The Vendor Numbers Everyone Cites and What They Actually Measure: eMarketer, GWI, and DataReportal Compared

Three numbers, three completely different evidentiary bases

A marketer preparing a deck can, within minutes, pull a "time spent with media" figure from eMarketer, a "social media penetration" figure from GWI, and a "number of internet users" figure from DataReportal, and stack all three into one slide as if they were reporting on the same underlying reality, measured to the same standard. They are not. Each of these three vendors sits on a different rung of an evidentiary ladder that runs from raw sampled survey data at one end to derived, non-transparent modeling at the other, and treating them as interchangeable is the single most common way vendor statistics get misused in this industry.

eMarketer: a modeled projection, not a raw survey

eMarketer's "time spent with media" figures are among the most quoted numbers in the advertising industry, and they are explicitly modeled estimates rather than raw survey output. The company's own 2025 report projects US adults will spend 12 hours and 44 minutes per day with media, broken down across categories such as traditional TV (2 hours 29 minutes), subscription streaming video (1 hour 58 minutes), and social networks (1 hour 31 minutes).

The headline number matters less than what eMarketer's own analysts read into the trend beneath it: total daily media time grew by only about two minutes in 2025, with roughly one more minute of growth projected for 2026. That is treated internally as a signal that attention is hitting a real ceiling, not a data artifact of the model. The underlying statistical methodology, exactly how raw inputs become that published daily-minutes figure, is largely proprietary and not fully disclosed in public materials, which is why this figure is tiered here as emerging rather than established: the trend it describes is credible and worth citing, but the precision of any single minute figure should be treated with real caution.

GWI: a recurring self-report panel, at global scale

GWI, formerly GlobalWebIndex, sits on a different evidentiary rung entirely. Its "Core" study interviews internet users across 48 or more countries in four waves per year, and sets quotas on age, gender, and education against national and international benchmarks for the 16 to 64 online population. A separate "brand and media" recontact module is completed by roughly half of respondents. This is self-report survey data at global scale, methodologically closer to a market-research panel than to a passive measurement instrument, and it is explicitly disclosed as such.

The distinction that matters here is evidentiary tier, not quality. GWI is transparent about being a non-probability access panel with stated quotas, which is a real methodological choice with known tradeoffs, self-selection into the panel and reliance on self-reported behavior among them. That transparency is precisely why GWI can be cited responsibly: a reader can see what kind of instrument produced the number and weigh it accordingly, which is not equally true of eMarketer's undisclosed modeling layer.

DataReportal: platform self-report, synthesized, with a stated margin of error

DataReportal's "Digital" series, compiled with We Are Social, produces the most widely cited "global internet and social media user" numbers in journalism and marketing decks. Its evidentiary basis is different again from both eMarketer and GWI: it is a secondary synthesis of platform-reported, advertiser-facing audience figures, the numbers platforms themselves publish to sell ad inventory, not a survey of actual people conducted by DataReportal.

To its credit, DataReportal's own methodology notes are unusually candid about this. They explicitly acknowledge that "user identities," meaning active accounts as platforms report them, are not the same thing as unique humans, and they estimate the resulting margin of error at roughly one to two percentage points when cross-checked against other third-party sources such as GWI, data.ai, Similarweb, and Semrush. The notes also flag that adoption data can lag by several months. Reading DataReportal's figures as "how many people," rather than "how many advertiser-facing accounts platforms report," is the single most common misreading of this particular source.

Why these three numbers cannot be added, averaged, or swapped

Line the three sources up and the incompatibility becomes obvious. eMarketer answers "how much daily time, modeled from proprietary inputs, does a population spend across media categories." GWI answers "what do a quota-sampled group of self-selected online panelists in 48-plus countries report about their own media behavior, four times a year." DataReportal answers "how many advertiser-facing accounts do platforms report existing in a given market, synthesized and cross-checked against other vendors." These are three different questions, answered with three different classes of evidence, and none of them can be validly summed, averaged, or substituted for one another, even when the resulting numbers happen to land in a similar range.

The practical failure mode this produces is common and specific: a deck that cites eMarketer's "time spent on social" next to GWI's "percent of population using social" next to DataReportal's "number of social media accounts" in a single chart, implying they triangulate on one truth, when in fact each number was produced by a different instrument answering a different question about a different unit of analysis, minutes, people, and accounts respectively.

The disciplined way to use vendor media-consumption data

All three are legitimate, widely used sources, and each is useful for the specific, narrower question it is actually built to answer. The discipline required is simple to state and easy to skip under deadline pressure: cite each figure with its source and its evidentiary tier attached, never combine figures from different vendors into one derived statistic without stating that they come from incompatible methods, and treat any modeled or undisclosed-methodology figure, eMarketer's daily-minutes precision being the clearest example here, as directional rather than exact.

This is the same discipline the field's more rigorous public instruments already practice. The Bureau of Labor Statistics' American Time Use Survey, the Pew Research Center's American Trends Panel, and Ofcom's Media Nations report all name their sample, disclose their weighting, and state their known limitations openly. Vendor "time spent" and "user count" figures deserve to be held to the same standard before they go into a business decision, because the underlying methods genuinely differ and a reader has no way to know that unless it is stated plainly.