One of the things that immediately flashed it as fiction for me was that the graphs had all the same shape, which if you've ever looked at trend graphs you will see they might all have a similar outlier quality to them but they build and sustain in different ways.
At Blekko (a search engine company) we did an interesting study on query traffic to see if you could "predict" the "hallmark" holidays based on search queries. The idea was that holidays like Valentines Day come up, people start thinking about plans or gifts before that, could we advise an advertiser when the "peak" planning session was so that they could maximize the impact of their advertising spend by focusing it during the peak? And if so what sorts of queries were people making that indicated they were doing holiday planning? The results were mixed. For things like Valentines it was easy, flowers, chocolates, bed & breakfast reservations sort of rose out of the general query stream, St. Patrick's Day? Not so much. But the data peaks all had different shapes appropriate for different levels of impact (Christmas shopping really starts in August among the back to school traffic for the really prepared). So looking at (and for) "interest spikes" like the ones in the story had a bunch of different shapes, some with slow onset and rapid decline, some with rapid onset and rapid decline, and some which were like soft swells on a breezy afternoon at the beach.
That said, the dataset made possible by Facebook's chat stream would be even better for those sorts of investigations.