Attention is the most over-quoted and under-sourced topic in advertising. It is also the one where the underlying research is genuinely good, once you separate it from the things people say about it at conferences. Part two of the ad creative series.
What you actually get
Facebook's own research put average dwell time on a feed ad at 1.7 seconds on mobile and 2.5 seconds on desktop, from an analysis combining 2,183 ratings across 965 ads with more than 200 Nielsen studies. The same work found that recall can begin forming in as little as a quarter of a second of exposure.
Two things follow, and they pull against each other. The window is very short. And something useful happens inside it even when it is very short — a quarter-second of exposure is not nothing.
That 2016 vintage matters. Feed mechanics, autoplay behaviour and video length norms have all moved since. Treat 1.7 seconds as the right order of magnitude rather than this year's number, because nobody has published a comparable replacement.
Recall rises with time on screen
Dentsu's attention work in Japan tracked more than 8,000 mobile users with eye tracking across seven ad formats and more than 45 creatives. Recall roughly doubled between the shortest and longest exposures.
The detail worth carrying is the last line of that caption. Ten seconds that someone chose to give you is worth substantially more than ten seconds you took. Skippable formats where the viewer stays are not a weaker version of unskippable formats — they are a different and better signal.
There is also a published academic curve for this, fitted from two controlled experiments, and it is drawn on the hub page against the fatigue curve. Its shape says the same thing with more precision: memory climbs steeply through roughly the first ten seconds, then keeps climbing more slowly.
The squeeze is real and it is old
The best long-run evidence on attention scarcity is Thales Teixeira's Harvard Business School working paper, which assembled a cost-per-thousand series running from 1966 to 2010 alongside compiled skip-rate and eye-tracking research.
Cost of attention up seven to nine times. Share of ads receiving full attention down from near-universal to under a fifth. That is the market every creative decision is made inside, and it explains why the first seconds carry so much weight — not because of anything about human biology, but because the alternative to your ad is one thumb-flick away and always has been, now.
What does not survive a check
Amplified Intelligence's work is widely cited for a 2.5-second attention-memory threshold, with 80 to 85% of ads failing to clear it, from a study of about 600 people across more than 3,300 impressions and 12 brands. I have used the figure, but with a flag: the page carrying it is internally inconsistent about its own numbers, and I could not resolve the discrepancy from the source. Directionally I think it is right and the sample is real. Treat the specific percentage as approximate.
Two things routinely quoted alongside it do not survive at all. There is no eight-second human attention span — that chain runs back through a defunct statistics aggregator to a report that did not contain the claim. And the "85% of video watched without sound" figure comes from publisher self-reports in 2016, not from any platform's data. Both are handled in full on the hub.
The reason to be strict here is that attention is exactly the topic where a plausible number does real damage. If you believe you have eight seconds, you will build a different ad than if you believe you have 1.7.
What actually changes in a hook between platforms
Less than people expect, and not in the direction they expect.
The mechanics of stopping a scroll have converged. Pattern interruption works across Meta, on LinkedIn, on Google's feed placements and now in conversational surfaces, and the first three to five seconds are where the work happens on all of them. A hook that earns a stop on one platform generally earns a stop on another.
What changes is intent — what a person is doing when your ad reaches them, and therefore which kind of ad reads as reasonable rather than intrusive. That is not a difference in hook technique. It is a difference in which archetype you should be reaching for: an offer and a guarantee, a comparison, a checklist, a proof format like a review or a case study, or an outright pattern interrupt built from something people do on their phones all day. The catalogue of those archetypes, with the job each one does, is in part three.
So the working rule I use is: hook technique travels, format choice does not.
What to read next
- Ad creative formats — 43 archetypes, grouped by the job they do.
- Ad creative testing — how many impressions a hook needs before you can read it.
FAQ
How long do I actually have in a feed? Facebook's own research put average dwell time on a feed ad at 1.7 seconds on mobile and 2.5 seconds on desktop, from 2,183 ratings across 965 ads combined with more than 200 Nielsen studies. The same work found recall can begin forming in as little as a quarter of a second. That research is from 2016 and feed mechanics have moved since, so treat 1.7 seconds as the right order of magnitude rather than this year's number — nobody has published a comparable replacement.
Is there really an eight-second human attention span? No. That chain runs back through a defunct statistics aggregator to a report that did not contain the claim. The frequently paired "85% of video watched without sound" figure comes from publisher self-reports in 2016, not from any platform's data. Attention is exactly the topic where a plausible number does real damage: if you believe you have eight seconds, you will build a different ad than if you believe you have 1.7.
Is longer viewing always better? Duration helps — Dentsu's eye-tracking work with more than 8,000 mobile users found recall roughly doubled, from 21% under one second to 41% over ten seconds. But the more interesting half of that study is that voluntary viewing beyond ten seconds outperformed forced viewing of the same length by 79%. Ten seconds someone chose to give you is worth substantially more than ten seconds you took.
Does a hook that works on one platform work on another? Generally yes. The mechanics of stopping a scroll have converged, and pattern interruption works across Meta, LinkedIn, Google's feed placements and conversational surfaces, with the first three to five seconds carrying the work on all of them. What changes between platforms is intent — what someone is doing when your ad reaches them, and therefore which archetype reads as reasonable rather than intrusive. Hook technique travels; format choice does not.
Sources on this page
- Facebook IQ, "Capturing Attention in Feed: The Science Behind Effective Video Creative", 2016. 2,183 ratings across 965 ads, fourth quarter 2014 to fourth quarter 2015, with more than 200 Nielsen studies; fieldwork by Fors Marsh.
- Dentsu Japan, "Introducing the Attention Economy", 2024. More than 8,000 mobile users, eye tracking, seven formats, 45-plus creatives.
- Teixeira, "The Rising Cost of Consumer Attention: Attention Economics", Harvard Business School Working Paper 14-055, 2014.
- Amplified Intelligence, "Why Does the Attention-Memory Threshold Matter". About 600 people, 3,300-plus impressions, 12 brands. Undated on the page and internally inconsistent about its own figures — cited here with that caveat.
- Goldstein, McAfee and Suri, "The Effects of Exposure Time on Memory of Display Advertisements", Proceedings of the 12th ACM Conference on Electronic Commerce, 2011.
