Device detection firm 51Degrees analyzed more than 11 million OpenRTB ad impressions carrying a device type signal between May 15 and 20, 2026, checking each one against its own device detection technology to confirm what was actually rendering the ad. The result: 4.92 percent of inventory labeled as connected TV was not being watched on a television set at all, while just 0.18 percent of genuine CTV inventory was mislabeled the other way, as something less valuable. Applied to eMarketer’s global market size figures, 51Degrees estimates the mislabeling costs advertisers close to 2 billion dollars a year in inflated prices.
The direction of the error is the story. Mislabeling runs almost entirely one way, toward overvaluing inventory, which means publishers stand to gain from it and advertisers absorb the cost every time. CTV commands a meaningfully higher CPM than standard video precisely because buyers trust the label, and this data set is a rare independent check on whether that trust is earned at the rate the premium implies.
For a media buyer, the original insight here is not that some mislabeling exists, it is how narrow the gap between honest and dishonest labeling has become to measure at all. A five percent error rate sitting almost entirely in one direction is closer to a systemic pricing problem than a rounding error, and it lands in the same conversation as the broader push toward treating measurement itself as something to audit, not simply trust, and the newer wave of AI-driven measurement layers now being built to catch exactly this kind of discrepancy faster than a manual review can.
Source: 51Degrees
