FreeWheel, the Comcast-owned television advertising technology company, published research on September 15, 2026 showing that the same streaming advertisement was recalled unprompted by 41% of viewers when it ran in a lightly loaded programme and by 26% when it ran alongside nine minutes of commercials per half hour. The study, conducted with MediaScience across 436 participants, also found that a single streaming advertisement produced 33% unaided brand recall against 12% for a social media placement.

In Short

FreeWheel ran a test with a research firm in which 436 people watched ads either inside a streaming TV show or while scrolling social media apps, and it tracked their eyes and asked them afterwards which brands they remembered. People remembered the TV ads far more often, and they remembered them even better when the show carried fewer minutes of ads. The company behind the test sells streaming ad technology, so the numbers come from an interested party and describe a controlled experiment rather than live campaigns.

What the study measured

The report, titled "Why Brands Need TV: How Streaming Captures Viewer Attention That Drives Performance," runs to nine pages. According to FreeWheel, its research partner MediaScience recruited 436 participants across both in-lab and in-home settings and exposed them to 30-second advertisements in two broad environments.

The social condition was itself split. One group scrolled through feeds on Instagram and/or YouTube, according to the deck. A second group scrolled through a vertical video feed of YouTube Shorts and/or Facebook Reels. Advertisements were embedded throughout both feeds.

The streaming condition worked differently. Participants chose a 30-minute programme from several options inside an unbranded streaming interface, with advertisements integrated throughout. No named service, publisher or app appears in the test design, which means the results describe a generic streaming environment rather than any specific platform on which FreeWheel serves advertising.

Brands were balanced. According to FreeWheel, viewers saw an even mix of well-known and lesser-known brands. Two instruments captured the response: biometrics including eye tracking, and a post-exposure survey. The deck states that all comparisons it shows are statistically significant at p<0.05, adding that "the observed differences are likely not due to chance but are meaningful and reliable."

That significance threshold is the only statistical detail disclosed. The deck does not publish per-condition sample sizes, confidence intervals, fieldwork dates, the split between lab and home participants, the number of advertisements each participant saw, or the brands and categories involved.

Eye tracking: 94% against 75%

The first set of results concerns visual attention. According to FreeWheel, 94% of participants in the streaming environment looked at the advertisement, compared with 75% in the social environment. That 19-point gap is expressed in the report as a relative increase of 26%.

Dwell time followed the same direction. Streaming viewers spent 71% of the advertisement's running time gazing at it, against 52% for social viewers, which the deck presents as 36% more time watching. Both figures come from eye tracking, a method that records where eyes fall on a screen. It does not record whether the content registered.

The deck draws a commercial conclusion from these numbers, stating that the capacity to command attention shows why premium environments deliver more effective impressions. That link from gaze to effectiveness is contested terrain. When the Media Rating Council and the IAB finalised attention measurement guidelines in November 2025, they cautioned against treating attention as a direct measure of outcomes. The same guidelines set minimum transparency requirements for vendors, and the FreeWheel deck discloses less about its eye-tracking method than those requirements contemplate - no definition of what counted as a "look," no fixation threshold, and no screen size for either condition.

Recall across one and two exposures

The second set of findings moves from the eye to memory, and the gaps widen considerably. According to FreeWheel, streaming generates "more memory per impression than social media."

The numbers behind that claim are specific. After one exposure, unaided brand recall - the share of participants naming a brand without being prompted - reached 33% for streaming and 12% for social. The ratio is 2.75, which the report rounds to 2.7X and which FreeWheel's covering email to press, distributed on September 17, 2026, used as its headline statistic.

A second exposure raised recall in both environments. Two social advertisements produced 24% recall, exactly double the single-exposure figure. Two streaming advertisements produced 53%, an increase of 20 points over one. The deck puts the two-exposure gap between streaming and social at 2.2X.

The mixed-sequence result

The report's most practically framed finding concerns combinations. According to FreeWheel, participants who saw one streaming advertisement and one social advertisement recalled the brand 42% of the time - 1.8 times the 24% recorded for two social advertisements.

Read against the rest of the chart, that figure carries a second implication the deck does not spell out. A mixed pair (42%) sits 11 points below two streaming exposures (53%). Adding a social impression to a streaming one lifted recall by nine points over a single streaming exposure; adding a second streaming impression lifted it by 20. The report's own data therefore places the mixed sequence between the two pure sequences, not above them. The deck does not state whether the streaming or the social advertisement came first in the mixed condition, or whether the order was randomised.

Break length and the 58% gap

The finding with the clearest operational consequence for publishers sits on page six. According to FreeWheel, when audiences saw a streaming advertisement within a lighter ad load environment, unaided recall was 58% higher than when they saw that same advertisement within a heavier one.

The two conditions were defined precisely. The heavier environment carried nine minutes of advertising per 30 minutes of programming - 30% of the running time, or the equivalent of 18 minutes an hour. The lighter environment carried three minutes and 45 seconds per 30 minutes, which is 12.5% of running time or seven and a half minutes an hour. The difference between them is five minutes and 15 seconds of commercial time per half hour. Recall was 26% in the heavier setting and 41% in the lighter one.

For context, the heavier condition exceeds what American audiences say they tolerate. A Magnite survey released on August 20, 2026 found that live streamers accept an average of 8.7 minutes of ads per hour, roughly half the rate tested in FreeWheel's heavy condition. The lighter condition sits just below that stated tolerance level.

The 26% and 41% figures raise a question the deck leaves open. Neither matches the 33% single-exposure streaming recall reported on the previous page. The report does not say whether the 33% pools participants across both load conditions, whether the load test used a separate cell of participants, or how many exposures the load comparison involved.

FreeWheel frames the result within its own earlier work. The deck states: "Previous research found that shorter ad pods can deliver up to 2X higher brand impact, leading innovative platforms to create premium curated packages around lighter ad loads to give advertisers opportunities to maximize engagement and recall." That previous research is not named in the deck, though it links to a FreeWheel report. The company and MediaScience published a study of more than 700 viewers in January 2024 which found that breaks of two minutes or less were preferred by viewers and associated with improved brand recall.

The ad load question has been running through the streaming market from several directions. YouTube, in August 2024, cited user research showing 79% of viewers preferred advertisements grouped together when it expanded fewer, longer breaks on connected televisions - a finding about break frequency rather than total volume. A consumer survey published by All About Cookies on April 22, 2026 found that 76% of U.S. streaming subscribers believed platforms carried too many ads, with only 24% saying they paid attention during breaks.

Lower-funnel claims rest on stated intent

Page seven extends the argument beyond memory. According to FreeWheel, compared with two social exposures, adding a streaming exposure to a social one made participants 2.1 times more likely to visit the brand website, 2.1 times more likely to look for more information, and 2.2 times more likely to keep an eye out for a deal or offering.

The deck adds a sentence on the counterfactual: "While an additional social impression does not deliver a significant lower-funnel lift, the addition of a streaming impression significantly boosts outcomes."

Two limitations apply. The first is that these are survey responses from a post-exposure questionnaire, not observed visits or searches. Participants reported what they were likely to do; the study did not track whether they did it. The second is that the deck publishes only the multipliers. With no base rates, there is no way to tell whether 2.1 times more likely means a move from 2% to 4.2% or from 20% to 42%. Measurement practitioners treat stated intent and incrementality as separate questions for exactly this reason: one captures what people say after exposure, the other what they do compared with an unexposed group.

What the deck leaves out

Several features of the design shape how far the numbers travel beyond the lab.

The two environments differ in more than advertising. Streaming participants chose their own 30-minute programme and watched it in what is typically a lean-back posture. Social participants scrolled feeds, where the mechanism of the platform is to move past content. The deck does not say whether social participants could scroll past the embedded 30-second units, a behaviour the feed format ordinarily allows, nor whether 30 seconds matches the length of advertisements normally served in Shorts and Reels. Nor does it say whether the social feeds were viewed on phones and the streaming programme on television screens. Each of those choices bears on eye-tracking and recall results independently of the channel.

The report also does not account for co-viewing. The deck does not say whether participants watched alone or with others, although trade bodies have elsewhere cited shared living-room viewing as a driver of recall.

Linear television is absent despite the title. The report is called "Why Brands Need TV," but its tested environment is streaming alone. No broadcast or cable condition appears.

The deck is marked "Confidential" in the footer of every page. It was nonetheless distributed by email to press subscribers, and the email's footnote points to a public report page on FreeWheel's website.

The closing slide contains recommendations. According to FreeWheel, advertisers can prioritise streaming for effective impressions, pair other channels such as social with streaming, and work with a platform that uses robust measurement to prove how streaming impressions deliver results. The deck adds that "as the ability to both buy and measure becomes easier, it's the most effective channel to deliver performance and measurable outcomes." That last phrase is a claim the study itself did not test: no cost, price or conversion data appear anywhere in the nine pages.

Who paid, and who sells

FreeWheel describes its technology as built for streaming and TV advertising, connecting buyers and sellers directly. Its position in the market is substantial. A PPC Land review of the video ad server landscape in November 2025 identified its platform among the leading systems for major broadcast and streaming publishers. Since then the company has placed an MCP server into premium video deal flow in March 2026, carried Amazon's retail signals into programmatic guaranteed deals from June 19, 2026, and in July 2026 added series-level content reporting for seven publishers inside Buyer Cloud. Its parent's advertising relationships are also widening; the NBCUniversal and YouTube agreement of July 27, 2026 deepened advertising technology work through FreeWheel.

A study concluding that streaming outperforms social media, sponsored by a company whose revenue depends on streaming advertising, is not by that fact wrong. Eye tracking and unaided recall are established instruments, and MediaScience conducts research for multiple clients. The company's research has, however, become a recurring input in industry advocacy. On August 28, 2026, the Video Advertising Bureau headlined a 94% completion rate sourced from 2023 FreeWheel data as a live planning benchmark. And on September 5, 2026, FreeWheel published a separate survey in which 48% of buyers overrated viewer dislike of AI-influenced advertising, making the streaming study its second research release in ten days.

MediaScience's name also appears across competing vendors' claims. LG Ad Solutions and MediaScience research reported that interactive CTV ads generated 138% higher unaided brand recall than standard units. Recall claims across CTV vendors vary widely and rarely share methodology; a PPC Land analysis of Disney's interactive formats noted that such studies seldom publish control group construction or exposure definitions, which makes cross-vendor comparison unreliable.

At 436 participants, the FreeWheel sample is larger than some comparable attention work. Teads' research with MediaMento Institute, which put HomeScreen ad attention at 48%, 16 percentage points above YouTube skippable pre-roll, used 100 Smart TV viewers. The sample is smaller than FreeWheel's own January 2024 pod study of more than 700.

Why this matters for buyers and publishers

The streaming versus social comparison will draw the headlines, but it answers a question most planning teams have already priced. Television screen attention has carried a premium over feed-based video for years, and the Video Advertising Bureau's work with TVision, released on February 24, 2026, made a related case against YouTube on connected televisions.

The ad load result is less settled and more consequential. It is a finding from a company that sells ad serving to publishers, telling those publishers that trimming breaks from nine minutes per half hour to three minutes and 45 seconds lifts advertisers' recall by 58%. Publishers monetise minutes. The same reduction removes more than 58% of sellable time, which would require prices on the remaining slots to rise sharply to hold revenue flat. The deck's reference to "premium curated packages around lighter ad loads" describes one commercial route: selling lighter breaks as a distinct, higher-priced product rather than lowering load across the board.

For buyers, the 26% versus 41% gap indicates that break density inside a streaming placement may matter as much as the choice of channel itself. Heavy-load streaming recall, at 26%, sits closer to two social exposures (24%) than to a single lightly loaded streaming exposure (41%). Whether that finding holds outside a single 30-minute lab programme, across genres, and with the co-viewing and distraction of real homes, is a question the nine pages do not answer.

Timeline

Summary

Who: FreeWheel, the Comcast-owned streaming and television advertising technology company, working with research firm MediaScience. The study involved 436 participants and targets advertisers, agencies and streaming publishers.

What: A lab and in-home experiment comparing 30-second advertisements in social media feeds (Instagram, YouTube, YouTube Shorts, Facebook Reels) with advertisements in a 30-minute programme inside an unbranded streaming interface, measured through eye tracking and a post-exposure survey. Streaming produced 94% ad looking against 75%, 33% unaided recall against 12% after one exposure, and 42% recall for a streaming-plus-social pair against 24% for two social exposures. Within streaming, recall was 41% under three minutes and 45 seconds of ads per half hour and 26% under nine minutes.

When: The report is dated September 15, 2026, and FreeWheel distributed it by email on September 17, 2026. Fieldwork dates are not disclosed.

Where: The deck does not state a market. The platforms tested and the company's New York mailing address point to the United States, and participants were tested both in a lab and in their homes.

Why: FreeWheel presents the findings as evidence that streaming delivers more attention and memory per impression than social media and that lighter commercial breaks raise recall. The results matter because break density is a revenue lever for publishers, and because the study, like much attention research, is published by a company with a commercial stake in its conclusion and discloses limited methodological detail.