This study is the biggest bunch of pseudoscience I've seen in ages. 18 participants, of which 17 knew which group they were in. Even if that was 10000 participants, that ratio means that the test is completely invalid!
I didn't read the study in depth, but the problem you describe is widespread and well-known. Well-known doesn't mean solved, so perhaps you have a better solution?
One option might come from clever experimental design, in which subjects do not know the hypothesis you wish to test. The problem is that this is often guessable - i.e. here subjects might not have known that AP was being tested, but they might have believed mental performance was, and if they thought they were in the treatment group they might have extra confidence, which could confound the results. That's hard to design around in practice, though.