Article Level Metrics – how reliable are they? (I prefer to read the paper)

I am on the board of a wonderful community voluntary organization – the Crystallography Open Database (COD) http://www.crystallography.net/ . For 10 years it has been collecting crystal structures from the literature and making them Open – more than 300,000. It’s the only Open database for small structures (the others CSD and ICSD are closed and based on subscriptions even though the data is taken from public papers. This morning we heard of a great paper using the COD for data-driven research. Here’s the landing page http://www.mdpi.com/2073-4352/5/4/617 :

 
The paper is trawling through hundreds of thousands of structures to find those with a high proportion of hydrogen atoms – a clever idea for finding possible Hydrogen Stores for Energy. The closed databases couldn’t be used without subscriptions.
I think this is a clever idea and tweeted it. I’m a crystallographer and structural chemist so it’s not surprising that a few other people retweeted it.
However I noticed that the article had a daily count of accesses and that there was a small glitch of 3 accesses today. I tweeted this and – surprise – the accesses went up. after 12 hours there have been over 100 accesses

 
You’ll see there have been over 100 accesses today because I and 3-4 others have tweeted it. This is nothing to do with the contents of the paper because not many have actually read it today. People have clicked to view the graph, and every time they visit the graph goes up. It’s nothing to do with the quality of the science (which I think is good) or the fact that the paper is Open Access – it’s just a Heisentwitter.
So what does 100 accesses today mean? Nothing.
What does my opinion of the paper count? Something, I hope (I would have recommended publication).
The point is that to decide whether science is good or useful
you have to read the paper