Hello and welcome to Digital Literacy Education. If you see this page, you were redirected here after clicking on a link that you were shown after watching an ad online. On this page, you can learn more about the purpose of the ad you watched and the scientific research behind it.
Digital Literacy Education is a project by Mythos Labs and the University of Cambridge, supported by the Alfred Landecker Foundation, Google Jigsaw, and others. We are trying to find out if it’s possible to improve people’s ability to spot persuasion techniques commonly used online. This idea is grounded in inoculation theory, a framework from social psychology which posits that it’s possible to build psychological resistance against future unwanted persuasion attempts through “prebunking” (or pre-emptive debunking). To learn more about how this works, click this link.
As part of this project, we’ve created a series of educational videos, each explaining a specific persuasion technique or logical fallacy commonly encountered online, such as using emotionally-charged language, spoofing, or perfect solution fallacy. Please click this link to watch the videos.
To test whether these videos actually improve people’s ability to spot potentially negative persuasive content, we ran a series of randomised controlled studies in a laboratory setting, with very positive results: compared to a control group, study participants who watched an educational video became significantly better at discerning persuasive social media content from the non-persuasive social media content, were more confident in their ability to assess the quality of information on social media, improved in their ability to identify trustworthy content, and indicated being less willing to share negatively persuasive content with others. The results of this study were published in Science Advances: https://www.science.org/doi/10.1126/sciadv.abo6254.
For this specific study, we were interested in finding out if the videos are effective at improving people’s ability to spot manipulative social media content if you run them as an advertisement online. To do so, we bought ads to show these five videos to a random sample of Indians online who met the following criteria: 1) 18 years or older; 2) Indian resident; 3) Hindi/Marathi/Telegu-speaking. Within this sample, people were randomly shown either the prebunking video or a control video as an ad. A random subset of people who were shown either video were subsequently shown a single survey question, where they were asked to identify which persuasion technique is being used in a fictional social media post. In total, we created 6 such social media posts, which we stripped of all source and other identifying information. Half of these posts (3) were phrased to be persuasive, whereas the other half were each persuasive post’s neutral (non-persuasive) pair. You can see an example of a persuasive post and its neutral pair below.
As you can see, both posts are related to cricket. However, the persuasive post (on top) contains words such as “cricket superstar” which is a type of identity-mimicking phrasing that research shows can increase the viral potential of online content. The post on the right is worded to be much more neutral and does not use language that mimicks anyone’s identity. In total, we created 3 such manipulative-neutral pairs.
In this study, we are testing the hypothesis that participants who watched the prebunking video as an ad are significantly better than a control group (a random group of users that meets the recruitment criteria and saw a different video) at correctly identifying the use of a particular persuasion technique in social media content. We will update this page once we know more about whether these hypotheses were confirmed.
If you answered a survey question after watching one of the above videos, your response will be recorded and used in our study. To ensure your privacy, we do not record any usernames or other personally identifying information. We are only interested in study participants’ responses to the survey questions as part of our scientific research and therefore do not collect any demographic or other data. Because we do not record any identifying information (and therefore can’t match responses to any individuals), we are unfortunately unable to remove any responses from our dataset upon request after they have been recorded. This study was reviewed by the Cambridge University Psychology Research Ethics Committee. Should you have any questions or concerns about this study, please contact the coordinator of this study, Dr. Jon Roozenbeek (jjr51@cam.ac.uk).
