You have met this statistic: human attention spans have fallen from twelve seconds to eight, which is shorter than a goldfish. It is the opening line of a large fraction of short-form content advice, and it is the justification for “you have three seconds”, “front-load everything” and most of what follows.
It is not a finding. It is a number that escaped into circulation and was never caught.
The trail
The claim is usually attributed to Microsoft, on the strength of a 2015 report from Microsoft Canada that did exist and did contain real survey and neurological work. The eight-second figure, however, was not a finding of that research. It appeared in an infographic, sourced to a third party called Statistic Brain.
When the BBC followed the chain in 2017, Statistic Brain could not produce a credible source for the number. The trail ends there — not in a disputed study, but in nothing at all. The account is summarised in Fast Company's write-up and in LinkedIn's own marketing blog, which is unusual in that it argues against a statistic its own industry finds useful.
What specialists say instead
Attention is not one quantity with one duration. It varies enormously by task, by motivation and by context — the same person will abandon a video in two seconds and read for three hours the same evening. Clinicians who work on attention reject the single-number framing directly: Northwell Health's summary notes the eight-second claim is not supported by peer-reviewed research, and that there is no good evidence attention has changed materially since it was first studied.
The goldfish comparison is worse still: nobody established the goldfish's figure either.
What the myth has cost
It has been expensive in a specific direction. If you believe attention is a fixed eight-second budget, the strategy follows mechanically: cut everything, front-load everything, never let a beat breathe, assume the viewer is incapable rather than unconvinced.
That is a description of a great deal of short-form content, and it is why so much of it is exhausting and forgettable. It also cannot explain the things that actually happen in these feeds — long videos that hold people, slow openings that work, formats built entirely on delayed payoff.
The better model
Attention is not a budget that runs out. It is a continuously renewed decision. The viewer is not running down a timer; they are asking, repeatedly, whether the next moment is worth the one they are in. That is a far more demanding standard than eight seconds, and a far more optimistic one, because it means the video can earn more time at any point rather than simply spending an allowance.
It also matches how the platforms measure. Watch time accumulates in seconds across a video and its replays — a model in which attention is continuously won, not a fixed quantity spent. See watch time is the signal.
The general lesson
This is the cleanest example we know of a statistic that survives because it is useful, not because it is true. It is specific, it is memorable, it flatters the reader's sense that things are getting worse, and it justifies advice people already wanted to give.
Those four properties are a reasonable checklist for suspicion. When a figure has all of them and you cannot find who measured it, assume nobody did — the same test we apply to platform claims.