The Rate Card Nobody Measured

No independently audited dataset of YouTube RPM by category exists in public. Every rate card in circulation in 2026 is a vendor aggregation, and of the four largest, exactly one discloses how many channels it counted. This chapter reads all four the way a hostile reader reads them: what population was measured, by whom, across what period, and whether the page holds one number consistent from its own table to its own FAQ.

None of the four is disinterested. vidIQ sells growth software to creators. AIR Media-Tech is a multi-channel network that sells paid channel audits. The remaining two publish their tables as free content on commercial sites. There is no neutral rate card in this category, because no neutral party has built one.

The loudest of the four is Influencer Marketing Hub's page "How Much do YouTubers Make in 2026?", updated 24 July 2026, which prices seven categories per million views. Its figures are reproduced below as that page carried them on 24 July 2026, with one column added: the same numbers divided by a thousand, which is this book's arithmetic and puts them in the units a Studio revenue report actually uses.

That page discloses no sample size, no method and no date range. It carries one hedge, and the hedge is the most informative sentence on it: "These are broad benchmarks rather than guaranteed payouts, but they reflect the monetization patterns repeatedly discussed by creators and industry tracking platforms."

Read that as a methods statement, because that is the only thing it can be. The evidence base named on the page is creators discussing patterns, plus industry tracking platforms it does not name. A page whose stated source is a conversation has disclosed its sampling frame, and the frame is a conversation.

The added column matters more than it looks. Per million views is not a unit any Studio report produces, and the choice of unit is doing work: $15,000 to $40,000 is a figure a reader cannot check against anything on their own screen, while the same range on that page of 24 July 2026 expressed per thousand views is $15 to $40 — a pair of numbers sitting in the same column of the same report the reader already has open. Nothing was altered in that conversion except the number of zeroes.

The two rows the folklore is built from sit on that same page of 24 July 2026. Finance & Investing at $15,000 to $40,000+ per million views, set against Gaming at $2,000 to $8,000 per million views, is a ratio of five at one end of those two ranges and seven and a half at the other, on that page's own figures as it read on 24 July 2026. The sentence in circulation says ten. It is not the arithmetic of the table it is most often attributed to, and neither the table nor the sentence carries a channel count.

The second table is the one most often re-quoted with its caveats stripped. vidIQ's guide "Highest-Paying YouTube Niches in 2026: CPM & RPM by Category" was published 20 April 2026, reviewed 29 May 2026, and bylined Darryl Rentz. Its methodology note reads: "CPM and RPM ranges in this guide are drawn from Google Ads benchmark data, public creator earnings disclosures, and vidIQ aggregate channel data, reflecting the period through Q1 2026. Rates move with advertiser demand and seasonality, so treat them as directional ranges rather than guarantees for any single channel."

That paragraph names three inputs and separates none of them. Google Ads benchmark data is advertiser-side, collected on the buying side of an auction, and describes what buyers bid rather than what sellers received. Public creator earnings disclosures are self-reports, and the page neither counts them nor dates them. Aggregate channel data is the one input that could be a measurement, and for it the page gives no channel count, no country mix and no window narrower than "the period through Q1 2026." A blend of three sources with no weights published is not three sources; it is one number whose provenance cannot be reconstructed by anybody outside the company that made it.

Its own figures, as that page published them on 20 April 2026 and reviewed them on 29 May 2026. The left-hand band is advertiser-side and is not creator pay, for reasons this chapter gives in full below.

Six of those nine rows carry an advertiser-side band and no creator-side one, on a page whose title promises both. The column that describes what a channel is paid is the column mostly left blank.

The body table on that same page of 20 April 2026 gives $5 to $17 for the same category. Both figures are published, on one address, separated by the length of an article, and an archived copy taken 6 August 2026 carries both of them. At the top of the range they differ by about 47%, which is this book's arithmetic on the page's own two numbers, and the top of the range is the end a reader plans with.

A page that cannot hold one number consistent from its table to its own FAQ is not a measurement. This is the cheapest of the objections in this chapter and the strongest, because it needs no counter-study and no second opinion: it needs the same page read twice, which takes about twelve minutes.

The rule that follows from that table is a standing one, and it governs every figure printed for the rest of this book. Never read a CPM table as creator pay.

YouTube's ad-revenue analytics help page, retrieved 5 September 2026, defines CPM as the cost per 1,000 ad impressions before YouTube's revenue share, counting ads and Premium only, and RPM as what the creator earned per 1,000 views after that share, counting memberships, Premium, Super Chat and Super Stickers alongside ads. On that same page retrieved 5 September 2026, Google publishes no conversion between the two.

Three things move between a CPM and an RPM and none of them is published as a coefficient. The platform's share comes off. The denominator changes from ad impressions to views, and most views never serve an ad. And RPM's numerator includes money that CPM excludes altogether.

The denominator swap does most of the damage, and the one 2026 dataset built from Studio revenue lines shows it in a single row. In AIR Media-Tech's RPM study of 300 channels and 3,595 channel-months, last updated 1 July 2026, Gadgets & Tech posted a median Studio CPM of $5.73 against a median RPM of $2.33, because the monetized-playback rate for that category across those 300 channels was 34%. A category can post a higher CPM and a lower RPM than the category beside it on that number alone, and a table that prints only the CPM column has printed the half of the chain that nobody is paid.

The measuring vendor is blunt about the gap between its own dashboard figures and the tables above. In that same study of 300 channels and 3,595 channel-months updated 1 July 2026, AIR Media-Tech states that advertiser-side estimates run roughly 1.5 to 2 times higher than Studio CPM; third-party advertiser-side estimates for Gadgets & Tech in that study ran $9 to $11.50 against the $5.73 Studio median from those same 300 channels. The chain runs downhill twice before it reaches anybody's bank, and neither reduction is published as a rate.

The ten-times comparison is much easier to find on the advertiser side of that chain, which is probably where it came from. On vidIQ's page of 20 April 2026, Personal Finance & Investing carries a CPM band of $15 to $50 while gaming and entertainment together carry "$1–8 CPM" — a ratio running from about fifteen to about fifty depending on which ends of the two bands are picked. Both of those are estimates of what advertisers bid. Neither is a payment to anybody. A ratio taken between two advertiser-side bands describes a difference in what buyers were willing to spend, and it reaches a creator's account only to the extent that the two categories serve ads at the same rate and carry the same ad load — a pair of assumptions no CPM table states and none of the four tables in this chapter tests.

The third table takes the confidence further than either of the first two. OutlierKit's most-profitable-niches page carries the words "verified June 2026" and no sample size anywhere on it. Its figures, as that page carried them under that June 2026 mark:

The cent is the tell. An RPM carried to two decimal places is a figure computed from a specific set of channels over a specific set of months, and that page of June 2026 names neither the channels nor how many of them there were nor which months. A median taken across three channels and a median taken across three thousand both print as $12.82 on a page marked June 2026.

The word "verified" is doing the same work the decimal places are doing. It appears with no object: verified by whom, against what record, and on whose revenue lines. A page can be verified against its own previous draft.

The granularity is the second thing worth noticing. Six sub-niches named that precisely, each with a two-decimal RPM, is either a very large dataset sliced very fine or a very small one described very confidently, and the page of June 2026 does not say which. The tables in this chapter get more specific as they get less sourced, which is the opposite of the direction sampling error runs.

Against those three sits the fourth, and it is the only one built from Studio revenue lines rather than from advertiser benchmarks. AIR Media-Tech's RPM study, last updated 1 July 2026, reads 300 channels on one multi-channel network's own roster across 3,595 channel-months from May 2025 to May 2026, in thirteen niches, at channel sizes running from 10,000 to 50 million subscribers. Every figure it publishes is a median.

One caveat travels with that dataset every time it is used, and it belongs to the vendor rather than to this book. The same company published a second 2026 study eight days earlier — 274 channels, 3,044 channel-months, dated 23 June 2026 — and the two disagree on the same metric in the same category. The study of 23 June 2026 puts Kids long-form monetized playback at 12.1%; the study updated 1 July 2026 puts Kids at 32%, a factor of 2.6 between one measurer's own two numbers. The same thing happens on RPM in Education: a median of $10.22 in the 300-channel study of 1 July 2026, against $14.97 for medium channels and $18.23 for small ones in the 274-channel study of 23 June

1. Different samples explain part of that and not the size of it.

The medians, as that study of 300 channels and 3,595 channel-months last updated 1 July 2026 published them, with the four niches its own authors flagged as resting on fewer than ten channels marked in the right-hand column:

Business & Finance sits eleventh of thirteen at $2.01 in that 300-channel study of 1 July 2026, on three channels. Gaming sits above it at $2.05 in that same study of 1 July 2026, across 300 channels and 3,595 channel-months, on a sample its authors did not flag. The all-niche median across all 300 channels in that study of 1 July 2026 was about $2.30.

Three channels is not a measurement of a category. It is three channels, and the study's own authors say so by flagging the row. This book does not use that row to refute the finance tables and no reader should: a median of three is a statement about three channels that happened to be on one network's roster inside one twelve-month window ending in May 2026, and the network's roster is heavily non-US in origin, which is a separate reason its finance row cannot stand in for a US finance channel.

Four pages, then, and what each of them discloses about how it arrived at its numbers, as all four read on 5 September 2026:

Three of the four disclose nothing about who was counted. The fourth discloses enough to be argued with, which is the only reason this chapter has been able to argue with it, and the same disclosure is what makes its three-channel finance row visible rather than buried inside a range. A table that publishes its weaknesses is not thereby a worse table than one that publishes none. It is the only one of the four that can be checked at all.

Two of the three unsourced tables agree with each other and disagree with the measured one, and that agreement is worth less than it looks. Influencer Marketing Hub on 24 July 2026 and vidIQ on 20 April 2026 both put finance and insurance at the top of their rankings and gaming and entertainment at the bottom. Both name creator-reported material among their inputs, and neither names a channel count anywhere. Two pages drawing on the same pool of self-reports are not independent of one another, and two rankings arriving at the same order from the same inputs are one measurement repeated rather than two measurements agreeing.

There is a fifth source in this ecosystem, and it publishes no table at all. Social Blade and the public-API estimators like it build earnings ranges out of subscriber and view time series against assumed CPM bands, and they are the engine underneath most of the how-much-does-this-channel-earn articles in circulation. No figure from any of them appears in this book, for one structural reason: a public-API estimator cannot see a channel's monetized-playback rate, its ad load, its audience geography or its revenue share in 2026, which are the four inputs that decide an RPM. It is estimating the output of a function from none of the function's arguments.

Which leaves the sentence a reader arrives holding. Finance pays ten times what gaming pays for the same million views: the folklore tables of 2026 are consistent with it, the one measured table of 2026 points the other way on three channels, and neither of those is evidence for anything. The word for the claim is unsupported.

Unsupported is not a synonym for false, and it is not a polite way of saying false. It is a statement about the state of the record. As of September 2026, no public dataset had measured enough finance channels and enough gaming channels off Studio revenue lines, inside one sample, to say which of them pays more. The claim may be right. Nobody has checked it, and the pages that assert it are the pages that disclose the least about how they know.

None of the four pages names a channel. The three unsourced ones name none because they have none to name. The measured one reports medians and names no channel, winner or loser, which is a choice its authors were entitled to make and which also means it publishes no case in which a category paid what the table said and no case in which it did not. No publisher of any of these four has printed a channel that picked a category off its rate card and then earned outside the range that rate card gave it. As of September 2026, no failure data was published beside any of them.

So the chapter ends without a rate card, and that is the honest end of it. What is missing is a second measurement, and its shape is specific enough to describe: a sample with more than twenty channels in every niche, weighted toward US audiences rather than toward one network's international roster, read off Studio revenue lines rather than advertiser benchmarks, and published with its sampling frame printed beside its medians. Failing that, a cohort of creators publishing their own twelve-month revenue exports with the month, the category and the audience geography attached — including the channels that stopped uploading, because a table assembled only from the channels that lasted runs the same survivorship filter the vendor pages already run. As of September 2026, this was unresolved. Neither dataset existed, no regulator had required one, and the four pages read in this chapter are what a reader choosing a category in 2026 has instead.