratios, kept apart from YouTube’s own figures (derived metrics policy). Whether a per-video score qualifies is YouTube’s call.

TikTok and Instagram: what the official APIs allow

TikTok. The Research API is for qualified researchers in the U.S., Europe, Canada and Brazil, at academic or not-for-profit organisations, independent of commercial interests (Research API), so a commercial tool cannot use it to search other people’s videos. The Display API lists an authorizing user’s own public videos, 20 a page (List Videos), each with view_count (Video Object): enough for your own scores.

Instagram. Business Discovery returns another professional account’s posts, named by username, with like_count, comments_count and view_count. It runs from your own professional account, skips age-gated accounts, and its view_count includes paid views, so a boosted post can look like an outlier (Business Discovery). Hashtag search returns no view count and no author (IG Hashtag Top Media). The Creator Discovery API searches creators and shows their recent media, but only for brands eligible for Instagram’s creator marketplace, after app review (Creator Discovery API).

Neither network documents an official search of public videos by topic that returns view counts, as search.list and videos.list do on YouTube.

Where Mellow Hub fits

Since 8 November 2026, Mellow Hub’s Trends works this ratio out for TikTok and Instagram: a video took off at ×2 its creator’s median views and at least 5,000 views, and none is shown when the creator has too few videos to tell. On YouTube it shows YouTube’s own figures only, until YouTube approves derived figures such as this one. Hub also comes in afterwards: once you or your AI agent have made the post, Hub checks it against each network’s rules and publishes it now or at a set time to Instagram, TikTok, YouTube, X, LinkedIn, Threads, Bluesky, Pinterest, Facebook and Telegram. An agent reaches Hub through its MCP server (what an MCP server is) or REST API, within the channels, mode, daily limit and expiry the account owner set.

Hub has no free plan: a new account starts with a 7-day free trial of a monthly plan. Starting it takes a card, and $0 is charged that day; cancel before the 7 days end and nothing is charged. After the trial the plan renews monthly, from $19 a month (Solo), and the first paid month is 20% off. A yearly plan is charged on the day you choose it, with no free trial. Publishing is unlimited on every plan, under a fair-use ceiling that stops automated floods (900 publications a month on Solo), where one publication is one post on one account, so a post sent to five networks counts as five toward it. There are no automatic overage charges.

Questions people also ask

What counts as a good outlier score?

No network sets a threshold. A score only means something next to its settings: ×3 against a steady median of 50,000 views is a bigger event than ×20 on 40 views. Pick a window, a baseline and a floor, keep them fixed, and compare scores only with scores made the same way.

Can I see this for my own channel without the API?

Partly. YouTube Studio’s Content tab compares each video’s views with your channel’s typical performance, and its retention card compares against your last 10 videos of a similar length. Studio shows only your own channel, so for anyone else you work the score out from public view counts.

Does a high outlier score tell me why the video did well?

No. It says the video reached far beyond that creator’s usual audience, not why. The topic, the title, the thumbnail, the timing, a link from outside the network or paid promotion can each be the cause; on Instagram, the view count Business Discovery returns includes paid views.

Is an outlier score the same as engagement rate?

No. Engagement rate divides interactions by views or impressions, and the networks define it differently (what engagement rate is). An outlier score divides views by the creator’s own usual views, so it says how far a video travelled, not how people reacted to it.

How many recent videos should the baseline use?

There is no right number, only a trade-off. Ten reflects the channel as it is now and is what YouTube Studio uses for typical retention; twenty or thirty are steadier but reach back into older uploads. On YouTube, a baseline that crosses late August 2026 may mix two ways of counting views.

Sources

Other companies’ features change. Every statement about them rests on these pages, as they read on the date above.