How quickly the world forgets
On October 9, 2024, Indian industrialist Ratan Tata died. Within hours, nearly three million people visited his Wikipedia article. The next day, that number dropped to 768,000. By the third day, 317,000. By the end of the week, 56,000. Within a month, his article had returned almost exactly to its pre-death baseline of about 5,000 daily views.
This pattern is everywhere. When the Francis Scott Key Bridge collapsed in Baltimore, when Alexei Navalny died in a Russian prison, when CrowdStrike crashed half the world's computers — in each case, millions of people simultaneously turned to the same source of information. And in each case, the surge of attention followed a strikingly similar arc: a sharp peak, then a rapid, predictable decay back to nothing.
I wanted to know: how fast does this happen, exactly? Is there a universal law governing how quickly we collectively move on? And does the type of event matter?
To find out, I collected Wikipedia pageview data for 96 notable events from 2024 — deaths, disasters, political moments, entertainment releases, and more. For each, I tracked how daily views evolved from the moment attention peaked through the weeks that followed.
The results are more striking than I expected.
Below, every one of the 96 events is plotted on the same axes. The vertical axis shows attention as a fraction of the peak (1.0 = maximum), and the horizontal axis shows days after the peak. Each thin line is one event. The gold line is the median.
The visual is arresting: despite spanning completely different domains — deaths, politics, sports, entertainment, disasters — nearly every event traces the same steep plunge. There is no event type that defies the pattern. Attention always decays, and it always decays fast.
The median half-life — the time for excess attention to drop by half — is 0.82 days, roughly 20 hours. By the next day, 58% of the surge is gone. After two days, 84%. After a week, just 3% of the original spike remains. Whatever it was that gripped millions of minds, the grip loosens in hours, not days.
Select any event to see its raw pageview timeseries and normalized decay curve. The vertical dashed line marks the peak day.
The decay is not exponential. If attention faded at a constant rate — like radioactive decay — the curve would be a smooth, even descent. Instead, the initial drop is much steeper than an exponential, followed by a long, slow tail. Attention doesn't just fade — it crashes, then lingers.
The best fit, for 99% of the events in this dataset, is a stretched exponential:
This is also known as the Kohlrausch function, named after the 19th-century physicist who first used it to describe how electric charge leaks out of a capacitor. It has two parameters: τ (the timescale) and β (the shape). When β = 1, it's a pure exponential. When β < 1, the initial decay is faster than exponential but the tail is heavier — the curve drops like a stone, then stretches out into a long, slow fade.
Across all 96 events, the median β is 0.50. Not a single event in the dataset has β ≥ 1. The decay of collective attention is universally faster-than-exponential in its initial phase, with a heavy tail.
The stretched exponential isn't just a curve that happens to fit. In physics, it emerges whenever a system is composed of many independent relaxation processes with different timescales. A glass cooling, a polymer relaxing, electric charge dissipating in a disordered material — all follow stretched exponentials because the underlying system is heterogeneous. There isn't one decay rate. There are many, superimposed.
Applied to attention, the interpretation is intuitive. When Ratan Tata dies, some fraction of viewers are people with a passing curiosity — they look once and never return. Others have a deeper connection — maybe they're Indian, or interested in business history — and they come back the next day, or the day after. A tiny fraction keep checking for weeks. The distribution of individual decay rates, superimposed, produces the characteristic stretched exponential curve: a sharp initial crash (the casual viewers disappearing) followed by a long tail (the committed few lingering).
The value β ≈ 0.5 has a specific meaning. It implies that the underlying distribution of individual attention timescales follows approximately an inverse Gaussian distribution — a strongly right-skewed distribution where most people forget almost immediately but a small minority persists far longer than the average.
Not all events decay at the same rate. The chart below shows the median normalized decay curve for each category of event.
The differences are real, if modest. Holidays decay fastest (median half-life: 0.64 days) — which makes intuitive sense; once the day is over, the day is over. Political events (0.69 days) also fade quickly, perhaps because political attention is driven by news cycles that move fast. Deaths (0.92 days) and conflicts (0.95 days) linger slightly longer — grief and geopolitical consequence sustain attention beyond the initial spike.
But the differences are small compared to the overwhelming commonality. Every category shows the same basic shape: a steep crash on day one, followed by a long, slow tail toward baseline. The type of event modulates the speed slightly, but the form is universal.
The half-life distribution itself is remarkably concentrated. Most events cluster between 0.5 and 1.5 days. A few outliers persist longer — Imane Khelif (1.8 days, sustained by ongoing Olympic controversy), Alexei Navalny (1.6 days, geopolitical significance), Manmohan Singh (1.6 days, national mourning). But no event in the dataset has a half-life longer than 4 days.
Juneteenth (half-life: 0.6 days), Nowruz (0.6 days), New Year's Day (0.6 days). Holidays are perfectly predictable spikes: attention peaks on the day itself and crashes immediately after. There is no "news" to follow, no story to develop. The information need is satisfied in a single visit.
Imane Khelif (1.8 days). The Algerian boxer at the centre of a gender eligibility controversy during the 2024 Olympics. Day-1 retention was 98% — attention on day 2 was almost as high as day 1. This is the only event in the dataset where the controversy sustained near-peak attention for more than 24 hours.
Alexei Navalny (1.6 days). The Russian opposition leader died in a penal colony in February 2024. The geopolitical implications — occurring during the war in Ukraine — sustained attention well beyond a typical death announcement.
J.D. Vance (5.1M peak views) received the largest single-day traffic of any article in the dataset, driven by his selection as Donald Trump's vice-presidential running mate. Despite the enormous spike, the half-life (1.2 days) was only modestly above average.
Ratan Tata (2.9M views) and Imane Khelif (2.7M views) round out the top three. The dataset includes 14 events with peak views exceeding one million.
Three things stand out.
First: attention decays faster than you think. The median half-life of roughly 20 hours means that by the time most people read a "day after" analysis piece, the majority of public attention has already moved on. The news cycle isn't just fast — it's faster than the time it takes to write about it.
Second: the decay is universal. Deaths, disasters, political moments, entertainment releases, sporting events, holidays — they all follow the same stretched exponential form with β ≈ 0.5. The specific content barely matters. Whatever mechanism drives collective attention — and its dissolution — operates the same way regardless of what we're paying attention to.
Third: the decay is heterogeneous, not uniform. The stretched exponential tells us this isn't a simple case of "everyone forgets at the same rate." It's a superposition of vastly different individual behaviours: the vast majority who glance and move on immediately, a smaller group who linger for a day or two, and a tiny minority who remain engaged for weeks. The aggregate curve — that characteristic sharp crash with a long tail — is an emergent property of this heterogeneity.
There is something both humbling and liberating in this. The events that seem, in the moment, like they'll define the year — the bridge that collapsed, the election that loomed, the person who died — are, to the collective attention of millions, a brief perturbation. Within days, the signal has returned to baseline. The world has moved on. Not because the events don't matter, but because that's what attention does.
It has a half-life, and it's about twenty hours.
Data and methods. Pageview data from the Wikimedia REST API. Top 200 articles per day collected for 342 days of 2024. Spike articles identified as those appearing in the top-200 on 1–30 days with peak views ≥100,000. Daily pageview timeseries collected for a 60-day window around each spike. Baseline estimated as median of pre-spike period (−14 to −7 days). Decay curves normalized by (views − baseline) / (peak − baseline). Half-lives computed empirically by interpolation. Stretched exponential fitted via grid search over β with linear regression on log-transformed data. 96 events passed quality filters (R² > 0.5, ≥5 post-peak data points).