NumberEight, the London based audience intelligence company, published a report on September 23, 2026 comparing traditional demographic targeting methods against approaches that avoid persistent identifiers altogether, finding that IP address and household level data consistently produced near uniform gender and age distributions across mobile games and television genres with well documented audience differences.

In Short

NumberEight tested whether the age and gender data advertisers buy actually matches who is watching or playing. In a mobile game called Fashion Battle, IP based data said the audience was 51 percent female, while NumberEight's own model, which does not use IP addresses at all, measured 82 percent female. Across genres from horror to romance and TV shows, IP and household level data kept returning numbers clustered near 50/50, which the report argues cannot be right for content with genuinely different audiences, and the conclusion for the marketing community is that some demographic data long treated as reliable may need a second look.

A report built on two environments and two identifiers

The company published the 20-page report, titled "Demographic Data in Advertising: Traditional vs. ID-less Methods," in September 2026, according to NumberEight. The document compares three long-standing methods for determining who consumes advertising content - self-reported data, IP based targeting and third-party identifiers - against what NumberEight calls ID-less methods, which infer demographic traits from behavioral and contextual signals rather than tracking individuals across sites and devices.

The report frames its central question early: does identifying and matching more users actually lead to better audience intelligence, or does the pursuit of identity itself introduce distortions that go unnoticed once data reaches a media plan. NumberEight tested that question against two signals, age and gender, in two environments, mobile gaming and connected television, using data drawn from campaign delivery records and a large third-party viewership dataset.

A companion press release distributed on September 23, 2026 and dated from London summarized the findings under the headline "Do Advertisers Really Know Who They're Reaching?" according to NumberEight. That release, along with a promotional landing page for the report, forms the second and third source document behind this article, alongside the report itself.

Self-reported data carries its own distortions

Before turning to IP and identifier based methods, the report examines self-reported data, meaning information a person types into a form, survey or account registration. The reliability of that information, the report states, is increasingly being scrutinized by researchers who study why people misreport personal details.

The report cites a 2013 Pew Research Center study of American teenagers, which found that over a quarter of teen social media users had posted false demographic information, such as a fake name commonly used as a proxy for gender, specifically to protect their privacy. A 2024 survey cited in the report found that 21 percent of online dating service users admitted to lying about their age on their profiles, while 55 percent of respondents in a separate study said they expected others to misrepresent their gender online. Ofcom, the UK communications regulator, found that a third of children aged 8 to 17 with a social media account had registered using an adult age, according to the report.

None of these figures come from NumberEight's own testing; they are drawn from prior academic and regulatory research that the report assembles as background before introducing its own comparisons. The report also cites a real-world example from NumberEight's own client base: during a test period in August 2024, a mobile gaming publisher found that 5 percent of users had self-reported their age as 99 or 100 years old, entries the publisher internally flagged as implausible. That detail illustrates, in the report's framing, why even directly declared demographic data requires validation before it can be trusted for targeting.

The IP address problem: one address, many people

The bulk of the report's argument rests on a more structural critique of IP based targeting, the practice of using a device's or household's internet protocol address as a proxy for identity when cookies, mobile advertising identifiers or other persistent identifiers are unavailable. An IP address, the report notes, is in ideal conditions a unique numeric label assigned to a household or mobile device, and in the absence of internet proxies or VPNs it is visible in full to websites and ad networks for technical reasons, which has allowed it to function as a rough approximator of both identity and location.

The report distinguishes between the two IP standards in current use. Under IPv4, the older and still more prevalent format, network providers frequently share a single address across multiple internet subscribers due to a longstanding shortage of available addresses. Under the newer IPv6 standard, a longer alphanumeric string can identify an individual device within a household rather than the household as a whole. In both formats, the report states, addresses typically change only every few days or weeks under normal conditions, which is often a long enough window for advertising purposes.

The deeper limitation, according to the report, is not address volatility but shared occupancy. Because IPv4 remains the dominant standard, an address typically identifies a household or shared network connection rather than an individual, and results become inaccurate whenever multiple people of different ages and genders share that connection - which describes most family households. One historical workaround involved reading detailed browser user agent information to distinguish devices within the same household. That method, the report states, was largely closed off by 2023, when many browsers revised the information they share to include only broad categories such as operating system and browser version, significantly limiting its usefulness for individual level fingerprinting.

Privacy behavior compounds the structural problem. The report cites data showing that 33 percent of Americans reported using a virtual private network in 2023, with VPN adoption exceeding 50 percent in some countries globally, further eroding the reliability of IP addresses as consistent match keys.

Third-party identifiers face their own accuracy gap

The report's third traditional method, third-party identifiers such as cookies and mobile advertising identifiers, has historically supported retargeting, frequency capping, attribution and audience segmentation by linking user activity across websites, apps and devices into persistent identity graphs. Those graphs, the report argues, are inherently difficult to maintain: cookies expire or are deleted, mobile advertising identifiers reset or become unavailable, and users routinely move across multiple browsers and devices in ways that cannot always be reliably linked back to the same individual.

The report cites an independent study by Truthset, the data validation firm, finding that 51 percent of data used for ad targeting is inaccurate, with average accuracy rates ranging from 32 percent to 69 percent across different data providers. PPC Land has reported separately on Truthset's broader State of Data Accuracy 2026 findings, which broke inaccuracy down by specific signal type and found that age segments in particular collapsed to just 13 percent accuracy, with gender segments at 61 percent and presence-of-children segments at 41 percent.

The NumberEight report notes that even deterministic identifiers, such as hashed email addresses and cookies generally considered more precise than probabilistic methods, can still suffer from attribution errors that lead to incorrect audience segmentation. Browser restrictions from Safari and Firefox blocking third-party cookies, user opt-outs enabled by initiatives such as Apple's App Tracking Transparency framework, and increasingly shorter identifier lifespans have collectively made accurate cross-device targeting and measurement progressively more difficult, according to the report.

The report also notes, in a detail that touches on one of the more closely watched policy reversals in digital advertising, that Google will no longer deprecate third-party cookies in Chrome - a decision the search giant confirmed in April 2025 after years of delay, as PPC Land reported at the time. Even with that reversal, the report argues, third-party identifiers still face major accuracy challenges and declining availability due to other browser restrictions, privacy regulations and widespread user opt-outs.

How the ID-less approach is constructed

Against this backdrop, the report positions ID-less methodologies as a category of solutions designed to support the same advertising use cases - targeting, measurement and frequency capping - without relying on any persistent identifier that could trace activity back to a specific person or device. According to IAB Tech Lab, whose ID-Less Solutions Guidance the report cites as a reference framework, ID-less solutions use first-party data to define audience categories rather than individuals, rely on contextual metadata such as content and environment, and share only group-level aggregated data structured so that individuals cannot be re-identified.

The report describes four common techniques falling under this umbrella: contextual targeting, which matches ads to content or setting rather than a user profile; cohort-based targeting, which groups users into anonymous segments based on shared behavioral patterns without identifying individuals; on-device AI modeling, in which machine learning models run locally on a device and build behavior-based signals without transmitting personal data off the device; and private aggregation techniques using encryption or differential privacy to enable targeting or measurement without revealing individual-level data.

NumberEight describes its own Affinity Audience solution as inferring demographic attributes such as age and gender through a combination of anonymized first-party demographic data, device metadata, app store content, content metadata and contextual data available in advertising requests and directly from publishers. According to the report, those inputs are used to build a large ID-less panel, which then trains machine learning models that are further calibrated against aggregate market and census-level data to help ensure accuracy and consistency across different markets. The report states that at no point in this process does any processing involve the use of personally identifiable information, IP addresses or device identifiers.

The mobile gaming comparison

To test its central claim, NumberEight compared gender predictions across five mobile games with what the report describes as distinctly different audiences, measuring outputs from its own ID-less model against an IP based model supplied by what the report identifies only as a leading United States data provider. The company selected games where its own model identified a clear gender skew, defined as at least 65 percent of predicted users being either male or female, and then manually validated each result against the game's genre and broader audience signals before including it in the comparison.

The results, presented in a table in the report, show a consistent pattern. For Cooking Fever, the IP based model returned 48 percent male and 52 percent female, while the ID-less model returned 33 percent male and 67 percent female. For Fashion Battle, the IP based model returned 49 percent male and 51 percent female, compared with 18 percent male and 82 percent female under the ID-less model. For Golf Clash, the IP based model returned 51 percent male and 49 percent female, against 74 percent male and 26 percent female from the ID-less model. Nitro Nation, a racing game the report describes as expected to skew male, registered exactly 50 percent male and 50 percent female under the IP based model, compared with 74 percent male and 26 percent female under the ID-less model. Mini Football showed the same 50/50 split under the IP based model, against 65 percent male and 35 percent female from the ID-less model.

"For example, Fashion Battle was identified as 51% female by the IP-based model, compared with 82% female using NumberEight's ID-less model," the company stated in its press release. "Nitro Nation, a game expected to skew male, was identified as 50% male using the IP-based model, compared with 74% male using the ID-less model."

Across all five games, the report notes, the IP based model consistently returned near 50/50 distributions regardless of the app's actual audience composition, while the ID-less model reflected clear variation that the report describes as aligned with real user behavior. The report frames the IP based results as reflecting the general population's overall gender split rather than the composition of the specific audience playing a given title.

The connected television dataset

The report's second comparison drew on a considerably larger dataset: a connected television viewership file covering 7,013 shows, with an average of approximately 1,400 records per show, representing more than 3 million unique users. The dataset was supplied, according to the report, by what it describes as a mainstream CTV data broker, and while that partner shared television-level identifiers with NumberEight, those identifiers were used as household-level rather than individual-level markers - meaning the underlying limitation is structurally the same one affecting IP based targeting.

NumberEight compared the gender distributions in that household-level dataset against publicly available research on expected audience composition by genre, drawing on studies including work by Wühr, Lange and Schwarz published in 2017, alongside research from Stephen Follows, Filmgrail and Parrot Analytics. Those benchmarks, the report is careful to note, were themselves derived largely from surveys, observed consumption trends or genre-level analysis rather than direct measurement of individual viewers, so they represent directional expectations rather than ground truth.

According to the published benchmark data, genres including horror, war, science fiction, action, western and fantasy are expected to skew male, with published skew estimates ranging from 54 to 55 percent male in those categories. Romance, drama, musical and soap opera content is expected to skew female by comparable margins. Yet the household-level dataset told a different story. Across every genre examined, gender distribution remained tightly clustered within a narrow range of approximately 53 to 55 percent female, according to the report - a spread that showed no meaningful differentiation between genres widely understood to have opposite gender skews.

Age data followed the same pattern. The report compared age distributions in the household-level dataset against published research on expected audience age profiles across nine television genres, including animation, science fiction, news, documentary, drama, action, comedy, mystery and war. Genres such as animation, action and comedy are commonly expected to skew younger, while news, documentary, drama, mystery and war are expected to skew older. Despite those expected differences, age distribution in the household-level data was nearly identical across every genre, with the 18-24 age bracket consistently registering around 3 percent, the 25-34 bracket around 12 percent, and viewers aged 55 and older collectively accounting for roughly 48 to 50 percent regardless of content type.

"This indicates that household-level IDs are indeed not reliable for determining demographic data at the individual level," the report states. "Similar to the mobile analysis, the outputs collapse into near-identical distributions across genres, suggesting that the reported gender data is effectively random rather than reflective of real audience differences."

What company executives said

Emma Raz, Commercial Director at NumberEight, commented on the findings in the company's press release. "Demographic data is one of the most widely used inputs in advertising, but we rarely stop to question how that data is actually generated or how well it reflects the person consuming the content," Raz said. "What stood out in this analysis was how consistently some traditional approaches produced almost identical demographic profiles across content we know has very diverse audience profiles. The issue isn't simply whether you can match an identifier. It's whether the data attached to that identifier actually tells you something meaningful about the audience."

Abhishek Sen, CEO and Co-founder of NumberEight, framed the findings in terms of a broader industry shift. "The shift away from persistent identifiers creates an opportunity to rethink how we understand audiences," Sen said. "Sophisticated AI-led approaches that combine a variety of data sources from various channels enable us to create holistic predictive models that cover demographic predictions that are omnichannel at their core. Instead of treating identity as a proxy for understanding, we can focus on whether the signals actually tell us something meaningful about the audience."

A comparative scorecard across five dimensions

The report closes its analytical section with a summary table classifying all four methodologies - self-reported, IP based, third-party identifiers and ID-less - across five qualitative dimensions: relevance, privacy, scale, recency and bias. Self-reported data scores medium on relevance, high on privacy, low on scale, medium on recency and high on bias. IP based methods score low on relevance, low on privacy, medium on scale, medium on recency and medium on bias. Third-party identifiers score medium on relevance, low on privacy, medium on scale (with a footnote noting that scale continues to decrease due to signal loss from browser restrictions and shorter identifier lifespans), low on recency and low on bias. ID-less methods score high on relevance, very high on privacy, high on scale, high on recency and low on bias.

The report acknowledges a caveat that complicates a simple reading of ID-less methods as strictly superior: because these models make probabilistic inferences rather than relying on declared or deterministic data, they may occasionally be less accurate for any single specific individual than a deterministic identifier would be, even as they deliver more consistent and representative results in aggregate. The report frames this as a tradeoff rather than a limitation unique to ID-less approaches, noting that deterministic methods carry their own well-documented individual-level failure modes, from misreporting to fragmented identity graphs.

Context from a wider accuracy debate

The findings sit within a period of sustained scrutiny of identity based advertising data that has produced several independently sourced studies reaching broadly similar conclusions through different methodologies. Research commissioned by the Coalition for Innovative Media Measurement and Go Addressable and conducted by Truthset, released on November 5, 2025, examined nearly one billion IP address records from six major data providers and found that IP-to-postal address linkages achieved just 13 percent accuracy on average, while providers agreed on the same IP-to-household linkage only 6.4 percent of the time.

FreeWheel, the video advertising platform, drew on that same underlying dataset in February 2026 to argue that IP-based targeting can miss up to 87 percent of households, while disclosing that IP-derived addresses account for 17 percent of the targeting on its own platform. Separately, identity resolution vendor Adstra and media agency InterMedia Advertising reported in July 2026 that only 23 percent of residential IP addresses reached their intended geographic target, with device-level connected television identifiers proving 24 percentage points more consistent over time than IP addresses in the same analysis.

Each of these studies, it is worth noting, originates from an organization with a commercial stake in the identity resolution or verification market - a pattern that also applies to NumberEight, whose business model depends on advertisers adopting ID-less audience intelligence in place of the traditional methods its own report critiques. That commercial context does not by itself invalidate the underlying data, but it is a detail advertisers evaluating these findings would reasonably want to weigh alongside the numbers themselves.

The regulatory backdrop

The report situates its findings against a regulatory landscape that has grown steadily more restrictive toward persistent identifiers, even as the most consequential single policy reversal in that landscape - Google's decision to abandon full third-party cookie deprecation in Chrome - moved in the opposite direction. That reversal, confirmed in April 2025, arrived five days after Google was found to be a monopolist in the US Department of Justice's advertising technology antitrust case, timing PPC Land noted at the time carried its own significance. Chrome subsequently retired most of the associated Privacy Sandbox technologies in October 2025, roughly six months after the cookie reversal, leaving federated identity components such as FedCM as the most durable survivors of a five-year initiative that never achieved broad industry adoption.

Despite cookies remaining technically available, the broader trend the NumberEight report describes - declining reliability of persistent identifiers generally, whether through browser restriction, regulatory pressure or user opt-out - has continued largely unaffected by that single reversal. Apple's App Tracking Transparency framework, introduced in 2021, remains a significant constraint on mobile identifier availability, and a study published in July 2026 found the framework had shifted only 0.07 percentage points of affected apps toward paid download models, suggesting its primary effect has been on measurement capability rather than monetization strategy.

Insight for the marketing community

For media buyers and planners, the practical significance of this report lies less in any single statistic and more in what it implies about data that has long been treated as settled infrastructure. Demographic targeting - age and gender segments layered onto a media plan - is one of the most routine and least scrutinized inputs in digital advertising, purchased and applied without the kind of validation typically reserved for newer or more experimental data sources. If household-level identifiers genuinely produce demographic outputs that are, in the report's own words, effectively random at the individual level, then campaigns built on those segments may be systematically misallocating budget across age and gender lines without any visible signal that anything has gone wrong, because the data appears complete and precisely formatted even when it is not measuring what it claims to measure.

This concern connects directly to a broader pattern PPC Land has documented across the identity resolution market over the past year: study after study, from different commercial actors with different products to sell, converging on the same structural finding that IP and household-level data cannot reliably resolve to individuals. What distinguishes the NumberEight report from some of that prior coverage is its focus specifically on demographic attributes - age and gender - rather than on geographic or household matching accuracy alone, extending the accuracy concern from where an ad lands to who within that location actually sees it.

The comparative scorecard the report offers, weighing relevance, privacy, scale, recency and bias across four methodologies, gives planners a vocabulary for that conversation even if the specific scores reflect NumberEight's own analytical framework rather than an independently audited standard. Advertisers evaluating any demographic data source, whether IP based, ID-less or otherwise, would likely benefit from asking the same questions the report raises: how was this attribute inferred, against what benchmark was it validated, and does the resulting distribution show meaningful variation across content where variation would genuinely be expected.

Timeline

  • 2013 - Pew Research Center publishes its study of American teenagers, finding over a quarter of teen social media users had posted false demographic information
  • 2017 - Wühr, Lange and Schwarz publish research on gender stereotypes and actual movie genre preferences, later used in the report as a benchmark
  • 2021 - Apple brings App Tracking Transparency to iOS, restricting mobile advertising identifier availability
  • August 2024 - NumberEight observes a mobile gaming client where 5 percent of users self-reported ages of 99 or 100
  • 2024 - A survey cited in the report finds 21 percent of online dating service users admitted lying about their age on their profiles
  • November 5, 2025 - Truthset, commissioned by CIMM and Go Addressable, publishes a study finding just 13 percent accuracy in IP-to-postal linkages across nearly one billion records
  • April 2025 - Google confirms Chrome will retain third-party cookies, ending its planned deprecation
  • October 17, 2025 - Chrome retires most Privacy Sandbox advertising technologies
  • February 2026 - FreeWheel publishes analysis arguing IP-based targeting can miss up to 87 percent of households
  • July 2026 - Adstra and InterMedia Advertising report that only 23 percent of residential IP addresses reach their intended geographic target
  • July 2026 - Truthset's State of Data Accuracy 2026 report finds age segments collapse to 13 percent accuracy
  • September 2026 - NumberEight publishes "Demographic Data in Advertising: Traditional vs. ID-less Methods"
  • September 23, 2026 - NumberEight distributes its press release summarizing the report's findings

Summary

Who: NumberEight, an AI-native audience intelligence company based in London, alongside Commercial Director Emma Raz and CEO and Co-founder Abhishek Sen, who commented on the findings.

What: A 20-page research report comparing traditional demographic data methods - self-reported data, IP based targeting and third-party identifiers - against ID-less approaches that infer age and gender from behavioral and contextual signals rather than persistent identifiers. The report found that IP and household-level data produced near-uniform demographic distributions across mobile games and television genres with well-documented audience differences, while ID-less methods showed meaningful variation more consistent with expected audience composition.

When: NumberEight published the report in September 2026 and distributed its accompanying press release on September 23, 2026. The underlying mobile gaming comparison drew on data from August 2024, and the connected television analysis drew on a large third-party viewership dataset of unspecified collection date.

Where: NumberEight is headquartered in Norfolk, United Kingdom, according to the company's registration details, with the press release dated from London. The report's mobile and connected television data appear to originate primarily from the United States market, based on the games and benchmark studies cited.

Why: The report matters to the marketing community because demographic targeting is one of the most widely used and least scrutinized inputs in digital advertising. If IP and household-level identifiers cannot reliably distinguish individual-level age and gender within a shared household or network, as the report argues, advertisers relying on those signals may be systematically misallocating budget across demographic segments without any visible indication that the underlying data has failed.