In an era when audiences are scattered across broadcast TV, streaming services, and OTT/FAST channels, keeping track of who has seen what, when, and where—not to mention anything they did afterwards—is, well, hard.
Imagine a scenario where someone is watching live sports on their living room TV before switching to a tablet, while their partner watches from another room. This means multiple viewing events, two people—and three opportunities for a measurement system to get its math wrong.
That’s one of the central issues facing broadcasters in 2026. There can be no doubt that there is more data than ever, but pulling all of the strands into a reliable picture of who watched, and then what any advertising achieved, remains a stubborn challenge.
To address the problem and get a clearer picture, broadcasters and advertisers are turning to new measurement technologies and techniques, with the help of AI. TV Tech spoke to executives and measurement specialists from six companies across the industry to make sense of this brave new world.
Who’s On the Sofa?
As you might guess, Nielsen and Comscore are at the very forefront of understanding where viewers watch, no matter the location or device.
Nielsen’s Big Data Plus Panel approach combines viewing data from more than 45 million households and 75 million devices with a representative panel of more than 100,000 people. Its large-scale viewing data, including ACR (automatic content recognition) and set-top box data, provides the breadth, while the panel then helps establish who is actually watching and corrects for potential biases.
Comscore combines large-scale viewing data with panels because they solve two very different problems, according to Chief Analytics and Technology Officer Smriti Sharma.
“Big data gives us scale. It tells us what content was watched, on which devices and when,” Sharma said. “But panels help us understand who is actually watching.
“For example, imagine a smart TV is turned on for three hours,” she continued. “The data tells us exactly what was playing. But it doesn’t necessarily tell us whether it was a 25-year-old watching alone, a family of four watching together, or someone who left the TV running while making dinner!”
Nielsen is tackling some of these limitations through its updated co-viewing methodology, introduced in August, which makes greater use of wrist-worn devices that recognize program audio and are worn by panel members.
Nielsen’s pilot—which began with the Super Bowl in February—found that accounting for additional co-viewers increased measured audience figures by an average of 4.19% across seven major televised events, although those pilot figures weren’t incorporated into official ratings.
Nielsen has also reduced the threshold for crediting a quarter-hour of local TV viewing from 5 minutes to 1. The previous methodology left upwards of 24% of tuning events unreported, the company said.
Making the Numbers Add Up
Figuring out who’s watching is tricky enough. Making sense of where they’re watching—and ensuring the numbers match across different platforms—is another matter entirely.
Nielsen has accelerated its Streaming Content Ratings from weekly reporting to delivery each business day, covering viewing from two days earlier, while Comscore has introduced daily program-level reporting across linear and CTV. Looking ahead to 2027, Nielsen said it is developing new cross-platform content-measurement products.
But knowing what people watched is not necessarily the same as knowing who saw an ad, or whether those same people were reached across different platforms.
Comscore combines large-scale viewing data with panels because they solve two very different problems, according to Chief Analytics and Technology Officer Smriti Sharma.
“Big data gives us scale. It tells us what content was watched, on which devices and when,” Sharma said. “But panels help us understand who is actually watching.
“The biggest challenge is the inconsistency in how audience data gets translated into targeting and measurement across the campaign lifecycle,” David Levy, CEO of advanced advertising provider OpenAP, said. “An audience can be defined one way, targeted differently by each publisher and measured using an entirely different set of identity rules. When that happens, you end up not measuring the same audience you set out to reach.”
According to OpenAP’s analysis, even when leading identity providers have the same IP address in their graphs, they agree on the corresponding postal address only about 10% of the time. To address these inconsistencies, OpenAP is working with publishers to standardize how identities are resolved across linear and streaming, helping advertisers measure audiences more consistently.
Consistent audience measurement matters for more than reporting viewing figures: If advertisers cannot reliably identify the same audience throughout a campaign, establishing whether an ad subsequently influenced their behavior becomes much harder.
Beyond the Eyeballs
Even with a clearer picture of who is watching and where, there is more work to do to determine whether an ad actually influenced a viewer’s behavior. “Reach and frequency still matter,” Sharma said. “They’re the foundation, but they’re no longer the finish line, and I think that’s the biggest change for our industry.”
Seeing an ad, visiting a website, and buying a product are three separate events. The challenge for measurement providers is connecting them without assuming that one necessarily caused the next.
As CTV (aka connected TV) has entered the picture in recent years, these events have added to the complexity.
“For years we’ve relied on last-click, last-touch attribution, and it still can’t tell you whether that person was going to convert anyway,” Dan Larkman, CEO of Keynes Digital, a company that analyzes CTV, said. “CTV has its own version of the problem. It’s usually the first touch, not the last, so the sale often happens weeks later and gets credited to search or social.”
What’s known as “incrementality testing”—comparing exposed viewers with a comparable control group that didn’t see the ad—can help establish whether advertising actually generated additional business.
Nielsen said its own role is changing, too. “We really see ourselves becoming more of a media intelligence company,” a company spokesperson told TV Tech. Its Outcomes Marketplace brings different campaign performance metrics alongside audience data.
OpenAP is also working with publishers to standardize conversion data and connect advertising exposure with outcomes across linear and streaming.
AI Enters the Frame
You knew it was coming: AI is everywhere in 2026, and TV companies are making good use of the tech at all levels of the stack. There is a difference between what AI can do today and what the industry hopes it might do tomorrow, but some of the more practical applications are already here.
A great example comes from Comscore, which uses agentic AI to reconcile program metadata across linear and streaming. An episode of “Friends” might be identified by its series title on one platform, its episode title on another and a different identifier elsewhere. AI can recognize these connections without requiring everything to be manually matched.
Automatic content recognition (ACR): Smart-TV services equipped with ACR can identify shows and ads on screen, which generates large-scale viewing data across linear and supported streaming services. However, ACR only identifies the content, not the viewer.
Set-top box data: Cable, satellite and other TV receivers can provide return-path tuning data, which shows the channels viewers select and when, but only device activity is recorded.
Audience panels: Measurement companies recruit representative samples of people whose viewing habits and demographics are then measured. These help in calibrating larger datasets and mitigating biases, as well as understanding co-viewing.
First-party streaming data: Streaming platforms collect data about accounts, viewing sessions and viewing duration. However, shared accounts and devices can make identifying individual viewers difficult.
Identity resolution and de-duplication: Links viewing signals across devices, households and platforms, and then uses identifiers and statistical modeling to estimate unique audiences and overlapping reach.
—Max Slater-Robins
“AI is not deciding how we measure audiences,” Sharma said. “Our measurement methodology still determines that. AI is helping us do the work faster, more consistently, and at a scale that simply wasn’t practical before.”
Elsewhere, AI is helping advertisers put data to work. Nielsen launched Ad Intel AI in July, helping clients explore ad spending and competitive activity through a conversational interface.
VideoAmp’s AI-powered reporting system similarly allows users to interrogate campaign measurement data using natural-language questions, rather than manually extracting and manipulating reports.
And Samba TV’s Project Gravity combines its identity graph with AI-driven audience modeling, letting advertisers activate first-party data across more than 40 platforms for targeting, frequency management and campaign optimization.
For VideoAmp Chief Product and Technology Officer Josh Hudgins, the real opportunity lies in taking this a step further, reducing the gap between measuring an audience and acting on the results.
“In practice, an agentic system could understand the advertiser’s objective and constraints, monitor performance against those goals, identify an opportunity and recommend or eventually execute an optimization,” he said. “Measurement then becomes the feedback loop that tells the system whether that decision worked and informs what happens next.”
For now, though, there’s a distinction between the AI tools broadcasters and advertisers already use and the prospect of fully autonomous campaign management. AI may be advancing, but handing over advertising decisions entirely is another matter.
Trust But Verify
“AI is only as good as the underlying data,” the Nielsen spokesperson said. “So while AI may help uncover new patterns, trends and insights, it’s critical that you ensure the underlying datasets are accurate, unbiased and complete.”
As OpenAP’s Levy explained, “[If] the underlying identity and measurement data is inconsistent, AI will simply automate the same fragmentation we have today.”
In order for advertisers to fully trust the numbers, measurement providers must be able to explain where their figures come from and how audiences were identified, as well as the underlying assumptions.
AI, of course, can accelerate analysis and optimization, but the underlying data and decisions should always remain transparent and auditable, with human oversight.
“In a world where AI can generate almost any answer, the real competitive advantage will be the ability to prove that your answer is right,” Sharma said.
