A study from Lumen Research, Amazon Ads and Cint puts a number on how much attention an advertising campaign must accumulate before it moves purchase intent: a median of 47 seconds, summed across multiple impressions and media. The research, drawn from 270 cross-media campaigns, drew wider attention today after Peter Buckley, Connection Planning Director at Meta, summarised its findings on LinkedIn, arguing that raw exposure counts barely predict brand lift while cumulative attention predicted it about eight times better.

In Short

Researchers looked at 270 ad campaigns and measured how many seconds people actually spent looking at the ads, added up across every time they saw them. They found that, in the middle case, it took about 47 seconds of looking in total before people said they were more likely to buy. That means counting how many times an ad was shown matters less than counting whether anyone looked, though the full report behind the numbers is only available through Amazon.

What the study says, and where it says it

The primary public source is thin. Lumen's landing page for the work, titled "Cumulative attention and brand lift," runs to three short paragraphs and a download form. It carries no publication date. According to Lumen, the study "examines how Cumulative Attention across multiple ad impressions contributes to brand lift," drawing on "270 cross-media campaigns" to quantify "the attention needed to lift ad recall, awareness, consideration, favourability and purchase intent."

Five metrics are covered, running from the top of the funnel to the bottom. The page names a threshold for only one of them. "The study found a median of 47 seconds of Cumulative Attention is needed for purchase intent," according to Lumen. The thresholds for the other four metrics, if the study produced them, are not on the public page.

Two further claims appear in the page copy. "Attention builds up across impressions, and each medium contributes a different amount," according to Lumen. The page does not say which media were included, how much each contributed, or in which markets the 270 campaigns ran. It closes by offering "shareable slides" in exchange for a name, email address, company and country, and states that the full white paper is available only from an Amazon representative.

There is a small inconsistency in the campaign count. The body text of the Lumen page says 270 cross-media campaigns. The page's meta description, the summary text served to search engines, refers to "270+ campaigns." The difference is minor, but it means the exact sample size is not fixed by Lumen's own materials.

The slide deck itself was not available for this article. A saved copy of the Google Drive page holding the file "Amazon Lumen Cint_Cumulative Attention_downloadable slides.pdf" contains only the Drive interface, not the slide content. Every figure attributed to the study below therefore comes from the Lumen page or from Buckley's LinkedIn post, and the distinction matters.

What a 47-second median does and does not tell buyers

Forty-seven seconds is a lot of looking. Most ad formats deliver a fraction of that per impression; Lumen's own earlier work with Snapchat, covered by PPC Land in March 2024, linked one second of visual attention per impression to a 20x increase in ad recall. The new figure only becomes reachable if seconds are summed across many exposures and several media. That is the point of the word "cumulative."

What the median measures is less clear from the public material. A median of 47 seconds could describe the attention needed per respondent, per campaign, or at some aggregate level across a campaign's audience. Mike Follett, chief executive of Lumen Research, offered one clue in a comment under Buckley's post, crediting Cint's "cross-media, respondent level BLS" as "essential for this analysis." BLS here refers to brand lift studies. Respondent-level data would allow attention to be summed per person and matched to that person's survey answers, which suggests the 47 seconds is a per-person figure. That is an inference; the Lumen page does not state it.

A median also hides the spread. Half the campaigns needed less than 47 seconds to shift purchase intent and half needed more, and without the distribution there is no way to tell how tight that spread is. For a planner trying to set a target, a median with a narrow range is a usable benchmark. A median with a wide range is closer to a reminder that attention matters.

The method used to measure attention in the 270 campaigns is not described on the page either. Lumen's business rests on eye-tracking panels and models trained on them. According to Cint, announcing a partnership with Lumen on April 16, 2026, the combined offering uses Lumen's "proprietary real-world eye-tracking data from 50+ countries," together with attention predictions and brand-specific attention models. Whether the 47 seconds is observed attention, predicted attention or a blend cannot be determined from the files.

Buckley's reading: frequency counted properly

Most of the detailed findings circulating today come not from Lumen but from Buckley, who works at Meta and has written about attention research for years. His post opens with a provocation. "Forget frequency," Buckley wrote. "A new study from Amazon and Lumen, covering 270 campaigns, found raw exposure counts barely predict brand lift. Cumulative attention, the total seconds people actually spent looking across all your ads, predicted it about eight times better."

The "eight times better" figure does not appear on the Lumen page. Buckley does not say what statistic underlies it, whether variance explained, correlation strength or something else. Nor does the post define the brand lift measure it was tested against. It is a vendor-study figure relayed by a third party, with the method sitting in a gated document.

Buckley then lists three findings he says stood out.

Order matters

"Seeing streaming TV then online ads gave 30% more brand lift than the reverse," according to Buckley. He ties this to the frequency illusion: "Once someone has properly noticed you, every glimpse after that lands harder." The 30% figure is not on the Lumen page, and the post does not say which brand metric it applies to or how many campaigns ran in each sequence.

Buckley describes the frequency illusion as "a bias named over 30 years ago." That dating is loose. The term "frequency illusion" is generally credited to the linguist Arnold Zwicky, who coined it in 2005. The older popular label for the same experience, the Baader-Meinhof phenomenon, dates to 1994. The underlying idea is not affected, but the "over 30 years" description fits the older nickname rather than the name Buckley uses.

The sequencing claim has a direct bearing on how Amazon sells its inventory. Amazon offers Prime Video ads alongside display and online video through Amazon DSP. If streaming exposure first and cheaper online formats second performs better than the reverse, the order of a campaign's flighting becomes a planning variable rather than an outcome of auction dynamics. Whether Amazon's tools can enforce that order at the individual level is a separate question the files do not address.

Attention is not a silver bullet

The second finding cuts against the headline. "Around a quarter of the highest-attention campaigns showed no lift in purchase intent," according to Buckley. "At least a quarter of the lowest-attention ones beat the typical high-attention campaign on awareness."

That is a substantial caveat. If a quarter of high-attention campaigns produced nothing on purchase intent, attention is at best a probabilistic input. Buckley attributes the rest to creative and to prior brand familiarity: "If they know and use your brand, they notice it more." Neither explanation is tested in the material available.

It is secretly a study about frequency

The third point reframes the whole exercise. "No single ad got anyone over the attention threshold," according to Buckley. "Cumulative attention is just frequency counted properly: every exposure, weighted by whether anyone looked." He ends on a reversal of his opening line: "So don't forget frequency. It matters more than ever. But the first exposure often shapes what all the others are worth."

A discrepancy over who did the study

Buckley's post describes the work as "a new study from Amazon and Lumen," and his own follow-up comment labels it an "Amazon Ads + Lumen Research study." Cint does not appear. Follett corrected that within hours: "don't forget Patrick Comer and the good people at Cint whose cross-media, respondent level BLS is essential for this analysis." The Lumen page itself names all three companies, with Lumen as lead author "in partnership with Amazon Ads and Cint."

The omission matters for interpreting the results. Attention data alone cannot show brand lift. The lift side of the equation comes from surveys of exposed and unexposed people, and in this case those surveys appear to have come from Cint. Cint's role in Amazon measurement is not new; its Lucid Measurement product has long run brand lift studies for platforms, and PPC Land reported on June 17, 2026, that Cint merged brand and sales lift into one live dashboard in a US open beta.

The Meta parallel, and a figure that has shifted

Buckley connects the Amazon work to Meta's own research. "It echoes what we've found at Meta, where we call it aggregate attention: multiple short views drive on average 21% more sales lift than one long continuous view," he wrote.

A version of that 21% figure has circulated before. At IAB Europe's Virtual Programmatic Day on July 3, 2025, a panellist cited research from Meta and Playground xyz finding that attention split into shorter bursts was 21% more effective than one continuous block of the same total time. That account described the result as "more effective." Buckley's post today describes it as "21% more sales lift." Both may refer to the same underlying study, but the outcome metric is stated differently, and neither source in the files provides the original methodology.

The Meta finding and the Amazon finding point the same way: many short exposures, summed, can match or beat fewer long ones. Both conclusions also suit the companies involved. Meta sells short-form, feed-based placements in which individual views are brief. Amazon sells a mix that runs from 30-second streaming spots to display banners. A framework in which every glance adds to a cumulative total makes inventory with low per-impression attention easier to justify.

Where this sits in Buckley's earlier arguments

Buckley has been a recurring voice on the limits of attention as a buying metric. In July 2024, commenting on a Kantar study of 873 multi-media campaigns worth more than $3.2 billion in media spend, he argued that some advertisers were buying too much attention. That Kantar research found that overall channel attention levels did not directly correlate with channel-level brand contributions or cost-effectiveness.

Today's post is consistent with that position rather than a reversal. Buckley does not argue for buying the highest-attention placement. He argues for counting attention across all placements, which is a frequency argument dressed in attention units.

Why the timing matters for attention buying

The study arrives as attention metrics move from post-campaign reporting into bidding. Adelaide brought attention pre-bid targeting to Amazon DSP on June 10, 2026, letting buyers filter display, online video and streaming TV inventory by attention tier. Google followed by putting Adelaide attention scores into DV360 custom bidding from mid-July 2026.

Those tools score impressions one at a time. The Lumen study's central claim, that "no single ad got anyone over the attention threshold" in Buckley's words, sits awkwardly with per-impression optimisation. A bidder that maximises attention per impression is not the same as a planner that manages attention per person across a campaign. The first is available in DSPs today; the second requires person-level exposure data across media, which is exactly what Amazon holds for its own properties.

Lumen itself has been expanding the inventory it measures. It brought attention measurement to Netflix ads in five European markets in March 2026, and Teads took exclusive access to Lumen's CTV HomeScreen attention measurement globally in May 2026. Lumen also supplies the eye-tracking data behind IAS's Quality Attention product, launched in January 2024. The cumulative framing gives Lumen a way to aggregate across all of those surfaces.

Other measurement firms have published single-number attention benchmarks this year. TVision, now owned by Viant, said in August 2026 that viewers spend more than 200 minutes a day with video, of which only nine minutes are attentive ad time. IAS and Stellantis reported in June 2026 on an attention programme measuring more than 13 billion impressions across 19 countries. None of these figures is directly comparable to the 47-second median, because each vendor defines and measures attention differently. That lack of a shared standard was the central complaint of the July 2025 IAB Europe panel.

How it fits with the frequency debate

The frequency question has been reopened from several directions in 2026. In February, Google researchers questioned advertising frequency orthodoxy, citing YouTube Brand Lift data in which second impacts indexed at 1.5 to 1.8 against a first impact, below the 2x that would mean equal value. Their conclusion was that the first impact is worth the most.

Buckley's closing line lands near the same place: "the first exposure often shapes what all the others are worth." The difference is that the Google analysis concerns diminishing returns from each additional exposure, while the cumulative attention work concerns whether exposures that people actually looked at add up to a threshold. The two can coexist. A planner can accept that each successive impression is worth less and still need enough of them, properly noticed, to cross a line. That bears on how reach and frequency goals are set.

Amazon has its own tools on the frequency side. Amazon Marketing Cloud launched Optimal Frequency in October 2024, recommending daily frequency based on cost and incremental conversions. That tool counts impressions and conversions; it does not count seconds of attention. Whether Amazon intends to bring a cumulative attention measure into AMC or Amazon DSP planning is not stated in any of the files.

Amazon's earlier brand lift findings

Amazon has published cross-format brand lift research before. In a meta-analysis of more than 3,000 brand lift measurements from Q4 2022 to Q3 2024, Amazon reported that combining Amazon DSP display ads with video, including streaming TV, produced a 2.2x increase in brand awareness compared with standalone video. That study measured format combinations. The cumulative attention study, as described, goes a step further by trying to explain why combinations work: each medium adds a different number of attentive seconds to a running total.

Comcast made a related claim this month, reporting that TV plus streaming campaigns delivered twice the incremental return of single-tactic buys in January to June 2026 data. Sellers of multi-format inventory have an obvious interest in research showing that multi-format campaigns work better, and the Amazon-Lumen-Cint study belongs to the same genre: the media owner co-produced it and controls access to the full paper.

What remains unknown

The gaps are considerable for a study cited this widely. The public material does not give the markets, the date range of the 270 campaigns, the media included, the ad recall, awareness, consideration or favourability thresholds, the statistic behind "eight times better," the sample behind the 30% sequencing figure, or the distribution around the 47-second median. All campaigns appear to have run with Amazon as a media partner, which limits how far the findings extend to campaigns without Amazon inventory.

One comment under Buckley's post raised a different objection. Rares Rusu, chief executive of 40Folds, asked whether the results simply show that creative quality drives outcomes, since exposure weighted by whether people look could be a proxy for consistently strong creative. The material available does not answer that question either.

The full white paper, according to Lumen, is available through Amazon representatives. Until it is published openly, the 47-second figure is a vendor-supplied benchmark whose construction cannot be independently checked.

Timeline

Summary

Who: Lumen Research, an attention measurement company led by chief executive Mike Follett, with Amazon Ads and survey and measurement company Cint. The findings were publicised by Peter Buckley, Connection Planning Director at Meta.

What: A study of 270 cross-media campaigns found that a median of 47 seconds of cumulative attention, summed across multiple ad impressions, was needed to lift purchase intent. Buckley's summary adds that cumulative attention predicted brand lift about eight times better than exposure counts, that streaming TV followed by online ads produced 30% more brand lift than the reverse order, and that about a quarter of the highest-attention campaigns showed no purchase intent lift. Those additional figures are not on Lumen's public page.

When: Buckley's LinkedIn post appeared today, October 13, 2026. The Lumen landing page carries no publication date.

Where: The findings are published on a Lumen Research landing page with gated slides; the full white paper is available only through Amazon representatives. The campaigns' markets are not disclosed.

Why: Attention metrics are moving into programmatic bidding in Amazon DSP and DV360, where they score individual impressions. The study argues that outcomes depend on attention accumulated per person across many impressions and media, a frequency question that per-impression tools do not answer, and one that favours media owners like Amazon who hold cross-media exposure data.