What is Video view?

A video ad is delivered 400,000 times. Under a two-second definition it records 96,000 views, a view rate of 24%. Under a fifteen-second definition the same delivery records 11,200 views, a rate of 2.8%. The retention between those two thresholds is 11,200 ÷ 96,000 = 11.7%, which is the number that says something about the video. Neither view count says it on its own.

A video view is recorded when playback crosses a threshold the platform has chosen. That is the entire definition, and the thresholds are not standardised.

Across the major video surfaces, a view has been defined at two seconds, three seconds, six seconds, fifteen seconds, thirty seconds and completion, sometimes with an additional condition about how much of the player has to be visible. Several platforms report two or three of these at once, in adjacent columns, under labels that differ by a word.

One campaign can therefore report ninety-six thousand views and eleven thousand views at the same time, both accurately. Which figure appears in a report is a reporting choice, and it changes every derived metric — view rate, cost per view, and any comparison to another campaign or another platform.

Because the thresholds differ, a view count from one platform cannot be compared with a view count from another. A platform using a two-second definition will always look more efficient than one using fifteen, and the difference says nothing about either.

This is one of the more commonly violated rules in reporting, partly because view counts are large and satisfying and partly because the threshold is rarely printed beside them. A cross-platform video summary adding view counts together produces a number with no referent at all.

What does travel is the shape. The proportion of people still watching at a given point is a property of the video, and while the measurement points differ, the pattern of decay is comparable in a way the raw counts are not.

Most in-feed video plays automatically and silently as a person scrolls past. Somebody moving through a feed at normal speed crosses a two-second threshold repeatedly without ever deciding to watch anything.

This is the mechanism behind almost every implausibly cheap view rate, and it is why in-feed, sound-off placements reliably produce the lowest cost per view in an account and the weakest evidence of attention.

Placements where a person pressed play, or where the ad interrupts content they chose to watch, produce far more expensive views and far better evidence. Skippable pre-roll occupies a useful middle position: the person who did not skip made a decision, which is why a view counted past the skip point is worth several counted before it.

An account optimising on view volume alone migrates toward the cheapest inventory automatically, because that is what the metric rewards.

The view count compresses a distribution into a single number and discards the shape, and the shape is where every instruction lives.

A curve that collapses in the first two seconds is a hook problem. The opening frame is static, the first words are throat-clearing, the logo came before the point, or the subject of the video is not visible in the thumbnail moment.

A curve that holds for five or six seconds and then falls is a promise problem. The opening earned attention and the rest did not repay it.

A curve that decays evenly all the way down usually means the video is longer than the idea in it, and a shorter cut will outperform it.

A curve with a sharp drop in the middle points at a specific moment — a scene change, a price, a claim, a tonal shift — which is locatable and fixable.

Two campaigns with identical view counts can have completely different curves and completely different instructions. Reading the count produces bid changes; reading the curve produces edits.

Design decisions follow directly from how views are generated.

Assume the sound is off. Most feed views begin silent, and a video whose first seconds depend on audio has lost its audience before the audio would have mattered. Captions are not an accessibility afterthought here; they are the primary channel.

Put the subject in the first frame. The thumbnail moment is the ad for the ad.

Open on motion. Static openings lose disproportionately in the first second, because motion is what stops a scroll.

Put the point before the branding. Brand-first openings perform reliably badly in feeds, where the cost of the first two seconds is the entire audience.

Views are a delivery metric. They say something was shown and partly watched; they say nothing about whether anybody remembered it or acted on it.

That makes them suitable as an objective in a narrow band of cases: upper-funnel work where the outcome is too far away to optimise against, and creative testing where the retention curve is the thing being tested.

Everywhere else they are a diagnostic. A campaign meant to produce sales should be bought on sales, and its view figures read the way a mechanic reads a gauge — useful for locating a problem, never the reason the vehicle exists.

The one place a view count carries value beyond diagnosis is as an audience definition.

People who watched a substantial share of a video have demonstrated something no impression can: they chose to keep watching. Audiences built from deep view thresholds — three quarters watched, or completion — are materially warmer than audiences built from impressions, and on some accounts they perform close to site-visitor pools while being far larger.

The threshold used to build the audience matters as much as it does in reporting. A pool built from two-second views is barely distinguishable from everyone the campaign reached, because it includes everyone who scrolled past. A pool built from fifteen seconds or completion is small and meaningfully self-selected.

This turns upper-funnel video from a spend with a distant payoff into a mechanism that produces a usable asset immediately, and it is the strongest available argument for buying video at all in a direct-response account.

Three details change the numbers more than people expect.

Looping. Some placements replay a short video automatically. Whether a second loop counts as a second view, or extends the first, differs by platform and affects short-form counts substantially.

Player size and position. Some definitions require a share of the player to be on screen; others do not. An ad playing in a small, partly visible player can accumulate views that a stricter definition would refuse.

Continuous play. On surfaces where content plays back to back, a video that begins while the person has left the room accrues watch time nobody spent. This is why completion rates on some placements look far better than the engagement behind them.

None of these are reasons to distrust the metric. They are reasons to compare like with like, and to prefer the retention ratio over the raw count whenever the comparison spans placements.

• Write the threshold beside every view count. A number without its definition cannot be compared with anything, including the same campaign last quarter.
• Read the retention curve rather than the view count. Where the curve falls is an instruction about the edit; the count is not.
• Assume sound is off in feed placements and design the first seconds for it. Most views begin without audio and many end before it would have mattered.
• Compare thresholds within one platform instead of counts across platforms. The ratio between two thresholds is comparable even when the raw counts are not.
• Treat a cheap view rate on autoplay inventory with suspicion. Autoplay produces views from people who never chose to watch anything.