On August 24, YouTube starts counting a view at the first frame, on every format, with no minimum watch time. Every metric with views in the denominator changes meaning overnight. Six days before the switch, we froze the baseline: engagement benchmarks for 1.47 million active channels, measured under the standard that is about to disappear.
On August 17, 2026, YouTube announced that from August 24 a public view will be counted the moment a video begins to play, "from the very first frame", on every format globally. Long-form video and live streams join Shorts, which have counted views this way since early 2025. The industry's working assumption that a YouTube view meant roughly 30 seconds of watch time stops being true in six days.
| Metric | Until Aug 23 | From Aug 24 | Who Sees It |
|---|---|---|---|
| Long-form public views | ~30s of playback | First frame | Everyone |
| Shorts public views | First frame (since 2025) | First frame | Everyone |
| Engaged views | Not separated | The old standard, renamed | Channel owner only, Analytics Advanced Mode |
| Earnings basis | Engaged metrics | Engaged metrics (unchanged) | Channel owner |
Three details matter for everything that follows. First, historical counts are not recalculated, so every video's public count becomes a blend of units accrued under two different definitions. Second, the stricter metric survives, but as "Engaged views" inside Analytics Advanced Mode, visible only to the channel owner: the public data ecosystem, every third-party tool, every media kit, every competitor comparison, loses the 30-second signal permanently. Third, YouTube itself expects the numbers to move: the announcement tells creators they "will likely see their total view counts increase faster moving forward". Earnings and Partner Program eligibility stay pinned to engaged metrics and do not change.
This is not a payout change, it is a measurement change. The view, the denominator under every engagement rate, CPM estimate, and rate card on the platform, gets redefined in six days. What follows is the baseline, measured before the switch.
Engagement rate here is the public-data definition used across the industry: likes plus comments, divided by views, averaged over a channel's 20 most recent long-form uploads. Measured by exact census across 1,460,009 active channels with 1,000+ subscribers and recent long-form output, the distribution under the outgoing 30-second standard:
Size changes what normal looks like. Engagement rate declines as channels scale, which is why comparing a 50k channel against a 2M channel on raw rate has always been a category error:
| Subscribers | Channels | P25 | Median | P75 |
|---|---|---|---|---|
| 1k to 10k | 94,777 | 1.98% | 3.77% | 6.81% |
| 10k to 100k | 983,194 | 2.02% | 3.84% | 6.69% |
| 100k to 1M | 327,222 | 1.69% | 3.29% | 5.77% |
| Over 1M | 47,725 | 1.23% | 2.56% | 4.50% |
These are the reference numbers. A 4% engagement rate is above the platform median at every size; a 7% rate is top-quartile everywhere; a mega-channel at 2.5% is exactly typical for its class. From August 24, the same audience behavior will produce lower published rates as the view denominator widens, which is precisely why this table needed to exist before the switch: it is the conversion anchor every post-change comparison will need.
Median 3.67%, middle half between 1.9% and 6.4%, declining with scale. If you quote an engagement rate after August 24, say which view standard it was measured under. This table is the old one's closing price.
Nobody has to guess what first-frame counting does to engagement metrics, because YouTube has been running both standards in parallel since early 2025. Shorts views count at the first frame; long-form views required 30 seconds. Same platform, same creators, same like button. Here are both distributions, measured by census on the same day with identical banding:
The first-frame distribution is not a scaled copy of the long-form one, it is compressed toward the low bands. Under the 30-second standard, one channel in five posts a long-form engagement rate above 7%. Under first-frame counting, fewer than one in ten does. The census medians: 3.67% long-form against 2.59% Shorts, and restricting both sides to the 1.26 million channels that post both formats barely moves it (3.76% against 2.59%).
Census comparisons can hide composition effects, so we also measured it within channels: a stratified 480-channel cohort, 397 of which actively post both formats. For the median dual-format channel, the Shorts engagement rate is 77.8% of its own long-form rate. The middle half of channels spans 0.47 to 1.29, and 64% sit below parity. The gap is real, but it is a distribution, not a constant: plenty of channels engage better on Shorts than on long-form.
Where the two standards coexist, the first-frame one reads lower: 29% lower at the platform median, 22% lower within the median channel. That is the observed scale of the world long-form metrics are about to enter, measured, not forecast.
The obvious explanation for that gap is denominator inflation: first-frame counting registers views the 30-second standard would have discarded, so every per-view rate dilutes. The within-channel data says the truth is stranger. Decomposing engagement into its two components for the 397 dual-format channels:
Likes per view are statistically indistinguishable across the two counting standards within the same channel. Whatever share of first-frame views are fleeting swipe-pasts, viewers still like short content at the same per-view rate they like long-form content. The entire engagement gap concentrates in comments, which run at less than a third of the long-form rate per view. Commenting is an investment behavior, and lean-back feed consumption does not produce it.
This reframes what August 24 does to metrics. If first-frame views were mostly phantom exposures, like rates would dilute alongside comment rates. They do not. The additional views the new standard counts are, by the like-button's testimony, real audience contact. What changes is the mix of behaviors a "view" implies, and any metric weighted toward comments will read the change hardest.
The first-frame view is not a fake view, it is a shallower one. Likes track it at parity; comments do not. Expect like-based metrics to hold up after August 24 far better than comment-based ones, and treat combined engagement rates as the blend they are.
The second structural difference between the standards is what they do to channel size. Under the 30-second standard, engagement rate falls steadily with scale. Under first-frame counting, it barely moves:
| Subscribers | Long-Form (30s Standard) | Shorts (First-Frame) | Within-Channel Ratio (Cohort) |
|---|---|---|---|
| 1k to 10k | 3.77% | 2.23% | 0.73 |
| 10k to 100k | 3.84% | 2.59% | 0.67 |
| 100k to 1M | 3.29% | 2.70% | 0.78 |
| Over 1M | 2.56% | 2.53% | 0.89 |
At the top of the platform the two standards nearly agree: for channels past a million subscribers, the medians are 2.56% and 2.53%, and the within-channel ratio is 0.89. The divergence lives in the small and mid-size bands, exactly where engagement rate does its economic work as the small creator's counterweight to reach. Under the 30-second standard, a 30k channel typically out-engages a 3M channel by half again; under the first-frame standard, that advantage largely evaporates from the public numbers.
The 30-second view rewarded channels whose viewers chose to stay. First-frame counting grades every channel on the same shallow contact, and it grades them remarkably alike. The metric that let small channels prove they punch above their reach is the one being retired.
Engagement rate is not a vanity number, it is a price input. Sponsorship rate cards, campaign reports, and influencer-marketing platforms quote it, and CPM-style deal math divides spend by views. Both sides of that arithmetic change units on August 24. The population with money on the line is large:
Two frictions arrive at once. The first is the vintage problem: a campaign measured in September will report views in a bigger unit than the July campaign it is compared against, flattering the later one. Every year-over-year report crossing August 24 inherits this discontinuity, permanently. The second is an information asymmetry that did not exist before: the metric closest to the old standard, engaged views, lives in Analytics Advanced Mode where only the creator can see it. Brands that want the stricter number must now ask for screenshots, and creators with strong engaged-view ratios have every incentive to volunteer them while others let the inflated public count do the talking.
Sponsored channels out-engaging the platform median cuts both ways. It says brands have been selecting for engagement, which is rational. It also means the channels being paid on these numbers sit in the part of the distribution with the most room to fall when the denominator widens.
121,002 sponsored channels, and every one of their rate conversations after August 24 involves a number that changed meaning mid-year. The winners will be the parties who name the unit first instead of discovering the discrepancy in a quarterly report.
The last service this dataset can perform is to record how wide "normal" already was, because every one of these numbers becomes a historical reference on August 24. By content category, under the 30-second standard:
| Category | Active Channels | Median ER |
|---|---|---|
| Education | 150,294 | 4.42% |
| Gaming | 133,146 | 4.40% |
| People & Blogs | 397,668 | 3.82% |
| Howto & Style | 69,113 | 3.81% |
| Entertainment | 233,505 | 3.63% |
| Comedy | 20,427 | 3.45% |
| News & Politics | 51,392 | 3.36% |
| Science & Technology | 42,897 | 3.34% |
| Sports | 45,958 | 3.24% |
| Music | 172,432 | 2.77% |
| Film & Animation | 47,408 | 2.68% |
And by creator country, where the spread is wider than most benchmark guides admit:
| Country | Active Channels | Median ER |
|---|---|---|
| Brazil | 62,959 | 7.58% |
| United Kingdom | 35,995 | 4.34% |
| France | 23,585 | 4.26% |
| United States | 248,070 | 4.10% |
| Thailand | 24,698 | 3.16% |
| South Korea | 40,051 | 3.03% |
| India | 232,427 | 2.97% |
| Japan | 53,320 | 2.55% |
| Indonesia | 56,903 | 2.36% |
| Vietnam | 28,818 | 1.97% |
A Brazilian channel at 5% is under-performing its market median; a Vietnamese channel at the same 5% is a standout. An Education channel at 4% is ordinary; a Music channel at 4% is exceptional. These context gaps are all larger than the 22% cross-standard gap this study measures, which is the quiet argument of the whole dataset: engagement rate has only ever meant something relative to a reference population, and from August 24 it also only means something relative to a date.
Benchmarks were already local. Now they are also dated. Label both, or the number is noise.
Nothing about audiences changes on August 24. The same people will watch, like, and comment at the same rates they did on August 23. What changes is the ruler, and the recorded history of the platform splits into a before and an after. Three ways to act on that:
Screenshot your channel's engagement rate and views this week: that is your number in the old unit, and this study's tables tell you where it stood. After the switch, quote Engaged Views from Advanced Mode when a stricter number serves you, and expect your public rate to drift down without your content getting worse.
Add a measurement-standard line to every campaign report and rate card from September onward, and re-baseline any benchmark that crosses August 24 rather than trending through it. When comparing creators, comment rates now carry more discriminating signal than combined engagement rates: the like survives the new standard at parity, the comment does not.
The 30-second view leaves the public record permanently, so the pre-change distributions in this study are as close to a preserved standard as the ecosystem gets. We have frozen our 480-channel panel and will re-measure the same channels after the switch to publish the actual inflation, measured, not modeled.
The last honest engagement rate is 3.67%. Not because the next one will lie, but because it will answer a different question. Know which question your numbers are answering.
Journalists are welcome to cite this study with attribution to CreatorDB. For methodology questions, custom cuts (by country, category, or channel size), or comment on the August 24 view-counting change, contact hello@creatordb.app. We can usually turn custom breakdowns around same-day. The post-change re-measurement of the 480-channel panel publishes on this page's sibling in early October 2026.
CreatorDB (2026). The Last Honest Engagement Rate: YouTube Engagement Benchmarks Before the First-Frame View Change. Retrieved from https://creatordb.app/research/youtube-view-count-change/
Every figure here comes from the CreatorDB index of 30M+ creators across YouTube, Instagram and TikTok, with per-channel performance, audience demographics and sponsorship history. The API gives you direct access to all of it.