Two of the four largest video advertising platforms refuse to recommend a duration at all. One of the two that does cites nothing. The other gives its reason openly, and the reason is placement eligibility, not attention.

That asymmetry is the most useful thing on this subject. The platforms holding the deepest video measurement in the industry have declined to publish a number, and the ones publishing numbers are not showing their work.

This page sets out what each platform actually says, what constrains your length in practice, and how to decide.

What each platform publishes

Read the official pages and the picture is not a consensus with edge cases. It is a genuine split.

The largest social platform recommends, without evidence. Its guidance says feed video ads can run up to 60 minutes, but that shorter videos under 15 seconds tend to perform best in feed, and that stories ads perform best under 10 seconds. Elsewhere it says shorter videos of 6 to 15 seconds are more effective, and that shorter is better but context matters. No study, no sample, no period appears alongside any of it.

The professional network recommends, and says why. It recommends uploading videos 15 to 30 seconds long, so they can qualify for all possible placements across its feed and its audience network. That is a stated, checkable, technical reason. It is also not a claim about how long people watch.

The video platform declines. Its creative effectiveness playbook, the flagship document on what makes video advertising work, contains zero occurrences of the words length, duration or shorter, and no reference to 15 or 30 second formats. It publishes format maxima only.

The short-video platform declines too. Its creative best practices page gives hook timing and nothing about overall duration. Its in-feed specification page states a ten minute limit for standard ads and no restriction for the boosted-post format, with no recommendation attached.

What that pattern means. The two organizations with the most video advertising measurement in existence have looked at this question and published nothing. That is a finding, and it should lower your confidence in anyone who has published a number.

Video advertisement duration recommendations published by four platforms and the evidence each attachesTable comparing what the four largest video advertising platforms publish about video advertisement duration, the reason each gives, and the evidence each attaches. The largest social platform recommends videos under fifteen seconds in feed and under ten seconds in stories, and states elsewhere that videos of six to fifteen seconds are more effective and that shorter is better while context matters; it gives no reason and cites no study, no sample size and no period. The professional network recommends videos of fifteen to thirty seconds and states its reason explicitly, which is that videos in that range qualify for all possible placements across its feed and its audience network, making the recommendation an eligibility rule rather than a claim about viewer attention. The video platform publishes no duration recommendation at all: its creative effectiveness playbook, which is its flagship document on what makes video advertising work, contains zero occurrences of the words length, duration or shorter, and it publishes only format maxima, namely sixty seconds for non-skippable in-stream, five to six seconds for bumper advertisements and under three minutes recommended for skippable in-stream. The short-video platform also publishes no duration recommendation: its creative best practices page addresses hook timing only, and its in-feed specification page states a ten minute maximum for standard advertisements and no restriction for the boosted-post format, with no recommendation attached to either. The significant pattern is that the two platforms holding the deepest video advertising measurement have examined the question and published nothing.Who recommends what, and on what basisPlatformRecommendsReason givenEvidenceLargest socialUnder 15 sNoneNone”shorter videos (6-15 seconds) are more effective”; stories under 10 sProfessional network15 to 30 sPlacement eligibilityN/A”so they can qualify for all possible placements”. A technical rule, not an attention claim.Video platformNothingDeclines to recommendN/AIts creative playbook: zero occurrences of “length”, “duration”, “shorter”Short-videoNothingDeclines to recommendN/ASpecification pages give maxima only, with no recommendation attachedThe two platforms with the most video measurement in the world published nothing.That should lower your confidence in anyone who has published a number.
The two platforms with the deepest video measurement decline to answer. That is worth more than either recommendation. Source : Platform creative guidance and specifications, read directly (2026)

The one recommendation with a real reason

The professional network’s guidance is the only one you can act on with confidence, and not for the reason people assume.

What it says. Upload videos 15 to 30 seconds long so they can qualify for all possible placements across the feed and the audience network.

What that is. An eligibility constraint. Certain placements will not accept your video outside that window, so a video outside it reaches a smaller inventory pool.

What it is not. Evidence that 15 to 30 seconds holds attention better. The platform does not claim that, and reading it that way turns a technical fact into a folk belief.

Why this is still the most useful number on the page. Because you can verify it, it does not depend on anyone’s study, and it has a direct consequence: a 45-second video competes in fewer auctions.

The general principle to take from it. Your practical duration window is usually decided by which placements you want to be eligible for. Work out the placements first, then take the intersection of their limits, then write to that.

The limits that actually bind

The specification pages vary far more than the recommendations do, and they are hard constraints rather than advice.

Desktop in-stream: 5 to 15 seconds. The same placement on mobile allows 5 seconds to 10 minutes. One placement, two device contexts, and a fortyfold difference in ceiling.

Stories: 1 to 120 seconds on the largest social platform.

Rewarded video: 3 to 61 seconds on its audience network.

Non-skippable in-stream: 60 seconds or shorter on the video platform.

Bumper: between 5 and 6 seconds, non-skippable, on the same platform.

Skippable in-stream: no maximum, with under 3 minutes recommended.

Professional network: 3 seconds to 30 minutes. An extremely wide window whose practical range is narrowed by the eligibility rule above, not by the specification.

Short-video platform: up to 10 minutes for standard in-feed ads, no restriction for the boosted-post format.

What to do with this. Write down the placements you want, list their windows, and take the intersection. That is your real duration constraint, and it is knowable without any research.

Technical duration limits for video advertising by placement across platformsTable of the technical duration limits that apply to video advertising placement by placement across platforms, which constrain creative far more tightly and far more variably than the published recommendations do. On the largest social platform, in-stream video placement on desktop accepts videos of five to fifteen seconds while the same placement on mobile accepts five seconds to ten minutes, a fortyfold difference in ceiling for one placement across two device contexts; stories placement accepts one to one hundred and twenty seconds; feed and marketplace placements accept up to two hundred and forty-one minutes; and rewarded video on its audience network accepts three to sixty-one seconds while its native, banner and interstitial formats accept one to one hundred and twenty seconds. On the video platform, non-skippable in-stream advertisements must be sixty seconds or shorter, bumper advertisements must run between five and six seconds and are non-skippable, and skippable in-stream advertisements have no maximum length with under three minutes recommended. On the professional network, the specification window runs from three seconds to thirty minutes, a very wide range whose practical extent is narrowed not by the specification but by the platform’s placement eligibility rule recommending fifteen to thirty seconds. On the short-video platform, standard in-feed advertisements accept up to ten minutes while the boosted-post format carries no duration restriction. The practical method is to list the placements the advertiser actually intends to run, list the duration window of each, and take the intersection of those windows as the real constraint on creative length.The hard limits, placement by placementThese bind. The recommendations do not.PlacementAccepted durationIn-stream, desktop5 to 15 secondsIn-stream, mobile (same placement)5 seconds to 10 minutesStories1 to 120 secondsRewarded video, audience network3 to 61 secondsNon-skippable in-stream60 seconds or shorterBumperBetween 5 and 6 secondsProfessional network, any placement3 seconds to 30 minutesMethod: list the placements you want, list their windows, take the intersection.That is your real constraint, and it needs no research to establish.
One placement can cap at 15 seconds on desktop and allow ten minutes on mobile. Work from the intersection of what you want to run. Source : Platform specification pages, read directly (2026)

The measurement problem underneath short video

One platform mechanic quietly makes short-video reporting incomparable, and almost nobody accounts for it.

The mechanic. 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 loops twice. Videos over 30 seconds play once.

What that does to completion. A short video accumulates completions from repeat plays that a longer video cannot. Its completion rate is not measuring the same event.

What that does to watch time. Average watch time on a short video there includes looped seconds, so a short video can report watch time approaching its longer competitor while showing the viewer the same eight seconds four times.

Why this interacts with the eligibility rule. The platform recommends 15 to 30 seconds, and 15 to 29 seconds is precisely the range that loops. The recommended range is the one where the metrics are hardest to read.

What to do. Never compare completion rates between a video under 30 seconds and one over it on that platform. Compare cost per outcome, which does not care how many times the file played.

And keep the general lesson. Before believing any duration comparison, check whether the two durations were measured on the same basis. Frequently they were not.

The one measurement that exists, and it complicates the case for short

There is a single open, sampled study of skip behaviour, and its central finding is proportional rather than absolute.

The study. HTTP traces from a campus network, March to November 2014, 99,658 video-ad exhibitions across 5,667 unique ads, desktop only.

The headline numbers. 29% of exhibitions were watched in full. 35% were skipped in under six seconds. Only 25% ran past ten seconds before the viewer skipped.

The finding that matters here. On average, a user skipped after 20% of the ad content had been shown, with a standard deviation of 19%.

Why that is awkward for the short-video argument. If departure is proportional to length rather than fixed in seconds, then a longer ad receives more absolute seconds of exposure, not fewer. Twenty percent of 60 seconds is twelve seconds. Twenty percent of 15 seconds is three.

The caveats that stop this becoming a rule. The study did not test length as a variable, so the proportionality could be an artefact of the ad mix rather than of viewer behaviour. It is 2014 desktop, skippable in-stream only. And the standard deviation is nearly as large as the mean, so there is no typical viewer to design for.

What it does establish. That the intuition behind short video, that people bail at a fixed number of seconds, is not what the one available measurement found. That is enough to justify testing a longer version rather than assuming it will fail.

Proportional skip behaviour and its implication for absolute exposure across advertisement lengthsDiagram presenting the central finding of the only open sampled study of video advertisement skip behaviour and working through its implication for advertisement length. The study captured HTTP traces 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 across five thousand six hundred and sixty-seven unique advertisements, on desktop only. Twenty-nine percent of exhibitions were streamed in full, thirty-five percent were skipped in under six seconds, and only twenty-five percent ran beyond ten seconds before the viewer skipped. The finding relevant to duration is that on average a user skipped after twenty percent of the advertisement content had been shown, with a standard deviation of nineteen percent, meaning departure was measured as a proportion of length rather than at a fixed number of seconds. If that proportionality holds, a longer advertisement receives more absolute seconds of exposure rather than fewer, since twenty percent of a sixty-second advertisement is twelve seconds while twenty percent of a fifteen-second advertisement is three seconds. Three caveats prevent this from becoming a rule: the study did not test length as an independent variable, so the proportionality may be an artefact of the advertisement mix rather than of viewer behaviour; the data is from two thousand and fourteen desktop traffic in skippable in-stream placements only, which differs from current mobile feed delivery; and the standard deviation is nearly as large as the mean, so no typical viewer exists to design for. What it does establish is that the intuition behind short video, namely that viewers depart at a fixed number of seconds, is not what the one available measurement found, which is sufficient reason to test a longer version rather than assume it will fail.Departure was proportional, not fixed in seconds99,658 ad exhibitions, 2014, desktop, skippable in-stream. Average skip at 20% of content.A 15-second ad3 seconds seenA 30-second ad6 seconds seenA 60-second ad12 seconds seenColoured = seen at the average skip point. Grey = never reached.What this suggestsIf skipping is proportional, longer ads getmore absolute exposure, not less. That is theopposite of the short-video intuition.Why it is not a ruleLength was not tested as a variable. It is 2014desktop, skippable only. And the deviation isnearly as large as the mean.Enough to justify testing a longer version rather than assuming it fails.
If departure is proportional rather than fixed in seconds, a longer ad receives more absolute exposure, not less. Source : Arantes, Figueiredo, Almeida, ACM Web Science (2016)

The lift figures on the platform blogs

There is one place where a platform does publish duration performance, and it deserves a specific warning.

What it claims. A platform blog states that video ads falling between 21 and 34 seconds received a 280% lift in conversion. The same page claims a 312% lift for videos shot at 720p or higher, 91% for the 9:16 aspect ratio, and 152% for text-format calls to action.

What the page contains in support. No footnote. No source. No sample size. No period. No description of the comparison group. No definition of lift.

Why the numbers themselves should worry you. A 312% conversion lift attributable to resolution is not a plausible creative effect. It is far more likely a selection artefact: advertisers who shoot at higher resolution differ from those who do not in every other respect too.

What the other platform’s unsourced figures look like. The largest social platform states that feed video ads perform significantly better with sound on, at 2.25x higher call-to-action click-through rate, and that 98% of users hold their phones vertically. Same pattern: specific figure, no source.

How to use any of this. As a statement of what the platform wants you to do. That is genuinely informative, because platforms optimise for their own inventory and their preferences are real. It is not evidence.

The test that settles it for you. Two lengths, same message, same audience, running concurrently. A week of spend beats every unsourced multiple in the category.

Conversion lift figures published on platform creative blogs without methodologyTable of the conversion lift figures published on advertising platform creative blogs alongside a note that none carries any supporting methodology. One platform blog claims a two hundred and eighty percent lift in conversion for video advertisements between twenty-one and thirty-four seconds in length, a three hundred and twelve percent lift for videos shot at seven hundred and twenty pixel resolution or higher, a ninety-one percent lift for videos shot in the nine by sixteen vertical aspect ratio, and a one hundred and fifty-two percent lift for calls to action presented in text format; it also claims lifts of thirty-eight percent for varied scenes, eighty percent for subtitles and eighty-seven percent for the combination in electronic commerce, and one hundred and seventy-one percent for five or more scenes and forty-three percent for text embedded within the first seven seconds in gaming. The page carries no footnote, no source, no sample size, no period, no description of the comparison group and no definition of lift. The magnitude of the figures is itself diagnostic: a three hundred and twelve percent conversion lift attributable to recording resolution is not a plausible creative effect and is far more likely a selection artefact, since advertisers who shoot at higher resolution differ from those who do not in many other respects that also affect conversion. The largest social platform publishes unsourced figures in the same pattern, stating that feed video advertisements perform significantly better with sound on at two point two five times higher call-to-action click-through rate, and that ninety-eight percent of users hold their phones vertically. Such figures are usefully read as statements of what each platform wants advertisers to do, which is genuine information, but not as evidence.Lift figures with no methodology anywhere on the pageClaimStated liftSample publishedVideo between 21 and 34 seconds+280%NoneShot at 720p or higher+312%NoneShot in 9:16+91%NoneCall to action in text format+152%NoneSound on (different platform)2.25x CTA CTRNoneThe size of the numbers is the tellA 312% conversion lift from recording resolution is not a creative effect. Advertisers who shootat higher resolution differ in every other respect too. That is selection, not causation.
A 312% conversion lift from resolution is not a creative effect. It is a selection artefact with a percentage sign. Source : Platform creative blog, read directly (2026)

How to actually decide

Six steps, none of which requires anybody to have published a threshold.

List the placements you want. Not the ones available. The ones you intend to run.

Take the intersection of their duration windows. That gives you a hard range, and it is often narrower than you expected.

Write to the message, not to the range. Inside the window, length should follow what the message needs. An ad that fills 30 seconds because 30 was allowed is worse than one that finishes at 12.

Front-load regardless of length. Departure begins immediately in a feed, so a 40-second ad and a 10-second ad have the same first four seconds problem.

Test two lengths concurrently. Same message, same audience, same landing page, running at the same time rather than one after the other.

Judge on cost per meeting held. Not on completion rate, which as shown above is not comparable across the 30-second boundary on at least one platform.

A method for deciding video advertisement length from placement constraints and message requirementsDiagram setting out a method for deciding video advertisement length that does not rely on any published duration threshold. The first step is to list the placements the advertiser actually intends to run rather than the placements available, since eligibility rather than attention is what genuinely constrains duration. The second step is to take the intersection of the duration windows of those placements, which produces a hard range that is frequently narrower than expected, given that a single placement can cap at fifteen seconds on desktop while permitting ten minutes on mobile. The third step is to write to what the message requires rather than to fill the available range, on the principle that an advertisement running thirty seconds because thirty seconds was permitted performs worse than one that finishes at twelve seconds because the message was complete. The fourth step is to front-load the message regardless of the chosen length, since departure in a feed begins immediately and a forty-second advertisement faces the same opening-seconds problem as a ten-second one. The fifth step is to test two lengths concurrently on the same message, the same audience and the same landing page, running simultaneously rather than sequentially so that market conditions are shared. The sixth step is to judge the result on cost per meeting held rather than on completion rate, because completion rate is not comparable across the thirty-second boundary on at least one major platform, where videos shorter than thirty seconds loop until thirty seconds of playback accumulate.Decide from placements and message, not from a threshold1. Placements you wantNot the ones available.The ones you will run.2. Intersect their windowsUsually narrower than youexpected. That is your range.3. Write to the messageFinishing at 12 beats filling30 because 30 was allowed.4. Front-load anywayA 40-second ad has the samefirst-four-seconds problem.5. Test two lengthsConcurrently, not one afterthe other.6. Judge downstreamCost per meeting held, notcompletion rate.Why step 6 matters more than it soundsCompletion rate is not comparable across the 30-second boundary on the professional network.A week of concurrent spend settles this better than every unsourced multiple in the category.And the answer will be specific to your message, which is the only place it was ever going to live.
Placements give you a hard window. The message decides where in the window you land. Source : Platform specification and eligibility documentation (2026)

Where to go next

You are working on the opening seconds. The ad hook.

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

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

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

You are choosing between platforms. LinkedIn Ads vs Google Ads.

You want the buying mechanics underneath. Media buying explained.

In short

  • Two of the four largest video platforms publish no duration recommendation at all, including the one with the deepest creative measurement in the industry.
  • The platform recommending under 15 seconds cites nothing. No study, no sample, no period.
  • The professional network’s 15 to 30 seconds is an eligibility rule, stated as such: videos in that range qualify for all possible placements.
  • The real constraints are placement specifications, and they vary far more: 5 to 15 seconds on desktop in-stream, 5 seconds to 10 minutes on the same placement on mobile.
  • Videos under 30 seconds loop on the professional network until 30 seconds of playback accumulates, so completion rates across that boundary are not comparable.
  • The recommended 15 to 30 second range is precisely the range that loops, which makes it the range where metrics are hardest to read.
  • The platform lift figures have no methodology. A 312% conversion lift attributed to recording resolution is a selection artefact, not a creative effect.
  • Decide from your placements and your message, then test two lengths concurrently and judge on cost per meeting held.

Work out your window, then let the message pick the length. Book a diagnostic, or see how we approach B2B paid acquisition.