You stop watching. The screen goes dark. You leave the room. You are no longer in the vicinity of the television.
In the connected television ecosystem, nothing has changed. The ad was served to the device — not to you — and the device is still reporting impressions. The measurement stack has no signal for an empty chair. A household of zero viewers generates the same bid response as a household of four. The room is gone. The impression remains.
This is not a glitch. It is the architecture.
DoubleVerify data shows that more than one in four top CTV environments continue serving impressions after the television has been powered off. The waste: approximately $700,000 per billion impressions — a figure combining TV-off impressions and invalid traffic, with the TV-off component not isolated. The company calls this a transparency problem. It is, more precisely, a structural accounting error at billion-dollar scale.
The GroupM-iSpot study from mid-2022 put sharper numbers on the device-level overcount: across streaming environments measured, an average of 8 to 10 percent of impressions were counted while the television was off. In dongle, stick, and gaming console environments — devices that treat the television as a dumb display — the overcount rate climbed to 17 percent. Native smart TV apps, where the app and the display share an operating system, showed near-zero overcount.
The distinction matters because it pinpoints the failure mode. The streaming stick in your HDMI port does not know the television it is plugged into has gone dark. It has no requirement to know. The HDMI-CEC specification — the protocol that allows a television to signal its power state to connected devices — makes implementation optional for software. The hardware can signal off. The device does not have to listen. The standard permits the gap.
The measurement gap spans three distinct technical layers, and no existing standard addresses all three simultaneously.
The first is hardware signaling. HDMI-CEC, mandated by the HDMI specification for electrical continuity, can carry power state information from television to connected device. The specification, however, makes CEC software implementation optional. A television can signal off. The stick on the other end has no obligation to receive the signal or act on it.
The second is operating system and application detection. Even if the streaming device receives a power-off signal, the OS and app layers have no standard interface to communicate that state to the impression measurement system. The app knows it is playing content. It does not know the screen it is outputting to has been deactivated.
The third is the impression measurement layer itself. Current standards — CMCD, SVTA Ad Creative Signaling, IAB OM SDK — operate at the streaming protocol layer. They measure what the app is doing. They have no mechanism to learn what the screen is doing.
This is why the TV-off problem has persisted despite years of standards development. Every standards body is working at layer three. The problem is at layers one and two.
Research from DoubleVerify found that 63 percent of marketers cannot verify whether their CTV ads are reaching real people or empty rooms — a direct consequence of the device-screen disconnect at the measurement layer. The structural problem is not that the industry lacks data. It is that the data being collected is device data, not viewer data.
The CTA published CMCD version 2 (CTA-5004-A) in February 2026, introducing an Event Mode for ad event reporting. It is a genuine improvement in streaming measurement infrastructure. It does not solve the TV-off problem. It operates at layer three.
The SVTA's ad creative signaling standards (SVTA2053/2053-1, SVTA5020) enable identification of ad content within SSAI and SGAI workflows — important for trafficking and verification. They do not carry device power state information. Layer three.
iSpot.tv took a different approach. Their patent (US 12,301,944, granted May 13, 2025), titled "Continuous Play Quantification," describes a system for correlating a device status feed with a content impression feed to identify continuous play events — and to exclude non-viewable impressions from reported metrics. This is layer two, with some layer one dependency on device-level reporting. It is the most direct technical solution to the measurement gap currently documented in the public record.
GroupM responded to the iSpot findings with a joint initiative involving Disney, Fox, Paramount, NBCUniversal, Warner Bros. Discovery, Vizio, and LG Ads, seeking new standards for cross-platform streaming measurement. An agency holding company coordinating major streamers on measurement methodology is unusual. The structural incentive for each party individually — platforms retain pricing power from measurement opacity, agencies derive billing complexity value from fragmentation — typically prevents this kind of coordination. The GroupM initiative suggests the waste has become visible enough to enough players that the coordination cost became worth paying.
McGranaghan, Liaukonyte, and Wilbur published a camera-based panel study in Marketing Science (2022) using TVision's infrastructure — 1,155 households, approximately four million ad exposures — measuring not just tuning but actual viewer presence during ad breaks.
The finding: approximately 30 percent of television ads play to a room with no one in it. Viewers do not typically change the channel during ad breaks. They leave the room. When they return, they find the content has continued. The television has been playing to no one while the measurement system has been counting everyone.
This is distinct from the TV-off problem but related at the conceptual level. Whether the viewer is absent in a room where the television is on (tuning away), or the television has been turned off but the streaming device continues reporting impressions, the measurement system records a viewable impression that did not occur. The 30-percent figure and the 25-percent figure are measuring different things at different layers of the same structural gap.
The academic evidence also suggests the detection problem is technically tractable. Ibrahim et al. demonstrated that mobile sensor fusion — microphone plus camera on a nearby device — can detect television on/off state at a 0.978 F-measure. The phone in the room knows the television is off. The measurement stack does not ask.
The honest answer is that solving the TV-off problem requires simultaneous coordination across hardware manufacturers, OS developers, streaming application developers, and measurement providers — on a definition of what constitutes a valid impression that none of them individually controls.
Hardware manufacturers (Samsung, LG, TCL, Vizio) control the HDMI-CEC implementation. OS developers (Google, Amazon, Roku, Apple) control what signals the streaming device can receive and act on. Streaming apps (Netflix, YouTube, Hulu, Peacock) control what those signals mean at the application layer. Measurement providers (DoubleVerify, iSpot, IAS, Nielsen) control how impressions are counted and reported. No single player can solve this alone.
The financial incentive structure does not reward solving it. Advertisers pay for impressions, including the ones that occurred when no one was watching. Platforms receive payment for those impressions. The measurement vendors are paid to count them. The $700,000-per-billion waste is distributed across a sufficiently large number of parties that no individual party bears the full cost.
The technical components for solving this already exist. HDMI-CEC hardware signaling is available. ACR (automatic content recognition) technology, used for content identification in smart TVs, functions even when the television is acting as a dumb display for an external device — reading the HDMI input to identify content. It cannot operate when the set is off entirely. iSpot's patent demonstrates the matching logic at the measurement layer.
What does not exist is the industry coordination to wire these components together into a unified standard. The CTA, IAB, SVTA, and JIC all publish standards. None of them has jurisdiction over layer one.
The current standards efforts — CMCDv2, SVTA signaling, IAB Europe's CTV measurement framework (in public comment through June 2026) — will improve layer-three measurement. They will not close the gap between "ad was served" and "ad was viewable" in CTV because they are working at the wrong layer of the stack.
The room is dark. The device is still billing. The standards bodies are in the wrong room.