The Verification Venue · a viral number with no source
The Fish Remembers; The Statistic Never Existed
The claim writes itself: the human attention span fell to 8 seconds, down from 12 in 2000, now one second under a goldfish's 9, all thanks to phones. It is fabricated on both halves. No study ever measured an 8-second human span, and no one has ever measured a goldfish's attention at all. Trace the citation until it dead-ends in nothing, measure your own focus swinging by an order of magnitude, and watch the goldfish's real trait, memory, bury the myth. Every number here is recomputed in your browser.
Primary sources backing the 8s / 9s / 12s stat
Verdict: fabricated on both halves. The BBC (More or Less, 2017) traced the figures to Statistic Brain, which produced no substantiating research; the bodies it named (the US National Library of Medicine / NCBI and the Associated Press) had no record of any such study.
1. Trace the citation to its dead end
Pick a figure. Each one is supposed to rest on research. Follow the trail node by node and watch where it actually leads. The counter at the bottom tallies how many primary sources you reach. Spoiler: it is a constant.
Live citation graph
Where does the number come from?
The Microsoft Canada report is real: its Consumer Insights team surveyed about 2,000 Canadians and ran EEG on 112 people in Spring 2015. But the 8 / 9 / 12-second graph was not its own finding: the report reproduced a figure it credited to Statistic Brain. So this is a story of sourceless reporting recirculated, not of Microsoft inventing a number.
2. Measure your own span (it has no single value)
Even if the 8 seconds had a source, the premise is a category error: attention span is not a fixed quantity you have. Sustained attention is task- and engagement-dependent. Run both tasks below. The dull one asks you to keep watching a counter that does nothing; press Bored / done the instant your mind wanders. The engaging one is a quick tap game; keep going as long as it holds you. Your own two numbers will not match, and the ratio is the point.
Live self-measurement
Your on-task time, two ways
A · the dull task
– s on task
B · the engaging task
– s on task
This is your data, not a lab claim, and that is exactly the point: the same person holds focus for seconds on something dull and far longer on something gripping (a film, a game, deep work). Because the number moves with the task, there is no single true human "attention span" to swap in for 8 seconds. Substituting one figure for another just repeats the original mistake.
3. The goldfish's real trait is memory, and it is good
The "9-second goldfish" was never a mismeasurement; the number was invented whole. What goldfish actually have, and what has actually been measured, is memory. Gee, Stephenson & Wright (1994) trained goldfish to press a lever for food during a fixed one-hour daily window. The fish learned to anticipate feeding time; the response held across a four-week schedule and persisted for days into extinction. Plot the myth's 3-second memory against that on a log axis and it vanishes.
Log time-axis · recomputed live
3 seconds vs a 4-week schedule
Myth: goldfish memory
3 s
Gee 1994: feeding schedule
– s
Gap (orders of magnitude)
–
The bar you can barely see on the left is the "3-second memory." The bar on the right is 28 days = 2,419,200 seconds: the four-week schedule across which the trained fish kept anticipating feeding time. The ratio, recomputed above, is about 806,400×, close to 6 orders of magnitude.
The honest reversal is a double one
Why "just use the real number" is still wrong
The tempting fix is to keep the format and swap the values: "actually humans focus for X and goldfish remember for Y." But the frame itself is broken. Ranking two species by a single-number "attention span" is meaningless, because the two halves fail for different reasons:
The human half: no fixed quantity
Attention is not a scalar you own. It is task- and engagement-dependent (Panel 2). There is no single true human number, so any one figure, 8 s or otherwise, is the wrong type of answer.
The goldfish half: never measured
No one ever measured a goldfish's attention. The 9 s was fabricated. The measurable trait is memory (Panel 3), and it is good: weeks, directly, per Gee 1994.
So the complete answer is not "both numbers are fake" (though they are). It is that the human decline is unsourced, the goldfish comparison is invented, and the very idea of a one-number species ranking of "attention span" is a category error, twice.
The three figures, followed to the end
| figure | claimed | credited to | where the trail ends | primary sources |
|---|---|---|---|---|
| 8 s | human span "today" | Statistic Brain (via MS 2015) | no substantiating study; NLM/NCBI & AP had no record (BBC 2017) | 0 |
| 12 s | human span "2000" | Statistic Brain (via MS 2015) | no dated study; same broken trail | 0 |
| 9 s | goldfish span | Statistic Brain (via MS 2015) | no goldfish attention was ever measured; invented | 0 |
The check: every number recomputed in front of you
Nothing here is a stored figure. The citation tally, the memory gap, and your own focus swing are all recomputed live from the definitions below:
The offline gate recomputes all of this, two independent ways where possible: node research/is-your-attention-span-shorter-than-a-goldfish/verify-is-your-attention-span-shorter-than-a-goldfish.mjs. Free choices & uncertainty: the primary-source tally is 0 by construction: the trail is the finding. The memory gap uses 28 days (the Gee 1994 four-week schedule) against the myth's 3 s; anchoring on weeks is deliberate (see the honesty notes). The "measure your own span" numbers are your inputs, so we verify the arithmetic (the ratio and its order of magnitude), not a preset value; if you skip a task the swing simply says so.
What's exactly true here, and the honesty notes
The stat is unsourced. Simon Maybin's BBC More or Less piece (10 March 2017) chased the 8 / 9 / 12-second figures to a Statistic Brain page, found no substantiating research, and checked the outside bodies Statistic Brain listed (the US National Library of Medicine / NCBI and the Associated Press), which had no record of any such study. The NLM/NCBI and AP appear here only as the broken trail; they never endorsed the number.
Microsoft reproduced, it did not fabricate. The Microsoft Canada "Attention Spans" report (Spring 2015) is a real study (about 2,000 surveyed, EEG on 112), but the widely-shared graph was lifted from Statistic Brain, not produced by Microsoft. Many later outlets (including Time, 2015) repeated the myth; that recirculation is not independent confirmation.
Attention has no single number. Researchers who study attention say sustained focus cannot be reduced to one species-wide value; it depends on the task and engagement. The Panel 2 exercise is an informal demonstration of that variability, using your own timing, not a controlled measurement of "your attention span."
Anchor goldfish memory in weeks. Gee, Stephenson & Wright (1994) directly demonstrate retention across a four-week feeding schedule, with anticipation persisting for days into extinction. "Weeks to months" and "three months" appear in secondary sources and other paradigms (for example avoidance learning); we do not hard-state a month figure. Weeks are what this primary study measured directly; months are reported in the wider literature.
Never measured, not mismeasured. No one has ever measured a goldfish's attention span. The 9 s is an invention, not an error bar. The goldfish's measurable, and good, trait is memory.