The full paper is: https://go.nature.com/2N7K4cL

Online advertising systems offer their clients fine-grained information on the effectiveness of their campaigns. Because online experiments can be conducted rapidly and at low cost, advertisers can test a multitude of new ideas and then refine them based upon how different consumers respond to different ads. When advertisers have information on a given user, it becomes possible to tailor ad content to that person’s tastes and preferences. Such capabilities need not be targeted only toward sales of consumer goods. They can also be used to convey and tailor public health messages. In this sense, individualized public health messages are somewhat analogous to personalized (or “precision”) medicine. 

In precision medicine, doctors can combine information about one’s demographic, genetic, and social experiences to tailor a given therapy such that treatment efficacy is maximized. Similarly, we set out to explore how user information, including his or her search history, determines how he or she might respond to different advertisements that were designed by JWT, a professional ad agency. Our goal was to utilize a system designed to create one behavioral change, e.g., purchasing of products, to create another behavioral change – a healthy lifestyle choice.

In our paper we describe the design, execution, and evaluation of an advertising campaign intended to encourage healthy living by reducing food consumption and increasing exercise. Two advertising executives and members of their creative team designed 20 different ads. We then showed the campaign ads to almost 800 thousand people who used Bing, an internet search engine, to search for terms associated with social stigma or diseases related to poor diet or low levels of exercise. The advertisements were designed to convey the need for behavioral change but do so in a manner which would be easy, actionable and fun. People who clicked the ads were directed to a website which provided more information through short movies featuring cartoon characters.  

One key advantage of advertising systems which are linked to search engines is that it is possible to measure the effect of an advertisement not only in terms of the reaction to it but also in how they modify the topics for which people search after seeing the ad. This is important, because searches reflect people’s behavior in the physical world, and so provide a measure of the induced behavioral change. Thus, in our work we tracked several thousand anonymous users to evaluate how their searches changed as a result of the ads we showed them, compared to people who searched for the same terms which would have made them eligible to see the ads, but the ads system chose to show them other ads. Our analysis shows that the rate at which ads were clicked varied significantly by the choice of the words shown in the ads, as well as by the demographics of the people who were exposed to them. More importantly, we were able to show that 48% of people who were exposed to our ads made future searches for weight loss information, compared with 32% of those in the control group — a 50% increase. We also saw a large difference in effectiveness among our advertisements, and that much of this difference could be accounted for in the lifestyle preferences and sociodemographic characteristics of individuals.

Thus, our results demonstrate how public health campaigns can be designed, fielded, and tuned in a short period of time and at a low budget so as to create highly effective, personalized, messaging to induced behavioral change.