CPM is not a rate you negotiate. On the major platforms it is the output of an auction, which means you observe it rather than set it.

The figure divides spend by impressions delivered and multiplies by a thousand. It prices exposure. It says nothing about who saw the ad, in what conditions, or what happened next.

This page covers what CPM measures, what it ignores, why the published US benchmarks disagree by a factor of three, and why a low CPM is frequently bad news in B2B.

CPM stopped being a rate card

The word survives from a market that priced space in advance. That market still exists in direct publisher deals. It is not the one most advertisers are in.

What replaced it. On the auction-based platforms, the winning ad is not the highest bid. The auction combines the bid with estimated action rates, meaning the platform’s prediction of what the person will do, and with ad quality. A more relevant ad can beat a higher bid.

What that does to your CPM. It becomes a consequence of decisions you did not all make: how many advertisers want the same audience today, how your creative is being received, and how narrow you made the targeting.

The practical implication for management. You cannot set a CPM target and hit it the way you hit a budget. You can influence it, through audience width and creative quality, and you can explain it after the fact. That is a different exercise from steering it.

Why this matters before any benchmark discussion. If CPM is an auction output, then a “market average CPM” is an average of outputs from auctions you did not take part in, on audiences that are not yours. It is descriptive of a market, never prescriptive for an account.

What CPM ignores, and it is most of what matters

The metric is honest about its own scope. The trouble starts when people ask it questions it was never built to answer.

It ignores who saw the ad. A thousand impressions on people who could never buy from you and a thousand impressions on your exact buyer profile carry the same CPM. In B2B, where the addressable market is a few hundred to a few thousand companies, that difference is the entire game.

It ignores how the ad was seen. A viewable impression has an accredited standard: for display, at least 50% of the ad’s pixels in the foreground for at least one continuous second, and for video, 50% of pixels with two continuous seconds of play. Your CPM does not distinguish an impression that met that bar from one that scrolled past.

It ignores what happened next. No click, no enquiry, no revenue enters the calculation. CPM prices the exposure and stops there, which is exactly what it says on the label.

Why this is not a criticism of the metric. A cost of exposure is a legitimate thing to know, especially when you are buying reach on purpose. The error is using it as a performance indicator, because it contains no information about performance by construction.

What cost per thousand impressions measures and what it ignoresDiagram distinguishing what the cost per thousand impressions metric measures from what it ignores by construction. It measures the price of exposure, obtained by dividing spend by impressions delivered and multiplying by a thousand. It ignores three things. First, who saw the advertisement, since a thousand impressions on people who could never buy and a thousand impressions on the exact buyer profile carry the same figure, a difference that is decisive in business-to-business advertising where the addressable market runs from a few hundred to a few thousand companies. Second, how the advertisement was seen, since a viewable impression carries an accredited standard requiring at least fifty percent of the advertisement’s pixels in the foreground for at least one continuous second for display, and fifty percent of pixels with two continuous seconds of play for video, whereas the cost per thousand does not distinguish an impression meeting that bar from one that scrolled past. Third, what happened next, since no click, enquiry or revenue enters the calculation. This is not a criticism of the metric, because a cost of exposure is a legitimate thing to know, particularly when buying reach deliberately. The error consists in using it as a performance indicator, since by construction it contains no information about performance.What it prices, and what it cannot seeIt measures one thing: the price of exposure.Spend ÷ impressions delivered × 1,000. That is the whole claim.It ignores who saw itA thousand impressions onpeople who could never buyand a thousand on your exactbuyer carry the same CPM.It ignores how it was seenA viewable impression has astandard: 50% of pixels forone second. Your CPM doesnot check it.It ignores what followedNo click, no enquiry, norevenue enters thecalculation. It stops at theimpression.None of this is a flaw. It is the label.The error is asking a cost of exposure to report on performance.
A cost of exposure is a legitimate thing to know. The error is using it as a performance indicator, because it contains no performance information. Source : MASTRATOS (2026)

Why the published US benchmarks disagree by a factor of three

They do not disagree by accident. They cover different things, and they declare different amounts about themselves.

The most detailed public tracker. Common Thread Collective publishes a column labelled “Portfolio Avg CPM” for Meta: $9.84 in January 2024, $13.00 in January 2025, $15.54 in January 2026, $18.01 in March 2026 and $17.01 in April 2026, with a November 2025 peak at $21.55.

What it declares about itself, in full. “Tracking monthly Meta CPMs across a portfolio spanning hundreds of DTC brands and hundreds of millions in annual ad spend.” That is the whole methodology. No count of brands, no statement of whether these are means or medians, no declared geography, and a direct-to-consumer perimeter that has nothing to do with a B2B account.

What it is still good for. The trend. A portfolio tracked consistently over two years showing CPMs rising from under ten dollars to the high teens tells you something real about the direction of Meta inventory pricing, even if the level does not transfer to you.

The measurement house with the best method does not publish the numbers. The most methodologically serious source in this space keeps its values inside its product, and its public methodology page returns an access refusal to automated readers. You cannot cite what you cannot open.

Two more that do not exist any more. One benchmarking product has been discontinued and its old URLs now serve a generic site shell that answers 200 to any address, which is a good way to believe you have verified something you have not. Another well-known figure is published behind a subscription and is labelled a forecast, meaning a model output rather than a measurement.

What to conclude before quoting any of them. No publisher combines a US CPM figure with a complete methodology. That is not a gap in your research. It is the state of the subject, and saying so is more useful than picking whichever number suits the argument.

What publishers of United States cost per thousand impressions figures declare about their methodTable setting out what each publisher of a United States cost per thousand impressions figure actually declares about its own method. The most detailed public tracker publishes a column labelled portfolio average cost per thousand impressions for Meta, giving nine dollars eighty-four in January two thousand and twenty-four, thirteen dollars in January two thousand and twenty-five, fifteen dollars fifty-four in January two thousand and twenty-six and eighteen dollars one cent in March two thousand and twenty-six, but its entire declared methodology is a single sentence stating that it tracks monthly Meta figures across a portfolio spanning hundreds of direct-to-consumer brands and hundreds of millions in annual advertising spend, with no count of brands, no statement of whether the figures are means or medians and no declared geography, on a direct-to-consumer perimeter unrelated to business-to-business advertising. The measurement house with the most rigorous published method keeps its values inside its own product and its public methodology page returns an access refusal to automated readers, so it cannot be cited. One benchmarking product has been discontinued and its former addresses now serve a generic site shell that answers successfully to any address, which makes it easy to believe a figure has been verified when it has not. Another widely quoted figure sits behind a subscription and is labelled a forecast, meaning a model output rather than a measurement. The conclusion is that no publisher combines a United States figure with a complete methodology.What each publisher declares about itselfSOURCEWHAT IT PUBLISHESWHAT IT DECLARESPortfolio trackerMeta: $9.84 to $18.01One sentence. DTC brands.No count, no mean/median, no geographyThe rigorous oneNothing publicValues sit inside the product.Method page refuses automated readersThe discontinued oneOld URLs still resolveServes a generic site shell.Easy to think you verified somethingThe subscription oneValues behind a paywallLabelled a forecast.A model output, not a measurementNobody combines a US CPM figure with a complete methodology.
No publisher combines a US CPM figure with a complete methodology. That is the state of the subject, not a gap in your research. Source : MASTRATOS from Common Thread Collective and public sources (2026)

What the industry body publishes is revenue, not price

There is one large, methodologically declared US dataset on digital advertising. It measures something else, and knowing that saves a lot of misquoting.

The figure. The IAB and PwC report US internet advertising revenue of $294.6 billion in 2025, up 13.9% year over year, in the thirtieth edition of a report running since 1996.

The breakdown. Search $114.2 billion, display $81.6 billion, digital video $78.0 billion, digital audio $8.4 billion. Two cross-cutting figures that do not add to the others: social at $117.7 billion, and programmatic at $162.4 billion, up 20.5%.

What it does not contain. No unit price. Not a CPM, not a CPC. The report measures revenue booked by companies selling advertising, not what an advertiser pays per thousand impressions. Anyone citing it for a price is citing it for something it does not say.

What its own methodology admits. Data is “reported directly to PwC from companies selling advertising on the internet as well as publicly available corporate data”, non-participating companies get “a conservative revenue estimate based on available public sources”, and the report states plainly that “PwC does not audit the information and provides no opinion or other form of assurance.” The number of responding companies is not disclosed.

The line that matters most for a small advertiser, and it is rarely quoted. The top ten companies hold 84.1% of that revenue, up 3.4 points since 2024. A market average therefore describes those ten companies far more than it describes the auction conditions your account meets.

United States internet advertising revenue for 2025 and its concentrationChart presenting the United States internet advertising revenue reported for the year two thousand and twenty-five by the industry body and its auditing partner, together with the concentration of that revenue. Total revenue reached two hundred and ninety-four point six billion dollars, an increase of thirteen point nine percent year over year, in the thirtieth edition of a report running since nineteen ninety-six. The breakdown gives search at one hundred and fourteen point two billion dollars, display at eighty-one point six billion, digital video at seventy-eight billion and digital audio at eight point four billion, alongside two cross-cutting figures that do not add to the others, social at one hundred and seventeen point seven billion and programmatic at one hundred and sixty-two point four billion with twenty point five percent growth. The report contains no unit price of any kind, neither a cost per thousand impressions nor a cost per click, because it measures revenue booked by companies selling advertising rather than what an advertiser pays. Its own methodology states that data is reported directly by companies selling advertising alongside publicly available corporate data, that non-participating companies receive a conservative revenue estimate based on available public sources, and that the auditing firm does not audit the information and provides no opinion or other form of assurance, the number of responding companies not being disclosed. The line most relevant to a small advertiser is that the top ten companies hold eighty-four point one percent of that revenue, up three point four points since two thousand and twenty-four.US internet advertising revenue, 2025$294.6B+13.9%Thirtieth edition. No unit price anywhere in it.By formatSearch $114.2BDisplay $81.6BDigital video $78.0BDigital audio $8.4BThe top ten companies hold 84.1% of it, up 3.4 points since 2024.A market average describes those ten far more than it describes your account’s auction conditions.
$294.6 billion in 2025, up 13.9%. No unit price anywhere in the report, and a market average describes ten companies more than it describes your account. Source : IAB and PwC (2026)

In B2B, a low CPM is usually the expensive option

This is the point that costs the most money, and it runs against the intuition.

Why broadening lowers CPM mechanically. Wide audiences sit on cheaper inventory and face less competition per impression. Loosen your targeting and your CPM falls without you having done anything skilful.

What falls with it. The share of impressions reaching someone who could ever buy from you. On a market of a few hundred addressable companies, that share can drop from meaningful to negligible while your CPM improves.

The arithmetic that settles it. Take a thousand impressions at a $20 CPM where one in ten viewers is in your target: you paid $20 to reach a hundred plausible people, so twenty cents each. Now take a thousand impressions at a $5 CPM where one in two hundred is in your target: you paid $5 to reach five plausible people, so a dollar each. The cheaper CPM costs five times more per useful person.

Why nobody notices. Because the CPM is on the dashboard and the share of useful impressions is not. Nothing in the reporting shows the second number, so the first one wins the argument by default.

The question that exposes it. Our CPM fell: what changed in the audience? If nobody can answer, the fall is not good news, it is an unknown.

What actually moves your CPM

Three things do most of the work, and bid settings are not among them.

Competition for your audience. More advertisers wanting the same people raises the clearing price, and it moves with the calendar. Retail peaks and large sporting events lift CPMs for everyone, including advertisers who have nothing to do with either.

How your ads are received. Auctions weight predicted engagement and ad quality, so a creative that people respond to wins impressions at a lower price than a creative that people ignore. This is the lever you actually control, and it is why creative fatigue shows up as a rising CPM before it shows up anywhere else.

How narrow your targeting is. Narrow audiences cost more per thousand and are usually worth it in B2B. That is a trade you should be making deliberately rather than discovering in a report.

What barely moves it. Bid strategy tinkering, budget pacing changes, and most of the settings people reach for first. If your CPM rose and none of the three above changed, the most likely explanation is that the market moved, not that someone made a mistake.

Why a lower cost per thousand impressions can cost more per useful personWorked comparison showing why a lower cost per thousand impressions frequently costs more per person who could actually buy, which is the most expensive misunderstanding in business-to-business advertising. In the first case, one thousand impressions are bought at a cost per thousand of twenty dollars on a narrow audience in which one viewer in ten belongs to the target, so twenty dollars buys a hundred plausible people, which works out at twenty cents per useful person. In the second case, one thousand impressions are bought at a cost per thousand of five dollars on a broad audience in which one viewer in two hundred belongs to the target, so five dollars buys five plausible people, which works out at one dollar per useful person. The cheaper cost per thousand therefore costs five times more per useful person. Broadening an audience lowers the cost per thousand mechanically because wide audiences sit on cheaper inventory and face less competition per impression, while the share of impressions reaching someone who could ever buy collapses, and on a market of a few hundred addressable companies that share can fall from meaningful to negligible. Nobody notices because the cost per thousand appears on the dashboard and the share of useful impressions does not, so the first number wins the argument by default. The question that exposes the problem is what changed in the audience when the cost per thousand fell.The cheaper CPM, and what it really costsNarrow audience1,000 impressions at a $20 CPM1 viewer in 10 is in your target$20 reaches 100 plausible people$0.20per useful personBroad audience1,000 impressions at a $5 CPM1 viewer in 200 is in your target$5 reaches 5 plausible people$1.00per useful personThe cheaper CPM costs five times more per person who could buy.Ask the question that exposes it: our CPM fell, what changed in the audience?
The share of useful impressions is not on the dashboard, so the CPM wins the argument by default. Source : MASTRATOS (2026)

What to track instead, and what CPM is still good for

It is not a bad metric. It is a metric with a narrow job, and the job is worth doing.

What it is genuinely good for. Explaining a cost movement. When your cost per enquiry rises, the first question is whether you are paying more for exposure or converting less of it. CPM answers the first half in one glance.

What it is not good for. Judging a campaign, comparing two channels, or assessing a vendor. All three require knowing what the exposure produced.

The pair to watch. CPM and cost per qualified enquiry, over the same period. A CPM rising while cost per enquiry stays flat means the market got more expensive and your creative absorbed it. Both rising means the creative is not absorbing it any more.

The log to keep. Record every change of audience, creative and bidding setup, with the date. Without it, a CPM movement is never attributable to a cause and you will be explaining noise for months.

The month-over-month comparison to avoid. Compare to the same month last year, not to last month. Advertising inventory is seasonal, and a December-to-January comparison mostly measures the calendar.

What moves a cost per thousand impressions and what does notDiagram distinguishing the factors that genuinely move a cost per thousand impressions from those that barely affect it. Three factors do most of the work. Competition for the audience raises the clearing price when more advertisers want the same people, and it moves with the calendar, retail peaks and large sporting events lifting the figure for every advertiser including those with no connection to either. How the advertisements are received matters because auctions weight predicted engagement and advertisement quality, so a creative that people respond to wins impressions at a lower price than a creative people ignore, which is the lever the advertiser actually controls and the reason creative fatigue appears as a rising cost per thousand before it appears anywhere else. How narrow the targeting is matters because narrow audiences cost more per thousand and are usually worth it in business-to-business advertising, a trade that should be made deliberately rather than discovered in a report. By contrast, bid strategy tinkering, budget pacing changes and most of the settings people reach for first barely move the figure, so where a cost per thousand has risen and none of the three genuine factors changed, the most likely explanation is that the market moved rather than that somebody made a mistake.What moves it, and what does notCompetition for your audienceMoves it a lotRetail peaks and large sporting events lift CPMs for everyone, including you.How your ads are receivedMoves it a lotThe one lever you control. Creative fatigue shows up here before anywhere else.How narrow your targeting isMoves it a lotNarrow costs more per thousand, and is usually worth it in B2B. Decide it, don’t discover it.Bid tinkering, pacing changes, most first-reach settingsBarely moves itIf none of the three above changed, the market moved. Nobody made a mistake.
Bid settings are the first thing people reach for and close to the last thing that matters. Creative quality is the lever you actually control. Source : MASTRATOS (2026)

The seasonality nobody budgets for

CPM is one of the few advertising numbers with a genuinely predictable annual shape, and most plans ignore it.

Why inventory prices move on a calendar. The supply of impressions is roughly stable while demand is not. When large advertisers concentrate spending into a few weeks, everyone bidding on the same inventory pays more, including advertisers selling something completely unrelated.

The predictable peaks. The retail run from late November through December is the largest, and major sporting events pull whole quarters upward in the markets they touch. A B2B advertiser has no commercial reason to be in those auctions and pays their price anyway.

What that means for a January review. A CPM that fell sharply between December and January is not an achievement. It is the calendar releasing pressure, and reading it as an improvement leads people to credit whatever change they happened to make in early January.

The comparison that survives it. Same month, previous year. It is the only one that holds the seasonal position constant, and it is the reason a CPM series is worth keeping for two years before it becomes genuinely useful.

What to do rather than react. Plan the annual envelope knowing the expensive weeks exist, and decide in advance whether you buy through them or step back. Cutting a campaign entirely during a peak costs history and algorithmic stability, which you pay for on restart. Reducing without stopping is usually the better trade.

How to track a CPM so that it says something

Four columns are enough, and most reports have three of them wrong.

The CPM itself, per platform and per campaign type. Aggregating a search campaign, a video campaign and a retargeting campaign into one figure produces a number that moves for reasons nobody can name. Split it before you plot it.

The audience width behind each line. Not a precise reach figure, which nobody has, but a stated intent: broad prospecting, narrowed by firmographics, retargeting. A CPM only becomes readable next to the width that produced it.

A dated change log. Every audience change, every new creative batch, every bidding change, with the date. This is the column people skip, and without it every explanation offered in a review is a reconstruction after the fact.

The cost per qualified enquiry, on the same rows. This is what turns the CPM from a curiosity into a diagnostic. Read together, the two say whether a cost movement came from the market or from your conversion.

The reading cadence. Monthly, compared with the same month a year earlier. Weekly readings on a small account measure noise, and month-over-month readings mostly measure the advertising calendar.

What this setup lets you say in a meeting. Not “our CPM is good” or “our CPM is bad”, which are not statements about anything. But “our CPM rose 30% year over year, our audience width did not change, and our cost per enquiry held, so the market got more expensive and the creative absorbed it.” That is a sentence someone can act on.

Questions to ask whoever presents you a CPM

Our CPM fell: what changed in the audience? This is the one that separates an improvement from a broadening. If nobody can answer, the fall is an unknown rather than good news.

What share of these impressions reached our target? Rarely reportable exactly, but the attempt reveals whether anyone has thought about it.

What does a qualified enquiry cost over the same period? The only question whose answer supports a decision.

What is the source of the benchmark you are showing me? If the answer is a portfolio of direct-to-consumer brands with no declared sample, that is worth knowing before anyone treats it as a target.

The signal that should worry you. A monthly report leading with CPM on the front page, before the number of enquiries received. The order of numbers in a report says what the vendor considers the result, and that is rarely an accident.

Where to go next

You want the cost of a click rather than an impression. Click-through rate benchmarks.

You want the cost of a contact. Cost per lead, and how to calculate CAC.

Your creative has been running a long time and costs are drifting. Creative fatigue metrics.

You want the real prices on Meta. How much do Facebook ads cost.

You are splitting a budget between platforms. How to split budget between Google and Meta in B2B.

You want to know which metric to run on. ROAS vs MER vs CAC vs LTV.

In short

  • CPM is no longer a rate card, it is an auction output combining bid, estimated action rates and ad quality. You observe it, you do not set it.
  • It ignores who saw the ad, how, and what followed. It prices exposure, which is exactly what it claims to do.
  • No publisher combines a US CPM with a complete methodology. The most detailed tracker declares one sentence and a direct-to-consumer perimeter; the most rigorous house does not publish its values.
  • The industry body publishes revenue, not prices: $294.6 billion in 2025, up 13.9%, on a basis its own methodology says is not audited.
  • The top ten companies hold 84.1% of that revenue. A market average describes them, not your account.
  • A low CPM can cost five times more per useful person, and in B2B that is the common case rather than the exception.
  • What moves it: competition for your audience, how your ads are received, how narrow your targeting is. Not your bid settings.
  • Compare to the same month last year, not to last month, because inventory is seasonal.

A CPM is never judged alone. Book a diagnostic, or see how we approach B2B paid acquisition.