How long does the internet care?
When something suddenly becomes news, people rush to read about it on Wikipedia. We measured how quickly that rush fades, for every big spike in two years of English Wikipedia page views.
The finding
Loading…
Look up a topic
Matches the start of the title. Only spikes big enough to rank are included (see how this was measured, below).
Show the numbers, day by day
| Day | Extra views | Share of the month | Running total |
|---|
Fastest forgotten, longest remembered
Among the biggest spikes (at least 20,000 extra views on their busiest day), by attention half-life. Click a title to see its curve.
Gone fastest
| Page | Date | Half-life |
|---|
Lingered longest
| Page | Date | Half-life |
|---|
How this was measured
What counts as a spike
We read every hour of English Wikipedia page views from September 2024 to September 2026 (desktop and mobile, human readers only, as Wikimedia classifies them). A page spikes on a day when it gets at least 1,000 views, at least 500 more than usual, and at least five times its usual daily views. "Usual" is the page's median day over the four weeks before.
What "half-life" means here
For each spike we add up all the extra views — above the page's usual level — over the 30 days after the surge began. The attention half-life is how many hours it took for half of that month's extra attention to arrive. A short half-life means the world looked, then moved on. A long one means people kept coming back.
The measure we threw away
Our first definition was the obvious one: how long until hourly views drop to half of the busiest hour. It passed every automated test — and gave answers that were plainly wrong.
| Event | Busiest-hour measure | Attention half-life |
|---|
The moment news breaks, one hour towers over everything after it, so the busiest-hour measure only timed that single hour. Liam Payne's page got more views on the day after his death than on the day itself, yet the busiest-hour measure said attention halved within an hour.
Spikes we flag and leave out of the rankings
- Bursts
- Some traffic is automated even though Wikimedia counts it as human. It looks like a cliff: millions of views in one or two hours, out of nowhere, then nothing. ("Schutzstaffel" got 7.8 million views in one hour, then 278.) Real news fades; it doesn't fall off a cliff.
- Rekindled
- A spike whose page drew an even bigger crowd later in the same month — usually a second event, like a golfer's spike in March followed by winning the Masters in April. Its "decay" would really be measuring the second event.
- Calendar pages
- Pages like "Deaths in September 2025" fill up over their month instead of fading.
Limits worth knowing
Counts are per page title: views of a redirect (an alternative title) are counted separately. Some automated bursts still get through when the page had a little traffic beforehand. Spikes too close to the end of the data (September 2026) are not given a half-life, since their month isn't over. For events people saw coming, the clock starts with the build-up, which makes them look slower.