The Ad That Never Served

AIR Media-Tech's RPM study — 300 channels on one multi-channel network's own roster, 3,595 channel-months of Studio revenue lines running May 2025 to May 2026, last updated 1 July 2026 — carries one sentence that does more work than the thirteen-row table underneath it: "A Gadgets channel can post a healthy $5.73 CPM and a $2.33 RPM at the same time, because only a third of its views ever serve an ad."

Read the two halves separately. The first half is two prices attached to the same thousand views in the same 300-channel sample published 1 July 2026, and they differ by a factor of about two and a half. The second half is the reason, and the reason is not a rate. It is a denominator.

Every figure in this chapter comes from one vendor, and that vendor's two 2026 studies disagree with each other by a factor of 2.6 on one of the numbers printed below. That disagreement is worked through in full later in the chapter rather than left in a note, because it decides how much weight any of this can carry.

First, what the sample is, because in this book the sample is part of every number that comes out of it. The measurer is a multi-channel network that sells paid channel audits, and the 300 channels in its study of 1 July 2026 are its own clients rather than a draw from the platform. In that 1 July 2026 study they run from 10,000 to 50 million subscribers across thirteen niches, and the study states that its figures come from Studio rather than from advertiser-side estimates, which is the reason this dataset is worth reading at all. Four of the thirteen niches in that 1 July 2026 table rest on fewer than ten channels each, and one of those four rests on three. A roster is not a platform, and the audience geography of that roster sits inside every rate below without being separated out anywhere on the page.

Estimated monetized playbacks is the metric the mechanism runs on, and YouTube defines it. YouTube's ad-revenue analytics help page, retrieved 5 September 2026, describes it as the number of times a video was watched with ads, and gives a worked example on the same page: ten views, eight of them with ads, is ten views and eight monetized playbacks. The two counts are not the same count, and the page keeps them apart.

The rest of the definitional work sits on that same page. On it, as retrieved 5 September 2026, RPM is divided by 1,000 views — every view, whether an ad ran on it or not — and is struck after the revenue share, with memberships, Premium, Super Chat and Super Stickers inside it. CPM, on the same page of 5 September 2026, is divided by 1,000 ad impressions, comes before the revenue share, and covers ads and Premium only. Two prices, two denominators, and only one of those denominators is a count of the audience the creator actually had.

The monetized playback rate is the ratio between those two counts: estimated monetized playbacks divided by views. It is arithmetic performed on two metrics that YouTube's ad-revenue analytics help page defines as retrieved 5 September 2026, rather than a finished figure that page hands over. In the ten-view example printed on that page as retrieved 5 September 2026, the rate would be 80%. Nothing measured in 2026 runs anywhere near that.

A view that serves no ad is not a fault in the channel. An impression requires an advertiser bidding on that viewer, in that country, at that moment. A YouTube Premium subscriber sees no advertising, so that view can serve no ad impression, and the money still reaches the creator, because YouTube's ad-revenue analytics help page, retrieved 5 September 2026, puts Premium revenue inside both CPM and RPM. There is also the video itself: YouTube's advertiser-friendly content guidelines, retrieved 5 September 2026, separate content that is demonetized from content that is limited, which is why a channel can be fully monetized in 2026 and still earn a fraction of its category's median rate. This volume names that distinction once, here, and does not develop it.

What nobody has published is the split. Neither the 300-channel study last updated 1 July 2026 nor any other document retrieved for this volume in 2026 decomposes a category's unmonetized views by cause. The rate is measured. The reasons behind it are not.

Across those 300 channels and 3,595 channel-months, the median monetized playback rate published 1 July 2026 was 53%. Slightly more than half of a median channel's views carried an ad, and the half that did not still counted in the denominator of that channel's RPM.

The full column, from AIR Media-Tech's RPM study of 300 channels and 3,595 channel-months last updated 1 July 2026, medians throughout, with the four niches resting on fewer than ten channels marked:

\* Fewer than ten channels, flagged directional by the vendor itself. Business & Finance rests on three.

The Studio CPM column is not pay and is not printed as pay. It sits before the revenue share, it is struck per 1,000 ad impressions rather than per 1,000 views, and the same vendor states in its study last updated 1 July 2026 that advertiser-side estimates run roughly 1.5 to 2 times above the Studio CPM its own 300 channels reported. A CPM table describes what advertisers spent. Nobody is paid one.

Read down the first column and the ordering does not hold. Education & Science monetized 77% of its views and posted a median RPM of $10.22 in that 300-channel table of 1 July 2026, while Lifestyle monetized 61% and posted $2.98, and Gadgets & Tech monetized 34% and posted $2.33 — a category with barely half Lifestyle's rate, landing within seventy cents of it. The monetized playback rate is one of three inputs, not a ranking, and a table sorted by it is not a table sorted by money.

The second input is ad load. Median ads per monetized view across the same 300 channels and 3,595 channel-months, published 1 July 2026, was 1.55. Transport ran highest at 2.55 on seven channels, and the constraint there is runtime rather than subject: a video has to be long enough to hold several mid-roll slots before it can serve several ads, and long car, aviation and build videos are. Business & Finance ran 1.39 on three channels, and Kids & Teens ran 1.29 in that same 300-channel table of 1 July 2026.

The two ratios move independently, which is easy to see in the table published 1 July 2026 and easy to miss in the argument. Gadgets & Tech and News & Politics both ran 1.71 ads per monetized view in the 300-channel sample published 1 July 2026 while their monetized playback rates were 34% and 53%. Business & Finance and Health & Sport both ran 1.39 in that same 300-channel table of 1 July 2026 while their rates were 56% and 44%. Ad load is not a consolation prize for a low monetization rate, and it is not a symptom of one. They are two separate numbers that happen to multiply.

Those two ratios are the whole of what this chapter prices. It does not go on to price ad slots by format, or to ask what a skippable impression is worth against a bumper, because a category's monetized playback rate and its ad load already carry the mechanism, and neither of the 2026 studies this chapter reads breaks a slot out by type.

Runtime is also a format decision rather than a subject decision. A category does not carry a high ad load because of what it is about; it carries a high ad load because the videos in it are long enough to hold the slots, and the length is chosen by the person uploading. That is the reason this volume refuses to separate the two questions it prices. The subject sets what advertisers will bid. The format sets how many of those bids ever land.

Multiply the two ratios and the product is ad impressions per view, which is the number that turns a CPM into a creator's revenue. In the 300-channel table of 1 July 2026, Gadgets & Tech runs 34% of views monetized at 1.71 ads each, which is about 0.58 impressions for every view, and that multiplication is this book's own arithmetic on two medians published on the same page that day. Crafting & Handmade runs 57% at 1.90, which is about 1.08 impressions per view on the same two medians from those 300 channels, published 1 July 2026. A Crafting view generated roughly twice the ad inventory of a Gadgets view in that sample before any price was applied to either.

Now put the prices back. In the same 300-channel sample published 1 July 2026, Crafting & Handmade posted a median Studio CPM of $3.42 against Gadgets & Tech at $5.73 — and a median RPM of $2.39 against Gadgets at $2.33. The category with the lower advertiser price paid its creators more per thousand views. That is the whole mechanism in one pair of rows, and it needs no theory about what advertisers want.

The reverse case is in the same table and is worth holding beside it. Gaming and Music both posted a median Studio CPM of $4.05 in that 300-channel sample of 1 July 2026, and both landed near 0.8 ad impressions per view on their published rates and loads, which is this book's own multiplication of two pairs of medians from that page. Their median RPMs were $2.05 and $2.28 in the same 300-channel study of 1 July 2026. Identical advertiser prices and near-identical ad inventory, and twenty-three cents of difference in what the creators were paid, which the two ratios do not explain and the study does not account for.

Kids & Teens is where the multipliers compound instead of cancelling. In that 300-channel table of 1 July 2026 it holds the lowest median Studio CPM at $1.59, the lowest ad load at 1.29, and the lowest monetized playback rate at 32%, and it posts the lowest median RPM at $0.33. Three low numbers multiplied produce a fourth that is lower than any of them looks, which is the arithmetic of the thing. It is not the whole of what makes that category different in 2026, and the rest of it is not a rate story at all.

The reconstruction can be pushed one step further, and it is worth pushing, because it is where the dataset starts to come apart. Take a niche's median Studio CPM from the 1 July 2026 table, multiply by its monetized playback rate and its ads per monetized view from the same table, then take 55% of the result, because YouTube's partner earnings overview, retrieved 5 September 2026, sets the creator's cut at 55% of net revenues from ads on public videos.

Three warnings ride on that arithmetic, and the first is fatal to reading it as a check. A median of a product is not the product of medians: each of these four figures is a middle value drawn from a different ordering of the same 300 channels, and multiplying them describes no channel in the sample. The second is that RPM's numerator is wider than ad revenue, since memberships, Premium, Super Chat and Super Stickers sit inside it per YouTube's ad-revenue analytics help page retrieved 5 September 2026, and the reconstruction below omits all of them. The third is the word net in that 55%, which is struck after costs no creator sees itemised in 2026. What follows is an illustration of the mechanism. It is not a test of the table.

Run on the figures published 1 July 2026 across 300 channels and 3,595 channel-months, and set against the same study's measured medians, the illustration comes out like this — every figure in the middle column being this book's own multiplication of four numbers on that one page:

Twelve of those thirteen rows land within about sixty cents of the median RPM measured across those 300 channels and 3,595 channel-months, and eight of them land within forty, which is closer than four multiplied medians have any right to come. The misses also point mostly one way: in ten of the thirteen rows of the 300-channel table of 1 July 2026 the reconstruction comes in under the measured RPM, which is what a calculation that omits memberships, Premium, Super Chat and Super Stickers should do in 2026.

Then there is Education & Science. It was measured at $10.22 in the 300-channel study of 1 July 2026, and the same study's own CPM, monetized playback rate and ad load reconstruct to about $4.61 per thousand views. That is short by more than a factor of two, on a page where every other niche lands within change. Something in the top row of that table is not being carried by advertising, and the study published 1 July 2026 does not say what.

That is where the vendor's other 2026 study arrives, and it makes the top row worse rather than better.

AIR Media-Tech's Shorts study — 274 channels, 3,044 channel-months, read through the YouTube Analytics API, published 23 June 2026 — reports Kids long-form monetized playback at 12.1%, phrased on that page as roughly one long-form view in eight carrying a monetized ad. The RPM study published eight days later, on 1 July 2026, across 300 channels and 3,595 channel-months, puts Kids & Teens at 32%. One vendor, one year, one metric, one category, and two numbers 2.6 times apart.

The same pair of 2026 studies disagrees on Education as well, and disagrees in the direction of the gap the reconstruction just opened. The 300-channel study of 1 July 2026 gives Education & Science a median RPM of $10.22. The 274-channel study of 23 June 2026 gives Education long-form RPM as $14.97 for medium channels and $18.23 for small ones. The category that will not reconcile against its own inputs is also the category the measurer could not hold still across eight days.

Part of that is explicable, and the explicable part is not large enough. The samples are different sizes, 274 against 300, drawn from the same roster but not from the same list. The 23 June 2026 study reports Education by channel-size tier while the 1 July 2026 study reports a single median for the niche, and a small-channel tier is not the same object as a whole-niche median. Different channel-months are pooled behind each. None of that produces a factor of 2.6 on a ratio as blunt as monetized playbacks over views, and the vendor does not reconcile the two figures on either page.

So the standing rule for the rest of this book is a grammatical one. No figure out of either 2026 study is "the 2026 figure," and none of them travels without its sample. The 53% median is a median across 300 channels on one network's roster over 3,595 channel-months to May 2026, and it is written that way every time it appears, because a sentence that drops the sample has quietly promoted a roster into the place where a platform belongs.

What survives all of this is the mechanism, and it survives because it is definitional rather than measured. A category's RPM is set by what advertisers pay per impression, by how many of its views serve an impression at all, and by how many impressions each of those views carries, with the revenue share taken off the end at 55% for ads on public videos per YouTube's partner earnings overview retrieved 5 September 2026. That chain is arithmetic. A category posting a higher CPM and a lower RPM than its neighbour is not a paradox to be explained away. It is the expected result when the first category monetizes about a third of its views and the second monetizes well over half, as Gadgets & Tech and Crafting & Handmade did in the 300-channel sample published 1 July 2026.

What does not survive is any specific rate. Whether Kids & Teens monetized 32% of its long-form views on a 300-channel sample or 12.1% of them on a 274-channel sample in 2026 is a measurement, and the only public measurement contradicts itself. The mechanism is sound, and it is well illustrated by the Gadgets chain the vendor itself printed. As a finding about any named category it is a hypothesis, and this book prints it as one.

The dataset also names nobody, which limits it in a second way. Both 2026 studies report medians and percentile boundaries and no channels at all. No winner is named and no failure case is published beside any median in either of them, so in the study of 1 July 2026 there is no channel in Gadgets & Tech whose 34% can be looked up and no channel in Education whose $10.22 can be checked. The boundaries that were published show how little a median describes: Gadgets & Tech ran a 25th percentile of $0.70 and a 75th of $3.71 in that 300-channel sample of 1 July 2026, so the $2.33 in the headline names a position in a list rather than a channel that exists.

As of September 2026, this was unresolved. No second measured dataset of monetized playback rates by category had been published, and the mechanism in this chapter therefore rests on one vendor measuring one roster twice and getting two answers eight days apart. What would close it is specific enough to build: a US-heavy sample with more than twenty channels in every niche, reporting estimated monetized playbacks and views as raw counts rather than as medians, so that the ratio can be computed by the reader instead of taken from the measurer. Nothing shaped like that existed in public when this was written.