Of every $1,000 entering a demand-side platform, $150 buys impressions whose visibility was never determined. It appears on no report, and it beats the fees.
Of every $1,000 that enters a demand-side platform, $150 buys impressions whose visibility was never determined. Not impressions that went unseen: impressions where nobody could establish whether they were seen at all. It is the single largest leak in the programmatic chain, ahead of demand-side platform fees, ahead of supply-side platform fees, ahead of fraud by a factor of thirty.
It appears on no campaign report, because the metric that would expose it is almost never displayed next to the one everybody looks at.
Two rates, two denominators, one of them displayed
The standard sorts impressions into three categories it calls mutually exclusive: viewable, non-viewable, and viewable status undetermined. That third category is not a quality judgment. It is the absence of a verdict:
“Undetermined Ad Impression: These represent served impressions where the viewable status cannot be determined because of any conditions that do not allow that decision to be made. For example: (1) if cross-domain I-Frames block the viewable determination, (2) clients not supporting Java script, and/or (3) any other issues where browser settings or serving conditions disallow verification of the ad position on the page or the time-viewed.”
From this follow two rates:
The measured rate divides measured impressions by total served.
The viewable rate divides viewable impressions by measured ones only.
The standard supplies its own worked example, and it is instructive. One thousand impressions: 300 viewable, 200 non-viewable, 500 undetermined.
The measured rate is 50%. The displayed viewable rate is 60%, because it is computed across the 500 impressions anyone managed to observe. The share genuinely viewable across the whole campaign is 30%.
The document then forbids calling that 30% the viewable rate:
“Viewable Ad Impression measurers should not refer to the 30% in the Impression Distribution category as the ‘Viewable Rate’ since it assumes that all undetermined were non-viewable, thought to be an overly conservative presentation.”
The reasoning is defensible: nobody knows whether the undetermined impressions were viewable, so counting them as failures would be unfair. The practical result is that a report can accurately claim 60% viewability on a campaign where the fate of half the impressions is unknown.
On real data the gap is comparable. A study covering 21 advertisers and 35.5 billion impressions found median measurability of 86% and viewability of 85% among measured impressions. Multiplying gives the only honest figure:
“Given that 86 percent of total impressions were measurable and 85 percent of measurable impressions were viewable, only 73 percent of total impressions were both measurable and viewable.”
With enormous dispersion between advertisers: measurability ran from 0.02% to 99.61%, and four participants sat below 20%.
Cross-domain iframes. An ad served inside an iframe whose domain differs from the host page cannot, for browser security reasons, query the geometry of the parent document. The measurement script is there and it runs, but it cannot establish where the ad sits on screen. The standard devotes a section to this and creates a dedicated metric, the see-through rate, to quantify how far a measurer can see into nested frames. An advisory published in 2012, based on a pilot covering 22 campaigns and more than three billion served impressions, already put a number on it: cross-domain iframes accounted for three quarters of unmeasured impressions in network placements, and measured rates in that pilot ran at 27% for publisher placements and 19% across ad networks.
Environments without scripting. Some video players and platforms do not execute JavaScript. The study adds a fourth reality to the technical list: placements where the seller has simply rejected the verification tags.
Apps and connected television. This is the most radical case, because the definition of viewability loses its meaning entirely: “50% of pixels in the viewable space of the browser tab” says nothing about a television set. The standard concedes the point:
“OTT often cannot be measured in the same way as traditional digital video delivery due to limitations on JavaScript and SDK in OTT environments, as well as additional challenges such as TV Off detection.”
What replaces pixel measurement there are state checks: detecting that the screen is off while the device stays active, detecting continuous play with nobody in the room. A standardized measurement SDK exists for apps, but its announced connected-TV coverage was still 40% of the market in May 2024, after extension to two major manufacturers.
The $150 deserves breaking down, because how it forms explains why it goes unnoticed.
The non-measurable line represents 14% of impressions but 21.5% of spend. The study gives the reason:
“Video as a format over-indexes for non-measurable inventory, and also has higher CPMs resulting in the delta we see here of 14 percent of impressions but 21.5 percent of spend.”
The problem concentrates exactly where the thousand costs most. You lose proportionally more money than impressions, which makes the line invisible to anyone reasoning in volume.
Against the other lines in the same waterfall, on $1,000 entering the demand-side platform: DSP platform fees $80, DSP additional fees $20, data fees $60, SSP platform fees $130. Then, against seller revenue: non-viewable $95, invalid traffic $5, non-measurable $150, made-for-advertising inventory $100.
The non-measurable line is the heaviest in the entire chain, ahead of the largest platform fee. And unlike the fees, it appears on no invoice.
There is a question nobody asks: does an impression that is “viewable” under the standard, meaning 50% of pixels for one second, produce anything at all?
Where the threshold came from. Not from a study of perception. From a negotiation among three trade associations in 2012, validated by a pilot covering 22 campaigns and over three billion served impressions. The standards body’s own executive has explained the reasoning publicly: the threshold was adopted because 80% of display ads meeting it would also have qualified under the stricter criterion of 100% of pixels in view for a second. That is a calibration between two measurement methods, not a duration validated as sufficient to produce an effect. The stated motivation on the trade side was equally practical: create a currency “that can get brands excited about investing in digital”.
What the research says. The only published causal experiment on exposure duration and memory, in the proceedings of an ACM conference, tests no duration shorter than five seconds. The regulatory threshold of one second therefore sits entirely below the range anyone has examined. The authors report that recall and recognition “increases sharply in the first 10 seconds”.
And the two studies that tested the threshold disagree. A randomized experiment, which remains a working paper and to my knowledge was never published in a peer-reviewed journal, concludes that “everything below 75% or 5 seconds seems to perform a lot worse”, and that “advertisers will have to understand that the MRC standard is not enough from an advertising effectiveness stance”. Its authors note that their own observational analysis of the same data suggested the opposite, and attribute the discrepancy to endogeneity bias. A 2025 article in the Journal of Advertising reaches the contrary conclusion, finding concurrent validity between the one-second threshold and attention measured by mobile eye-tracking, plus predictive validity for brand recall.
What a 2025 causal study in the Journal of Marketing Research does establish, on 1,013 participants with webcam eye-tracking: one additional second of actual attention raises the probability of recall by 3.4 percentage points and the probability of choosing the brand by 0.7 points. Note the difference: that is seconds of real gaze, not the crossing of a declarative threshold.
There is a broader reason to hold all of this loosely. Twenty-five large-scale field experiments published in the Quarterly Journal of Economics found that “the median confidence interval on return on investment is over 100 percentage points wide”, and that detecting a 10% difference in advertising ROI would require campaigns roughly sixty times larger than those typically run. The authors are blunt about why observational methods fail here: “selection bias, due to the targeted nature of advertising, is a crippling concern for widely employed observational methods.” A separate field experiment published in Econometrica measured the gap directly: standard regression put the return on paid search at over 4,100%, while the experimental estimate was -63%.
Fourteen years into using viewability as a transactional currency, there is no scientific consensus on whether it predicts anything.
There is a way to settle the debate without taking sides: look at what the bodies that created the measure are now doing.
In November 2025, the IAB and the MRC jointly published an entirely new framework, devoted to measuring attention. Building a further measurement layer fourteen years after making viewability a trading currency is itself an answer. The document carries a caution the viewability standard never had:
“Attention is not intended to serve as a standalone measure of ad effectiveness of business results.”
And it explicitly files viewability under delivery verification, alongside impression counting, rather than under quality or effect measurement.
That classification is not new. It was implicit in the original definition of the viewable impression:
“It is recognized that an ‘opportunity to see’ the ad exists with a viewable ad impression, which may or may not be the case with a served ad impression.”
An opportunity to be seen. The body that built the metric never claimed it established that anyone had seen anything.
Ask for the measured rate alongside the viewable rate. This is the one action here that costs nothing and changes everything. A viewability figure without its measurability figure is uninterpretable, and the standards body publishes both formulas precisely so they are read together.
Multiply the two. 86% measurability and 85% viewability do not make 85%. They make 73%. That is the number describing your campaign.
Treat video separately. That is where the non-measurable concentrates, and where the thousand costs most. A report aggregating display and video hides the problem by construction.
Do not expect viewable CPM to fix it. Google’s documentation is explicit that in that mode “a small portion of served impressions might be charged”.
Select for publishers that accept verification tags. The study’s recommendation is to prioritize them, aiming for 100% of impressions to be measurable, and to pay only for measurable impressions. That is an inventory selection criterion, not a technical setting, and it is negotiable. It is one of the constraints we buy under, which is why our B2B paid acquisition work is judged on cost per lead rather than on thousands of impressions delivered.
One closing caution, less advice than a reading instruction. Even at 100% measurability and 100% viewability, you will not have shown that anyone looked at your ad. You will have shown it was delivered under conditions where looking was possible. That is useful, it is necessary, and it is not the same thing.
Frequently asked questions
What is the difference between measurability and viewability?
They have different denominators. The measured rate divides measured impressions by total served. The viewable rate divides viewable impressions by measured ones only. A 60% viewable rate can therefore describe half your impressions and say nothing about the other half.
Why can an impression not be measured?
Three documented causes: cross-domain iframes that stop the script from seeing where the ad sits, environments that do not run JavaScript such as some video players, and apps that do not implement the standard measurement SDK. Connected TV combines all three.
What does it actually cost?
$150 of every $1,000 entering a demand-side platform, per a study covering 35.5 billion impressions. That is more than any single fee line, including supply-side platform fees at $130.
Where did the 50% for one second threshold come from?
From a negotiation among three trade associations in 2012, validated by a technical feasibility pilot. The standards body's executive explained it was chosen because 80% of ads meeting it also qualified under a stricter criterion. It is a calibration between measurement methods, not a duration validated by research.
Does buying on viewable CPM solve it?
It reduces it. Google's own documentation states that in viewable CPM buying, a small portion of served impressions might still be charged. You exclude the measured non-viewable, not the never-measured.