At an abstract level:
If one can technically verify level of belief, then one can both insist upon it in coercive capacities (the NSA does not want any employee who has any doubts about its holy mission), and tune arguments for it in non-coercive capacities (The breakfast cereal marketing department's learning algorithms will be able to test jingles against a fully verifiable 100% accurate human model).
Blowing things off and consenting because you just want to move on, ceases to be an option once it's detectable. Instead, the opponent can iterate on strategies of persuasion until the job is genuinely done. This doesn't produce an easily swayed populace for the same reason an unsecured computer doesn't produce an egalitarian botnet host: It is in the interest of every memetic attack to install defenses against other memes. Will you tell your new girlfriend, that her brown hair looks nice when your preexisting bias in favor of blondes has been hyped up into a whole obsessive fandom by your last girlfriend, who jealously insisted on the mental calcification of True Belief?
Mindreading in the form of an accurate lie detector of some sort is a large portion of mind control, once you extrapolate it to routine applications throughout society.
.
More specific to the article's point:
Datamining truth verification is more likely to be a problem for us due to the asymmetry: It benefits large organizations and only large organizations. It benefits the people with the best AI to search your history and only the people with the best AI to search your history. The NSA is already capable of performing a background check, utilizing the whole of its resources constructively, that would point out any doubts you have; But we are not similarly capable of deducing facts about the NSA.