Co-viewing describes what happens when more than one person sits in front of a screen at the moment an advertisement plays. One device receives one ad. Two, three or four people may see it. The term exists because television advertising has always been sold on audiences while connected television advertising is delivered to devices, and something has to close the distance between those two units. Co-viewing is the name for that distance, and for the collection of panels, surveys, sensors and models used to estimate its size.

The commercial consequence runs in both directions. A publisher selling a living-room impression to a family of four wants credit for four exposures rather than one. A buyer wants to know that a reported reach of ten million people is not really six million screens with a multiplier applied to them. Neither party can settle the question by observation, because the thing being measured, the number of bodies in a room, is invisible to an ad server.

How the number is produced

Co-viewing enters reporting as a factor applied on top of delivered impressions, or as an adjustment to unique reach. In linear television planning the factor is usually expressed as viewers per viewing household, abbreviated VPVH, which counts how many people are watching per household that has the set tuned to a given programme. Multiply households by VPVH and a household rating becomes a person rating. The IAB UK has described the traditional linear approach as a flat rate, a single average applied across inventory rather than a figure that varies by programme or daypart.

Four methods dominate the estimation, and they do not agree.

Panels with active log-in. Nielsen's people meter, introduced nationally in the United States in 1987, assigns each household member a button. Viewers press to register presence. According to Nielsen, individual viewer data of this kind is what allows co-viewing to be estimated inside much larger big-data sets that carry no person-level signal at all.

Passive sensing. TVision runs an in-home panel using computer vision, performing person and facial recognition alongside automatic content recognition to identify what is on screen. Nielsen began piloting wrist-worn wearables at Super Bowl LX on February 8, 2026, devices resembling a smart watch that capture audio from programming. According to Nielsen's August 19, 2026 announcement, the passive collection removes the formal log-in step and should produce a more accurate count of how many people are watching a given programme.

Surveys and modelling. Google measures co-viewing on TV screens using a combination of connected television panel data and real-time census-level surveys covering more than 100 countries in over 70 languages, feeding the result into a dynamic reach model that adjusts for day of week, time of day, country and user demographics, according to Google Ads Help.

Deterministic identification. Comscore's multi-year partnership with HyphaMetrics, announced July 15, 2025, pairs the Unified Individual Media Metric with Comscore's personification methodology to produce deterministic VPVH and person-level co-viewing across devices within a household.

Inside a buying platform the output surfaces as named fields. Google's Reach Planning API exposes total_coview_reach, on_target_coview_reach, total_coview_impressions and on_target_coview_impressions alongside their non-co-viewed equivalents, splitting reach that matches the targeting criteria from reach that spills outside it. Spill matters here: a co-viewer is by definition not the signed-in account holder, so demographic targeting accuracy degrades as the co-viewing factor rises.

From people meters to platform self-measurement

Co-viewing predates programmatic entirely. Households watched television together long before anyone tried to count them, and the 1987 shift from paper diaries to electronic people meters was in large part an attempt to convert household tuning into person-level audience. Diary measurement relied on recall; the meter recorded who was logged in, minute by minute.

Digital video arrived with the opposite problem. Cookies, mobile advertising identifiers and connected television device identifiers all describe hardware or accounts, not rooms. For a decade that mismatch was tolerable because most digital video played on personal screens. It stopped being tolerable when the television set became a significant digital surface.

Google announced co-viewing metrics for its planning and measurement tools in April 2022, with a global rollout across Google Ads and Display and Video 360 scheduled by the end of the second quarter. The company cited a study with Nielsen finding that multiple viewers aged 18 and over watched YouTube together on the TV screen 26 per cent of the time, against 22 per cent on linear television.

The next step drew objections. In August 2023, according to Ad Age reporting summarised by eMarketer, YouTube told clients it would move away from third-party measurement of co-viewing and trade on its own surveys and panel data from January. Nielsen said at the time it did not plan to incorporate Google's co-viewing figures into its own measurement. Agencies read the move as a platform grading its own homework on a metric that directly inflates reported reach.

Google completed the definitional shift on June 2, 2026, when the Google Ads API adopted a Total Co-view definition for seven campaign-level metrics: metrics.unique_users, metrics.average_impression_frequency_per_user and five frequency-threshold variants running from two-plus to ten-plus exposures. Mattia Tommasone of Google's Advertising and Measurement Developer Relations team wrote on the Ads Developer Blog on May 27, 2026 that no code changes were required, but that reported reach and frequency values would move from the launch date forward. The same 48-hour window carried a separate three-year retention limit on those granular reach metrics.

Nielsen followed on August 31, 2026, folding wearable-based co-viewing into Big Data plus Panel as one of seven simultaneous currency changes, negotiated with Media Rating Council oversight. The company paired the release with a statement that no ratings increase was guaranteed.

Why buyers and sellers argue about it

Co-viewing sits directly on price. If the factor rises, cost per thousand falls for the same spend, because the denominator grows without any change in delivery. If it falls, inventory that was sold as reaching four people is reaching fewer. The Video Advertising Bureau noted in its July 14, 2026 explainer that minor fluctuations in co-viewing can have significant impact on impression counts and estimates.

Measured differences between platforms are large. A VAB and TVision study published in February 2026, drawing on data collected between July 2024 and June 2025 across 21 premium video platforms, defined co-viewing as the share of impressions with two or more viewers holding overlapping active sessions of at least five minutes. Premium platforms averaged 60 per cent; YouTube recorded 45 per cent, ranking 18th of 19 entries. Spanish-language free ad-supported streaming platforms topped the chart at 75 per cent.

Content type matters more than platform in some cuts. TVision figures from September 2025, cited by CIMM, put American football at 68.2 per cent, hockey at 64.0 per cent and golf at 56.1 per cent, against 52.8 per cent for non-sports content. Magnite's August 20, 2026 survey of 835 American adults found 69 per cent of live connected television viewers always or often watch with others, with an average of three companions, though that figure is self-reported behaviour rather than measured impressions.

Limitations and disputes

The central criticism is that co-viewing numbers are modelled rather than observed. TVision put the point sharply in a 2025 paper, arguing that the same household watching the same show can be counted as 1.5 viewers by one provider and 1.2 by another, not because behaviour changed but because the arithmetic did. Models that overstate presence inflate reach and understate frequency per person, which is the direction of error most convenient for sellers.

There is also no single accredited method. As of September 2026 at least three approaches operate in the same American living rooms: Nielsen's wearables inside currency, Google's survey-and-panel model inside its own reporting, and TVision's computer vision sold as an independent layer. A buyer running linear, YouTube and open connected television in one campaign receives three co-viewing figures that cannot be reconciled, each treated as authoritative within its own environment.

Frequency management degrades in parallel. A co-viewer counted in reach cannot be capped, because the platform has no identifier for that person. Higher co-viewing therefore produces cleaner-looking frequency distributions without any corresponding change in exposure control.

Distinguishing co-viewing from adjacent terms

Co-watching refers to synchronised remote viewing through features such as watch parties, where participants sit at separate screens. That produces multiple impressions on multiple devices and requires no adjustment.

Out-of-home viewing covers screens watched outside the measured household, in bars, hotels and offices. Nielsen estimates it may account for 10 to 15 per cent of total viewership for many sporting events. It is a separate undercount with separate methodology.

VPVH is a metric expressing the co-viewing factor as a ratio, not a synonym for the behaviour itself.

Attention measures whether eyes are on screen. Presence and attention diverge sharply: TVision reported on August 25, 2026 that of roughly 200 daily minutes with video content, only nine are attentive advertising minutes.

Recent developments

Nielsen extended four-screen deduplication in Nielsen ONE Ads to all measurable connected television inventory in Japan on August 3, 2026, including co-viewing in total impression delivery. Ownership of the observational layer has also shifted: Viant acquired TVision for $40 million, announced April 15, 2026 and closed in May, moving panel signals covering who was in the room and whether people were co-viewing inside a single demand-side platform. TVision had previously licensed that data to iSpot, VideoAmp and Oracle.

Timeline

  • 1953: Nielsen begins paper viewer diaries, recording household self-reported viewing
  • 1987: Nielsen deploys the people meter nationally, capturing person-level presence
  • April 2022: Google announces co-viewing metrics for Google Ads and Display and Video 360, citing a Nielsen study measuring 26 per cent co-viewing on YouTube CTV
  • August 2023: YouTube tells clients it will trade on its own co-viewing measurement from January 2024
  • July 15, 2025: Comscore and HyphaMetrics announce a person-level co-viewing partnership
  • September 2, 2025: Nielsen Big Data plus Panel becomes currency for the United States television season
  • February 8, 2026: Nielsen pilots wrist-worn wearables for co-viewing at Super Bowl LX
  • February 2026: VAB and TVision publish co-viewing shares of 60 per cent for premium video and 45 per cent for YouTube
  • June 2, 2026: Google Ads API adopts the Total Co-view definition across seven reach metrics
  • August 19, 2026: Nielsen announces seven currency enhancements including wearable-based co-viewing
  • August 31, 2026: The enhancements take effect in Big Data plus Panel

Summary

Who: Measurement providers including Nielsen, Comscore, TVision and HyphaMetrics; platforms including Google and YouTube; and the buyers, publishers and trade bodies that transact against the resulting numbers.

What: The presence of more than one person at a screen when an advertisement plays, expressed as a multiplier on impressions or an adjustment to unique reach, and estimated through panels, surveys, sensors and models rather than observed directly.

When: Person-level television measurement dates to the 1987 people meter. The connected television version of the problem became commercially material from 2022, when Google introduced co-viewing metrics, and reached currency status on August 31, 2026.

Where: Primarily the shared household screen in linear television and connected television markets, with the sharpest disputes in the United States, where three measurement methods now operate in parallel.

Why: Digital video counts devices while television is sold on people. Co-viewing is the adjustment that reconciles the two, and because it moves reported reach, frequency and effective cost per thousand without changing delivery, its methodology is contested by everyone it prices.