The main paper they're basing the headline on is this one (kudos to them for actually linking it): http://conference.nber.org/confer/2013/EoDs13/Tadelis.pdf

That paper concludes that a substantial proportion of conversions via search-engine ads seem to be sales to loyal users who convert via another channel even when you take away the ads, so the conversion rates themselves may be significant overestimates if what you really want to measure is the net impact of advertising.