This article will no doubt feature highly in the blogs (click here) and create much controversy! It reminded me of the Isabel decision support tool.
Basically, the authors took a years worth (n=26) of diagnostic cases from the NEJM and, after reading the article, they selected 3-5 pertinent search terms. Added these to google and somehow selected the 3 most likely diagnoses. They then compared these 3 diagnoses with the actual result from the NEJM.
Overall result - 'correct' in 58% of cases (CI - 38-77%).
I'm not sure if I find this disturbing or not, it's certainly too small to make conclusions. But once again Google (not deliberately on their part) have set a standard.....