Three platforms, three different answers, for the same decision. One says three seconds, one says six, one says five. Only one of the three cites any research, and when you open the citation, its number turns out to be the timing of its own skip button.

Underneath all of it sits the eight-second attention span, which has no primary source at all.

This page traces each number to where it actually comes from, gives the one open measurement of skip behaviour that exists, and says what to do instead.

What each platform actually publishes

Read the official pages rather than the summaries and the disagreement is immediate.

The largest social platform: three seconds. Its video best-practice guidance says to feature your brand and key message within the first three seconds, and that early branding increases recall and drives better results. It also advises using motion or a compelling visual in the first frame.

The short-video platform: six, and also three. Its creative best practices say to prioritise your hook in the first six seconds to boost engagement and increase watch time, and separately to introduce your content proposition in the first three seconds for better recall and awareness. Two thresholds on one page for two different jobs.

The video platform: five. Its creative playbook advises two or more shots in the first five seconds, and introducing your product or brand in the first five.

Two of the three cite nothing. No study, no sample, no period appears on either page in support of the number.

One of them cites three sources by name. And that changes how you should read all of this, so it is worth setting out properly.

The one platform that shows its work

The footnotes are on the last page of the playbook, and they are specific.

A neuroscience database. Declared as global, 2013 to 2016, n = 6,000 ads.

A creative drivers analysis. Run with a named research agency, global, fielded October to December 2018, n = 5,000 skippable video ads across 11 verticals.

A facial coding meta-analysis. Declared as global, 2018, drawn from a database of more than 15,000 ads, with facial coding results for all ads considered.

What that establishes. That this platform’s creative guidance rests on measurement it is willing to name, and that the other two platforms’ guidance does not. On provenance alone, this is not a fair fight.

What it does not establish. That five seconds is the right number, for a reason that only appears when you look at the product.

Five seconds is the skip button

This is the finding that reframes the whole convention.

Where the number comes from. The platform’s own advertising documentation states that after five seconds, the viewer has an option to skip the ad. Five seconds is when the button appears.

What follows. The creative rule is downstream of a product constraint. You have five seconds because that is how long the platform makes people stay, not because five seconds is a property of human attention.

Why the research does not close the gap. The cited studies measure what happens in ads that were themselves built to that constraint. Every ad in a skippable-format database was made by someone who knew about the five-second button.

Whether that makes it useless. No. It makes it a real operating constraint on that platform and a poor general principle. Five seconds is true where the button is, and undefined everywhere else.

The version to carry across platforms. Not “you have five seconds” but “you have until the viewer can leave, and each platform decides when that is.”

The opening-seconds thresholds published by three advertising platforms and the evidence each citesTable comparing the opening-seconds creative thresholds published by the three largest video advertising platforms and the evidence each attaches to its number. The largest social platform advises featuring the brand and key message within the first three seconds, stating that early branding increases recall and drives better results, and cites no study, no sample and no period; that three-second figure coincides exactly with the duration of the platform’s own video play billing metric, which makes the rule and the measurement mutually reinforcing rather than independent. The short-video platform publishes two thresholds on the same page, advising that the hook be prioritised in the first six seconds to boost engagement and increase watch time, and that the content proposition be introduced in the first three seconds for better recall and awareness, and cites no evidence for either. The video platform advises two or more shots in the first five seconds and introducing the product or brand within the first five seconds, and is the only one of the three to declare sources: a neuroscience database of six thousand advertisements collected globally between two thousand and thirteen and two thousand and sixteen, a creative drivers analysis of five thousand skippable video advertisements across eleven verticals fielded globally between October and December two thousand and eighteen, and a facial coding meta-analysis drawn from a database of more than fifteen thousand advertisements. However, that platform’s own advertising documentation states that after five seconds the viewer has an option to skip the advertisement, which means its five-second creative rule is the timing of its skip button rather than a finding about human attention, and the cited studies measure advertisements that were themselves built to that constraint.Three platforms, three numbers, one citationSame creative decision. No agreement on the threshold, and almost no evidence.PlatformSaysCitesWhere it comes fromLargest social3 sNothingIts own billing metric”feature your brand and key message within the first 3 seconds”Short-video6 s, and 3 sNothingUnstatedTwo thresholds on one page: hook at 6, proposition at 3Video platform5 sThree named studiesIts skip buttonn=6,000 ads (2013-16), n=5,000 ads (Oct-Dec 2018), >15,000 ads facial codingThe five seconds is not a fact about attentionThe platform’s own documentation: “After 5 seconds, the viewer has an option to skip the ad.”Every ad in the cited databases was made by someone who already knew about that button.The rule is real where the button is, and undefined everywhere else.
The only platform that sources its number turns out to be describing its own skip button. Source : Platform creative guidance, read directly (2026)

Three seconds is a billing metric

The other unsourced number has an equally revealing origin.

The rule. Brand and message within three seconds.

The metric. The same platform counts a video play when the video has played for three seconds, or for 97% of its length if shorter.

Why the coincidence matters. The creative rule and the counting rule are the same number. An advertiser optimising to the metric is optimising to the rule, and vice versa, so neither can serve as evidence for the other.

What that does to reported performance. Front-load the brand and your three-second play count improves, because the metric measures duration of play, not attention or recall. The measurement rewards the behaviour the guidance recommends, and the loop closes.

How to break the loop. Judge hooks on something the platform is not simultaneously selling you: the proportion of viewers still present when your actual message lands, or cost per meeting held.

The eight-second attention span does not exist

Almost every article on this subject leans on it. It has no primary source.

Where it appears. A 2015 consumer-insights report from a large technology company, on a page giving 12 seconds in 2000, 8 seconds in 2013, and 9 seconds for a goldfish.

What that page actually credits. The line under the figures reads “Source: Statistic brain”. The company did not produce the number. It reprinted a third party’s chart.

What the company’s own research was. Per its methodology pages: a gamified online survey of 2,000 Canadian respondents fielded in Q4 2014, plus electroencephalography on 112 Canadian respondents in December 2014. Neither measured a general attention span in seconds.

What happened when a broadcaster followed the chain. A public broadcaster’s statistics programme contacted the two organizations the aggregator had listed. Its report states that neither can find any record of research that backs up the stats, and that attempts to contact the aggregator itself came to nothing.

What an attention researcher said about the premise. A psychology lecturer quoted in the same report said she did not think it was true at all, because it is not something researchers interested in attention would try to measure and quantify that way, and that how much attention we apply is task-dependent.

And the goldfish. A fish behaviour professor quoted in the same piece noted that goldfish can perform all the kinds of learning described for mammals and birds, and have become a model system for studying learning and memory formation precisely because they have both.

Tracing the eight-second attention span statistic back through its citation chainDiagram following the eight-second attention span statistic back through each step of its citation chain to show that no primary research sits at the end of it. The statistic is normally attributed to a large technology company, on the strength of a two thousand and fifteen consumer insights report whose chart gives twelve seconds in the year two thousand, eight seconds in two thousand and thirteen, and nine seconds for a goldfish. The line printed under that chart credits a statistics aggregator, not the company, so the company reprinted a third party’s figure rather than producing it. The company’s own research, described in the same report’s methodology pages, consisted of a gamified online survey of two thousand Canadian respondents fielded in the fourth quarter of two thousand and fourteen together with electroencephalography on one hundred and twelve Canadian respondents in December two thousand and fourteen, neither of which measured a general attention span expressed in seconds. When a public broadcaster’s statistics programme investigated in March two thousand and seventeen, it contacted the two organizations that the aggregator had listed as its sources, a United States medical research body and a news agency, and reported that neither could find any record of research backing up the statistics, while attempts to contact the aggregator itself came to nothing. A psychology lecturer quoted in that report stated that she did not think the claim was true at all, because it is not something researchers interested in attention would try to measure and quantify in that way, and that the amount of attention applied is task-dependent. A fish behaviour professor quoted in the same report noted that goldfish can perform all the kinds of learning described for mammals and birds and have become a model system for studying learning and memory formation precisely because they possess both.Following the eight seconds back to nothingWhat everyone says”A major technology company found the human attention span is now 8 seconds.”↓What the 2015 report page actually saysThe chart carries the line “Source: Statistic brain”. The company reprinted it.↓What that aggregator listed as its sourcesA US medical research body, and a news agency.↓What happened when a broadcaster asked them, March 2017”neither can find any record of research that backs up the stats”The company’s actual research2,000 survey respondents, 112 EEG. Not an attention span.The goldfish halfA model system for studying memory, precisely because it has one.
A chart citation to a defunct aggregator, whose own listed sources say the research does not exist. Source : BBC, Busting the attention span myth (2017)

What is actually measured about skipping

One open academic study covers this properly, and its findings are more useful than any platform threshold.

The sample. HTTP traces from a university campus network, March to November 2014. 99,658 video-ad exhibitions matched out of 114,709 requests, across 5,667 unique ads. Desktop only, because the authors could not capture mobile.

How many people watched the whole thing. 29,442 exhibitions were streamed in full. In 29% of cases, users chose not to skip.

How fast the rest left. In more than a third of cases, 35%, users skipped in under six seconds, one second above the minimum. In only 25% of cases did users wait more than ten seconds before skipping.

How much of the ad played on average. A user skipped after 20% of the content had been exhibited, with a standard deviation of 19%.

What this does and does not support. It supports front-loading, clearly: a third of the audience is gone by the six-second mark. It does not support any specific threshold, because the distribution is continuous and its 2014 desktop context is not yours.

The honest way to use it. As the only open, sampled, dated measurement of skip behaviour anyone has published. If you cite one number in this whole subject, cite this one.

The one real standard says the opposite

There is an industry viewability standard for video, and it makes a point that cuts against the entire hook convention.

What it requires. For a video ad impression to count as viewable, two continuous seconds of the video must play, meeting the same pixel requirement as a viewable display ad.

The clause nobody quotes. The standard states explicitly that this required time is not necessarily the first two seconds of the video ad: any unduplicated content of the ad comprising two continuous seconds qualifies.

Why that is striking. The only standardised attention threshold in the industry deliberately refuses to privilege the opening.

What it implies for creative. Attention is not exclusively front-loaded, and a measurement body that spent years on this declined to say it was.

What it does not imply. That the opening does not matter. It matters for a simpler reason than any standard: people who leave early never see the rest.

Measured skip behaviour across ninety-nine thousand video advertisement exhibitionsChart presenting the measured skip behaviour reported by an open academic study of video advertising, based on HTTP traces captured on a university campus network between March and November two thousand and fourteen, covering ninety-nine thousand six hundred and fifty-eight video advertisement exhibitions matched out of one hundred and fourteen thousand seven hundred and nine requests, across five thousand six hundred and sixty-seven unique advertisements, on desktop only because the authors could not capture mobile traffic. Twenty-nine percent of exhibitions were streamed in full, meaning users chose not to skip in roughly three cases out of ten. Thirty-five percent of exhibitions were skipped in under six seconds, one second above the minimum permitted by the skip button. Only twenty-five percent of exhibitions ran more than ten seconds before the user skipped. Averaged across all exhibitions, a user skipped after twenty percent of the advertisement content had been shown, with a standard deviation of nineteen percent, meaning the variation between viewers is nearly as large as the average itself. The distribution supports front-loading a message, since roughly a third of the audience has left by the six-second mark, but it does not support any specific creative threshold, because the distribution is continuous rather than stepped and because its two thousand and fourteen desktop context differs from current mobile-dominated delivery. It remains the only open, sampled and dated measurement of skip behaviour published on this question.What actually happens: 99,658 ad exhibitionsCampus network traces, March to November 2014, 5,667 unique ads, desktop only.Watched in full, not skipped29%Skipped in under 6 seconds35%Waited more than 10 seconds25%Average skip point: 20% of the ad played, standard deviation 19%The spread is nearly as large as the average. There is no typical viewer here.What this licensesFront-loading. A third of the audience isgone before six seconds.What it does not licenseAny threshold. The distribution iscontinuous, and this is 2014 desktop.If you cite one number on this subject, cite this study. It is the only open one.
A third of the audience is gone at six seconds. That supports front-loading, not any particular threshold. Source : Arantes, Figueiredo, Almeida, ACM Web Science (2016)

In a feed there is no button at all

Every threshold in this article comes from a placement with a skip control. Most B2B video does not run in one.

What a skippable format is. An ad that plays before or during content the viewer came for, with a control that appears after a fixed delay. The five-second convention is a property of that control.

What a feed is. A surface the viewer is already scrolling. There is no gate, no fixed delay and no moment at which leaving becomes possible, because leaving was always possible.

Why importing the threshold breaks. In a feed, the equivalent of the skip button fires at zero seconds. A rule built around a five-second gate has nothing to attach to.

What replaces it. The scroll, which is continuous and starts immediately. That argues for front-loading harder than the skippable case does, not for a different number.

One placement mechanic that quietly ruins the measurement. On the professional network, videos under 30 seconds loop until there are at least 30 seconds of playback: a 10-second video loops three times, a 20-second video twice. Any completion or watch-time figure on a short video there is counting repeats.

The consequence for reporting. A 10-second ad and a 40-second ad on that platform are not measured on the same basis, so their completion rates cannot be compared. Compare them on cost per outcome instead.

Why opening-seconds thresholds drawn from skippable formats do not transfer to feed placementsDiagram contrasting skippable video advertising placements with feed placements to show why the opening-seconds thresholds published by platforms do not transfer between them. A skippable placement runs an advertisement before or during content the viewer came to watch, and presents a skip control after a fixed delay, typically five seconds; the five-second creative convention is a property of that control rather than a property of human attention, and it gives the advertiser a guaranteed minimum exposure. A feed placement has no gate at all: the viewer is already scrolling when the advertisement appears, leaving is possible from the first frame, and the effective equivalent of the skip button fires at zero seconds. Importing a five-second rule into a feed therefore attaches it to nothing, and the correct conclusion is not a different threshold but harder front-loading, since the departure risk begins immediately rather than after a delay. A separate mechanic on the professional network distorts any measurement taken in that environment: videos shorter than thirty seconds loop until at least thirty seconds of playback have accumulated, so a ten-second video loops three times and a twenty-second video loops twice, while videos longer than thirty seconds play once. Completion rates and watch-time figures for short videos on that platform therefore include repeated plays, which means a ten-second advertisement and a forty-second advertisement are not measured on the same basis and their completion rates cannot be compared with each other; the comparison should be made on cost per business outcome instead.The thresholds come from a placement you may not be usingSkippable in-streamPlays before content the viewer wanted.Skip control appears after a fixed delay.Guaranteed minimum exposure: 5 secondsFeedAppears while the viewer is already scrolling.No gate, no delay, no control to wait for.Guaranteed minimum exposure: noneThe answer is not a different number. It is harder front-loading, because departure starts at zero.And one mechanic that makes short-video metrics incomparableOn the professional network, videos under 30 seconds loop until 30 seconds of playback accumulate.A 10-second video loops three times. A 20-second video loops twice. Over 30 seconds plays once.So completion rates for short videos there are counting repeats. Compare on cost per outcome.
A rule built around a five-second gate has nothing to attach to in a feed, where leaving was always possible. Source : Platform placement documentation, read directly (2026)

What to do instead of picking a number

The disagreement between platforms is not a problem to resolve. It tells you the threshold is not the thing.

Front-load, for the reason that does not need a study. People who leave early never see the rest. That has been true of every medium and needs no neuroscience.

Stop citing seconds you cannot source. Saying “you have three seconds” in a brief imports a platform’s billing metric into your creative process. Say “open with the point” instead.

Define your own hold point. Identify the timestamp at which your actual message lands, and measure what proportion of viewers is still there. That number is specific to your creative and nobody is selling it to you.

Test openings against each other, not against a rule. Two versions of the same ad, different first four seconds, same everything else. That resolves the question for your audience in a week.

Judge on the downstream number. If a stronger opening produces better attention, it shows up in cost per meeting held. If it does not show up there, the improvement was in a metric, not in the business.

Do not let the constraint set the format. A skip button at five seconds is a fact about one placement. It should not decide how you write for a feed where there is no button at all.

Replacing a platform threshold with a hold point measured on the advertiser’s own creativeDiagram contrasting the conventional approach of writing creative to a platform’s published opening-seconds threshold with the alternative of measuring a hold point specific to the advertiser’s own creative. The conventional approach imports a number the advertiser cannot source: three seconds from one platform, which coincides with that platform’s own video play billing metric; five seconds from another, which is the timing of that platform’s skip button; six seconds from a third, which cites nothing at all. Writing to any of these numbers optimises toward a measurement the platform is simultaneously selling, and the disagreement between the three shows that none of them describes a property of human attention. The alternative is to identify the timestamp at which the advertisement’s actual message lands, measure what proportion of viewers remains present at that timestamp, and treat that proportion as the number to improve. This measure is specific to the creative in question, is not being marketed by any platform, and can be compared between two versions of the same advertisement that differ only in their opening seconds. The final judgement should be made on the downstream business measure, cost per meeting held, on the principle that an improvement in attention which does not appear in the downstream measure was an improvement in a metric rather than in the business. The industry viewability standard supports this reading, since it requires two continuous seconds of video play and states explicitly that these need not be the first two seconds of the advertisement.Stop picking a threshold. Measure a hold point.The convention”You have 3 seconds”= a billing metric”You have 5 seconds”= a skip button”You have 6 seconds” = nothing citedYour hold point1. Find the second your message lands2. Measure who is still there3. Improve that proportionSpecific to your creative. Nobody is selling it to you.Then judge on cost per meeting heldIf a better opening does not show up there, the improvement was in a metric, not in the business.The one industry standard on this refuses to privilege the openingTwo continuous seconds of play, and “this required time is not necessarily thefirst 2 seconds of the video ad”.
The hold point is specific to your creative, and no platform is simultaneously selling it to you. Source : MRC Viewable Ad Impression Measurement Guidelines (2015)

Where to go next

You want to know what a view actually counts. Hold rate and video metrics.

You are deciding how long the ad should be. Video ad length.

You are choosing between video and static. Video or static ads.

Your creative has been running a long time. Creative fatigue metrics.

You want the buying mechanics underneath. Media buying explained.

Your click-through rate is the number in dispute. Click-through rate benchmarks.

In short

  • Three platforms, three thresholds: 3 seconds, 5 seconds, and 6 seconds for the same decision.
  • Two of the three cite nothing at all. No study, no sample, no period.
  • The one that cites research names three sources: 6,000 ads, 5,000 ads across 11 verticals, and a facial coding database above 15,000.
  • Its five seconds is its skip button. The creative rule is downstream of a product constraint, and the cited ads were all built to it.
  • The other’s three seconds is its billing metric. Rule and measurement are the same number, so neither can evidence the other.
  • The eight-second attention span has no source. A 2015 chart credited to a statistics aggregator whose own listed sources, when contacted, could find no record of the research.
  • The one open measurement: 29% of ads watched in full, 35% skipped in under six seconds, average skip after 20% of the ad.
  • The industry viewability standard explicitly does not require the first two seconds. Any unduplicated two continuous seconds qualifies.

Front-load because people leave, not because a platform said three. Book a diagnostic, or see how we approach B2B paid acquisition.