Advertising on internet-delivered television is measured differently from both broadcast and desktop video. The unit being counted is usually a device in a household, and that has consequences throughout the report.
The device is the identifier
A television application reports an identifier belonging to the hardware or the platform account, not to whichever person is watching at that moment.
Several people may watch the same impression together, and the same person may appear as different identifiers on different devices in the same home.
Reach figures therefore describe screens reached rather than people reached, and translating between the two requires assumptions about household composition.
Co-viewing is estimated, not observed
Because the device cannot see the room, the number of viewers per impression is modelled from panel data and applied as a multiplier.
Those multipliers differ by content type and by time of day, since a live event assembles more people in front of one screen than a weekday afternoon programme.
Two vendors applying different assumptions will report different audience sizes from identical delivery, which is a definitional gap rather than a measurement error.
Frequency control is harder than on the open web
Inventory is fragmented across many applications and platform storefronts, each maintaining its own view of the household identifier.
Without a shared identity layer, capping applied inside one application does nothing about exposure delivered through another.
Households at the heavy end of viewing consequently absorb far more repetition than planned, which is the most frequent complaint about the format.
Completion is near-universal and uninformative
Most television placements are non-skippable, so completion rates cluster near the top of the range for every advertisement in every campaign.
A metric that does not vary cannot distinguish between executions, and using it as a performance indicator produces confident conclusions from no signal.
More useful comparisons come from outcomes measured elsewhere, such as branded search activity or site visits observed in the same household window.
Server-side insertion changes what can be verified
Many television advertisements are stitched into the video stream before it reaches the device, which makes the delivery indistinguishable from the programme itself.
This improves the viewing experience and defeats blocking, but it also removes the client-side signals that verification vendors traditionally rely on.
Verification in this environment therefore depends on the stitching platform reporting honestly, which is why authorised-seller declarations carry unusual weight in television buying.