A view runs from zero to thirty seconds depending on who sells it. On two platforms a views objective bills by impression. Completion is tied to no sale.
You may be paying for impressions under an objective called views. That is not an inference. It is what the documentation says, on two of the largest platforms, in a table most buyers never open.
Video is not a format. It is a billing category covering objects measured under different rules, sold under a shared vocabulary whose terms nobody agrees on, and optimized against a metric that has never been tied to a commercial outcome in a peer-reviewed venue.
The objective is not the billing basis
Start here, because it is the one thing on this page you can act on this afternoon.
Meta. The documentation on its completion objective states:
”15s ThruPlay is available on CPM or ThruPlay billing. 6s ThruPlay is available on CPM billing.”
The six-second objective bills by impression only. You optimize toward a six-second watch and you pay per thousand impressions.
LinkedIn. Its official table, under the video views objective, sets out three bidding strategies and what each one charges: maximum delivery, charged by impressions; cost cap, charged by impressions; manual bidding, charged by video views. The two automated modes, which are the defaults most buyers use, bill by impression under an objective named views.
Google. Cost-per-view bidding genuinely charges on watching: “you pay when a viewer watches 30 seconds of your video (or the full duration of the video if it’s shorter than 30 seconds) or interacts with your video, whichever comes first.” A skip before the thirtieth second costs nothing. But the same inventory can also be bought on target CPM, so the format does not determine the basis either.
TikTok. Its views objective does bill on views, at six or fifteen seconds or completion, and explicitly excludes interactions in the first second from billing.
Four platforms, four logics, one vocabulary. The question to ask is not what a view costs. It is whether you are billed per impression or per view, and from which second.
Underneath the billing question sits a definitional one. Here is what each platform calls a view, per its own documentation.
Platform
What counts as a view
YouTube, organic content
the moment playback starts
LinkedIn
2 continuous seconds, at 50% on screen
X
2 seconds at 50%, or 3 seconds at 100%, or 6 seconds
Meta
3 seconds, with 15 seconds for the billed metric
TikTok
playback start, with 6 and 15 second variants
YouTube, paid campaign
30 seconds or the end of the video
Zero to thirty seconds, and three definitions sold separately in one campaign manager at a single company.
Two of these track the industry viewability threshold of 50% on screen for two seconds. That alignment is voluntary. Nothing in the standards requires a platform to define a view that way, and most do not.
Which means: comparing view counts across platforms compares nothing. The word is shared. The object is not.
The video viewability threshold is 50% of pixels for two continuous seconds, against one second for display. The standard adds a detail few buyers know:
“This required time is not necessarily the first 2 seconds of the video ad; any unduplicated content of the ad comprising 2 continuous seconds qualifies in this regard.”
Two seconds anywhere in the ad. Not the opening.
And the threshold is justified nowhere. No normative document motivates it. The only public explanation is an executive of the standards body telling a trade publication that users “really didn’t act upon the ad, in itself, until the two- or three-second mark. That’s why we came up with the two-second mark.” Never formalized, never published with data.
That is precisely the situation of the display threshold, which came out of a trade association negotiation. Video is not better founded. It is only different.
Audio is not required. The standard says so plainly: “detection of audio is not currently a requirement for a viewable video ad impression under these guidelines.”
But a second standard, covering combined television and digital measurement, requires both audio and 100% of pixels. The document describes the two regimes side by side: “digital video viewability using a 50% pixel criteria with no consideration of audio vs. cross-media video viewability using a 100% pixel criteria with consideration of audio.”
Two official, incompatible definitions of a viewable video impression. The same spot can count as seen under one and not the other.
One more comparability problem, technical but consequential: the standard permits measuring 50% of the pixels of the ad or 50% of the pixels of the player, provided the choice is disclosed and shown to be immaterial. Two vendors can both claim compliance while measuring different surfaces.
The rule dates from 2015 and lives in a footnote: “Banner ads with video embedded within them generally are covered by the display ad criteria for viewable impression measurement.”
A video playing in a display slot is measured as display. It becomes viewable after one second, not two. The 2022 format guidelines carry the classification forward unchanged, filing these objects under display video and treating them as rich media.
And the industry changed how it names video because the old naming was being used to overcharge. The technical field that classified video in bid requests was deprecated in March 2023. It sorted by placement: in-stream, in-banner, in-article, in-feed. The documented reason for the change is that this classification allowed out-stream video to be presented as in-stream video, which sells for more.
The replacement classifies by user intent and by sound signal. To qualify as in-stream, a video must have sound on by default at player start. And the in-banner value was removed from the list entirely.
That is a doctrine change, not a cosmetic update: video is no longer classified by where it appears, but by what the user came to do.
On a television, viewability stops existing
In connected TV the notion of a percentage of pixels in a browser tab has no meaning. The reference document for that environment, published in August 2021, contains no occurrence of the words pixel or viewability.
A figure of 100% of pixels for two seconds does circulate for connected TV. It exists, dated and signed, but it belongs to the cross-media standard, not to any connected-TV-specific standard. The questionnaire the same body supplies to buyers asks vendors whether they “have or are developing CTV viewability capabilities”, which settles it.
What replaces pixel measurement are state checks, and one of them deserves reading twice:
“Certain CTV devices may include dedicated power sources and as a result, may be independent of the power state of the TVs used to display their content. In such environments, CTV video content and advertising may be played while corresponding TV sets are off. […] detection of TV Off is not currently a requirement for CTV video impression measurement.”
The device has its own power supply. Ads can therefore run while the screen is off, and detecting that is not a measurement requirement. It is encouraged.
Server-side ad insertion compounds it. When content and ads are stitched server-side, the player cannot process ad tracking and all traffic originates from one address. The document draws the conclusion without hedging: “CTV environments especially may be more vulnerable to ad fraud than other formats”, and “IVT perpetuators may disguise themselves as SSAI providers.”
The best-documented case was disclosed in April 2020: an operation impersonating more than two million people across thirty countries, counterfeiting over three hundred publishers, and accounting at its peak for 66% of programmatic server-side-inserted traffic in the segment observed.
One correction worth making here, because it cuts against the easy narrative: fraud rates do not differ by format. The advertiser association study states it plainly: “There was no meaningful difference in IVT rates between: Display (which was 82 percent of activity) and video.” What differs is measurability, not fraud.
Completion rate is the central indicator of video advertising. It measures the share of ads played to the end. Nobody has demonstrated, in a peer-reviewed venue, that it predicts anything commercial.
The foundational study appeared in 2013 in the proceedings of an ACM conference. It covers 65 million unique viewers, 362 million videos and 257 million advertisements, and its authors present it as “the first rigorous scientific study of video ads and their effectiveness”. Its findings on what drives completion are solid: a fifteen-second ad completes 2.9% more often than a twenty-second one, a mid-roll completes 18.1% more often than a pre-roll.
Then comes the sentence nobody quotes:
“Our current data set does not currently allow us to measure CTRs or survey responses. But, comparing the different metrics of ad effectiveness is an interesting avenue for future work.”
The most rigorous study ever published on completion rate says explicitly that it could not test the link to an outcome, and files the question as future work. That was thirteen years ago, and no peer-reviewed article identified since has closed it.
What circulates instead comes from video ad technology vendors, and none of it addresses the obvious problem: people who finish a video were probably more interested in it to begin with, independent of any effect of finishing.
The same void covers format comparison. No published causal experiment compares video to another digital format on a commercial outcome at equal budget. The closest observational study does not even separate video from display.
And when video’s persuasive advantage was tested properly, it proved modest. A 2021 paper in the proceedings of the US National Academy of Sciences, based on 7,609 people and 26,584 observations across 72 different messages, randomly assigning a video, its written transcript, or a control:
“individuals are more likely to believe an event occurred when it is presented in video versus textual form, but the impact on attitudes and behavioral intentions is much smaller […] the difference between the video and text conditions is comparable to, if not smaller than, the difference between the text and control conditions”
An important caveat: that study covers political persuasion, not brand advertising. It does not transfer mechanically. But it is the largest and most rigorous published test of the claim that video is inherently more persuasive, and it largely refutes it. No commercial study of comparable scale exists to settle the same question for a brand.
Two smaller points, for completeness. The claim that people retain 95% of a message seen on video against 10% read as text has no identifiable academic source. And the widely quoted figure that 85% of social video is watched without sound traces back to a 2016 trade article reporting two publishers’ self-declared numbers, not to a measurement.
One last fact, because it says more about this category than any threshold.
A trade association ran an annual television production cost survey for twenty-five years. It was discontinued in 2013. The reason given by an association executive at the time:
“Who shoots just a 30-second commercial (alone) anymore? […] And what is a ‘commercial’ anyway?”
The profession had an instrument for measuring what a video ad costs to make, and retired it because the object it measured stopped being definable. Nobody has replaced it. Which is why no institutional source exists today for the production cost of a video ad, and why every figure you find comes from a production company’s own website.
That is the whole article in one anecdote. Video did not become harder to measure because measurement got worse. It became harder to measure because “a video ad” stopped being one thing.
What to do
Ask which second triggers billing, and whether it is per view or per impression. First question, before budget, before targeting. On two major platforms the automated bidding modes bill by impression under an objective called views.
Never compare view counts across platforms. Zero to thirty seconds. The only comparison carrying information is one platform against itself, over time, at constant format and placement.
Check that your video is running in a video slot. A video in a banner is measured as display. If your deal says video and delivery happens in display, you are paying a video price for a display object, and the industry rewrote its classification in 2023 precisely because that confusion was being exploited.
In connected TV, ask what is actually detected. No viewability threshold, no requirement to detect a dark screen, and an environment the standard itself calls more vulnerable to fraud. Ask your vendor what it detects, how, and what it does not.
And stop optimizing on completion rate as though it were a result. It is a delivery indicator whose link to a sale has never been established in a peer-reviewed venue, and whose foundational study says outright that it could not test that link. It tells you the video played to the end. It does not tell you it did anything.
The only measurement that belongs to you, in video as everywhere else, is the one your own system can produce: what the campaign cost, and what it produced on your side. The rest is a vocabulary whose definitions you do not control. Owning that measurement is the first phase of our B2B paid acquisition work, where the conversion tracking is built on your own side and the video is then judged on what it produced there.
Frequently asked questions
What counts as a video view?
It depends on who is selling it. On one platform a view is counted the instant playback starts, meaning zero seconds. On another it requires thirty seconds or the end of the video. In between: two, three, six or fifteen seconds. One platform sells three different definitions in the same campaign manager.
Am I billed for views or impressions?
Often impressions, even under a views objective. Meta's documentation states its six-second objective is available on impression billing only. LinkedIn's official table shows that under a video views objective, the two automated bidding modes are charged by impressions and only manual bidding is charged by views.
Where does the two-second video threshold come from?
No published document justifies it. The only explanation on record is an executive of the standards body telling a trade publication that users did not react to the ad before the two or three second mark. It was never formalized with data.
Does completion rate predict anything?
Nobody has demonstrated it in a peer-reviewed venue. The foundational study on the subject, covering 257 million ads, writes plainly that its data could not measure commercial outcomes and files the question as future work. That was 2013.
Is video more effective than other formats?
No published causal comparison establishes it on a commercial outcome at equal budget. The largest test of video's persuasive advantage over text, across more than seven thousand people, finds the advantage far smaller than intuition suggests.