The Top 1% of Videos Captured 61% of Plays. What Does Your Brand Report Miss?

Across 50,182,617 social video posts in dig’s corpus over 90 days, the top 1 percent of posts captured 61.4 percent of all plays. The top 10 percent captured 92.0 percent. The bottom half captured 0.50 percent between them.

If your brand report relies on the highest-play videos, it shows where attention concentrates. But recurring complaints, questions and experiences can appear across smaller posts without any one of them becoming a hit. Our analysis examines why those posts deserve a separate look.

A short note on terminology and timing before the figures.

Throughout this analysis, we use recorded play counts, not unique audience reach or interactions such as likes, comments and shares. The field is video_play_count, taken from the most recent scrape of each post. It is the one engagement field reliably populated across TikTok, Instagram and YouTube. It does not establish how many distinct people watched a post, and it does not establish identical measurement across platforms; the method note at the end records what we did and did not treat as equivalent.

The 90-day window defines which posts entered the analysis. Play counts are cumulative totals recorded at each post’s latest scrape, not plays generated during those 90 days. A video that was published two years ago and is still being scraped carries its lifetime play count into this pool. That matters when reading "captured 61 percent of plays": the figure describes the share of all recorded plays across the posts in this pool, not plays earned inside the window.

What 50 million videos look like when you line them up by plays

The set is 50,182,617 video posts analysed by dig between 13 July and 11 October 2026 across TikTok, Instagram and YouTube, split roughly 47 percent TikTok, 41 percent Instagram and 11 percent YouTube. The window describes when each post was last scraped by dig, not when it was published; publication dates vary and are not what the pool was built from.

We ranked the posts by plays and asked a simple question: if you sorted every video in this set by plays, where do the plays actually sit?

For the broader shape of what people are actually saying about your brand on social, this piece sits one step upstream. It describes the population from which any such view has to be drawn.

The top one percent captured sixty-one percent of plays

In an analysis of 50,182,617 social video posts across more than 240 projects over 90 days, the top 1 percent of posts captured 61.4 percent of all plays and the top 10 percent captured 92.0 percent. The bottom half of all posts captured 0.50 percent between them. The median post received about 3,967 plays against a mean of about 134,046.

501,910 posts, one percent of the set, captured 61.4 percent of all plays. Widen the window to the top ten percent, 5,018,953 posts, and you reach 92.0 percent of plays. The other 90 percent of posts in the set, roughly 45 million videos, divide 8 percent of the plays between them.

Widening from the top 1 percent to the top 10 percent adds only 31 points of coverage of plays. It adds 4.5 million posts.

What that means for a report built on the top posts

If your report relies on the highest-play examples, it emphasises the small share of posts where plays concentrate. That is a legitimate view. It is different from a view of every theme people posted about, and the gap between the two is what this piece is about.

That is one reason why manual TikTok scrolling fails as a strategy for understanding a brand: scrolling arrives at the top either way, from an algorithm or from a human picking the loudest clips.

Half the posts captured half a percent of recorded plays

The bottom half of the set, 25,089,136 videos, captured 0.50 percent of total plays between them. The median post received about 3,967 plays.

That shape has a practical consequence. If a report analyses only the top of a plays-ranked list, 25 million posts are present in the pool and absent from the analysis. Whether that matters depends on what you are trying to see. It matters a lot if what you are trying to see is a pattern that showed up across many small posts without any single one travelling.

Low plays are not automatically low signal

A product fault described by hundreds of people in smaller videos can disappear from a report that analyses only the highest-play posts. The same fault described once by someone with three million plays will dominate it.

High-play posts show where attention concentrates. Recurring themes across smaller posts show patterns that attention alone can hide. Both findings can be true at the same time, and they answer different questions.

Why the mean is a poor answer to "what does a typical post receive?"

Median plays per post: about 3,967. Mean plays per post: about 134,046. The mean is roughly 34 times the median.

The distribution is highly skewed: a small share of posts accounts for most plays. The mean therefore gives a very different picture from the median.

The mean remains useful for some questions. It is the right number when you want to know the total divided by the count. It is a poor answer to "what does a typical post receive?", because it is set by the top of the curve and describes almost nothing in the middle of it. The median is the right benchmark for a typical post, and the distribution shape should be reported alongside any single figure.

What this does to a sentiment read

Sentiment can be read three ways from the same pool, and each answers a different question.

Sentiment per post counts each post once, regardless of how many plays it received. It describes the tone of what was posted. It gives equal weight to a creator with four million followers and a reviewer with four hundred.

Sentiment per author aggregates posts to the account that made them before scoring. It describes the tone of distinct voices. It stops a single prolific account from swinging the number by posting ten times.

Play-weighted sentiment gives more weight to posts with more recorded plays. It does not establish which sentiment individual viewers encountered or retained.

A single large angry account and several thousand smaller measured ones will produce different results across the three. The useful move is to report all three on anything consequential, and state what each measure describes. For the record of sentiment analysis for video content read from the full pool rather than only the top, the shape to look for is the agreement or disagreement between the three.

In this dataset, half the posts accounted for just 0.5 percent of recorded plays. Smaller posts can still contain recurring themes worth investigating.

What to ask about any report you are handed

Four questions.

• How many posts is this built on?
• How were they selected: by plays, at random, or all of them?
• What is the median post in the set, not the average?
• Can you show me the themes that recur in smaller posts, not only the ones that travelled?

If the answer to the last one is "we would have to read them", that is where a pipeline that reads what is said and shown inside each post stops being a convenience and starts being a scalable way to identify recurring themes across the monitored videos, not only the concentrated attention at the top. For the shape of that record, see reading what is said, shown and felt inside a video.

Key takeaways

• Across 50,182,617 social video posts analysed by dig over 90 days, the top 1 percent of posts captured 61.4 percent of all plays and the top 10 percent captured 92.0 percent.
• The bottom half of all posts, more than 25 million videos, captured 0.50 percent of plays between them.
• The distribution is highly skewed: a small share of posts accounts for most plays. The mean of about 134,046 plays per post against a median of about 3,967 is a signature of that skew. Averages set by the top of the curve describe the typical post poorly.
• A report that surfaces only the highest-play posts can overlook patterns that recur across thousands of smaller ones. Whether that matters depends on the question you asked.
• Plays measure recorded views at last scrape, not unique audience reach or interactions. Three sentiment reads are possible from the same pool (per post, per author, play-weighted) and they answer different questions.

Frequently asked questions

What share of social video plays comes from the top posts?

In dig’s analysis of 50,182,617 social video posts whose latest scrape fell inside a 90-day window, the top 1 percent of posts captured 61.4 percent of all recorded plays and the top 10 percent captured 92.0 percent. The distribution is highly skewed, with the mean about 34 times the median.

Widening from the top 1 percent to the top 10 percent adds only 31 points of coverage of plays while adding 4.5 million posts. The practical consequence for a report that analyses only the highest-play examples is that it can overlook patterns which recur across thousands of smaller videos.

Is a social listening report representative of the whole conversation?

That depends on how the report selects posts, not on how many it was built from. A report that analyses only the top of a plays-ranked list describes where attention concentrates. In dig’s analysis the bottom half of all video posts captured just 0.50 percent of total plays, so recurring themes in that half will not appear in an analysis limited to the highest-play posts, even though those posts number more than 25 million.

Representativeness is a function of selection. A report built on millions of posts can still analyse only the loud end of them. The useful question is which selection the report used.

Why is average plays per post a misleading benchmark?

Because the distribution is highly skewed. In dig’s sample the mean was about 134,046 plays per post against a median of about 3,967. The average is set by the top of the curve and describes the typical post poorly.

Use the median as the typical-post benchmark, and report the distribution shape alongside any single figure. The mean remains useful when you want the total divided by the count; it is a poor answer to "what does a typical post receive?"

Method note

Figures come from a single cross-client aggregate query over the dig corpus, run on 2026-10-11. The base pool is posts on TikTok, YouTube and Instagram with a recorded play count greater than zero whose most recent scrape sits within the preceding 90 days (approximately 13 July to 11 October 2026). The 90-day window selects which posts entered the analysis; the play counts attached to those posts are cumulative totals recorded at each post’s latest scrape, not plays generated during the window.

Posts are keyed by platform post ID and are not duplicated across projects in the aggregate: a video monitored by two client projects is counted once in the figures above. Zero and missing play counts are excluded by the greater-than-zero filter.

The three cohort shares (top 1 percent, top 10 percent, bottom 50 percent) are derived by percentile boundaries on the pool’s play distribution. Top 1 percent boundary: about 2.1 million plays; 501,910 posts above the boundary captured 61.4 percent of plays. Top 10 percent boundary: about 117,000 plays; 5,018,953 posts above the boundary captured 92.0 percent of plays. Bottom 50 percent: 25,089,136 posts at or below the median (about 3,967 plays), capturing 0.50 percent of plays.

Platform view and play metrics have different definitions, which we did not attempt to normalise here. We pooled the recorded counts to describe concentration within this corpus, rather than treating them as equivalent measures of audience exposure. Readers should treat the pooled figures as a description of the pool as dig scraped it, not as a cross-platform audience statistic.

The project count is a cardinality estimate with roughly 2 percent error, which is why the piece says "more than 240 projects" rather than naming an exact number. The sample is dig’s corpus, which is what dig’s customers asked us to monitor, and skews toward brands, categories and events under active watch. No individual customer is identifiable and none is named.