The types of "neurotargeting" describes in the article - using data from emojis, personality tests, OCEAN models, fMRI studies - are all very noisy compared to what actually does work in advertising. You perturb the ad shown, measure conversions, and then just directly optimize for conversions. It's a variant of A/B testing described in the article, but you aren't comparing just two variants and having a human decide based on metrics, you're comparing thousands of variants and then training a machine learning model to predict conversion events.

All of the major online ad networks - Google, Facebook, etc - actually function like this, and have for the past 15 years. And it is effective to the tune of about $300B/year, looking at Google's ad revenue.