Can We Anticipate Viral Content Before It Goes Live?

What we know for certain is that people are at the heart of viral outcomes (unless you are leveraging bots or other automated methods to drive traffic, in which case, we can’t help you). From previous marketing research, we know that the emotions evoked in readers are critical to the likelihood a piece of content will go viral and that people are more likely to share content that is highly arousing. These findings are important, but how can you test whether content is likely to go viral without subjecting it to the scrutiny of the internet?

Typically, predictions around how content will perform are formulated using a combination of content topic analysis and self-report of measures such as “likelihood to share.” We decided to extend these metrics to include a known physiological measure, GSR, to see whether there was predictive value in how the body itself responded to pieces of our own agency’s content that had highly variable success in the field.

Why GSR? Electrophysiology measures have been around for a long time, but the technical and economic barriers to using them in marketing research were fairly high until recently. The average marketer was skeptical of their utility, particularly since these costly types of methods had not been definitively shown to produce a clear advantage over cheaper, more accessible behavioral methods. Over time, prices have decreased, and though these devices still require a strong technical understanding, they are not as difficult to use as they once were. Galvanic skin response, which measures the skin’s resistance to a very mild electrical current, has been demonstrated to be a strong predictor of emotional arousal, and emotional arousal is known to be a crucial ingredient for viral content.¹

Getting people to self-report their emotional arousal has long been an attractive option for researchers: These surveys are cheap and make it easy to collect data from a large group of people. However, there are several limitations to self-reported data. Telescoping, selective memory, and the availability heuristic are just three of the dozens of ways certain participants can obscure the results of your study. Not to mention the “bot crisis”—large-scale corruption of data validity from participants either not paying attention as they complete surveys or leveraging bots to auto-complete surveys with fake data—which has plagued the online platforms that marketers and academics utilize to recruit participants.

In general, people are just pretty bad at knowing how they feel and why they feel that way. Just look into the misattribution effect for proof. People constantly misinterpret their physiological state to the emotional context they find themselves in. This effect explains why you may think a fear response is actually sexual attraction or why a rainy day might make your bad mood worse. It is also why drinking coffee or watching horror movies makes great dates.

In our study, we examined content from 15 highly successful Fractl content marketing campaigns (average social shares = 21,358.07) and 15 of our low-performing campaigns (average social shares = 11.07). We compiled the graphics most prominently featured in the publisher coverage earned by each campaign and showed them to 22 participants. We used only most prominently featured assets for two reasons. First, this allowed us to include a wider variety of stimuli in the study; if we used entire projects with several images each, the number of campaigns sampled would need to be much smaller. Second, because of where these graphics were placed, they were the ones most likely to be viewed and shared by readers.

Prior to coming into the lab, participants were given a qualifying/demographic survey. Additionally, we collected self-reported interest in all the content verticals we explored (this included items like sports data, political data, health and wellness data, workplace data, etc.). We used a Shimmer GSR unit to gather galvanic skin response data.

After viewing each image, participants filled out a brief survey where they reported their levels of interest, enjoyment, surprise, and understanding of the content. We also asked how likely participants were to share each image and how engaging they found each image overall. Participants rated almost every qualifier on a 5-point Likert—a scale with an odd number of answer anchors that allow researchers to assess the attitudes of participants (for example, a five-star rating system for businesses). The question regarding how engaging the content was to them was rated on a dichotomous 9-point scale—a scale that ranges from one adjective to its polar opposite (for example, 1 for extremely boring to 9 for extremely engaging).

Our hypotheses were as follows:

1. The content in the highly viral group will elicit more emotional arousal in our participants, leading to greater electrodermal activity.
2. Participants’ electrodermal activity would be more predictive of viral outcomes if the participant had previously expressed interest in that content vertical.
3. The behavioral survey methods will be less predictive of viral outcomes than the biometrics.

What we found was: There was a significant difference between the high and low viral content for phasic galvanic skin response (phasic GSR refers to the portion of the data that corresponds directly to a participant’s response to a stimulus).

This effect was not reduced by whether or not participants reported being interested in a given piece of content: both “low interest” and “high interest” items showed the same pattern of galvanic skin response. There was no significant difference in the predictive ability of participants’ galvanic skin response if they were interested in the content vertical or not.

In plain English, this means that whether or not someone actually told us they were interested in a type of content, their body’s response still predicted how that piece of content eventually performed on the internet (see figure 7-1).

We also found that participants’ self-reported data—typically a cornerstone of marketing research—was not able to predict which content was eventually successful. There was no significant difference between the high-viral and low-viral content in terms of how participants rated their understanding, how likely they would be to share it, how surprising they found it, how much they enjoyed it, and how engaging they found it.

Can Virality Be Predicted?

The more viral a campaign was, the greater galvanic skin response our participants had while viewing it. This held true regardless of self-reported interest in the campaign vertical; in fact, all our self-reported metrics were much less useful than galvanic skin response. Factors such as how well our participants understood the content, how much they enjoyed it, how surprised they were by it, how likely they were to share it personally, or how engaging they found it did not correlate with how likely the content was to go viral.

FIGURE 7-1

Skin response to viral content categorized by participants’ self-reported interest

Galvanic skin response is a technique that measures emotional arousal based on the skin’s resistance to electricity. It was able to predict which content eventually went viral regardless of whether or not study participants found the topic interesting.

The intersection of electrophysiology measures and marketing is a new and exciting field. Introducing biometrics such as galvanic skin response, eye tracking, and EEG (electroencephalogram) to a market research repertoire may yield insights previously unattainable to marketers. Though we still cannot perfectly predict what makes content go viral online, neurometrics help us get a clearer view into the minds of digital content consumers.

TAKEAWAYS

What separates an article that goes viral from the rest depends on a single moment: a person deciding whether or not to click the share button. Being able to predict what will happen would be akin to discovering the holy grail of marketing research.

• According to research, self-reported factors such as how well participants understood the content, how much they enjoyed it, how surprised they were by it, how likely they were to share it personally, or how engaging they found it did not correlate with how likely the content was to go viral.
• Physiological markers, however, can be used to measure emotional responses to content. By recording an electrophysiological signal called galvanic skin response (GSR), researchers were able to predict the viral outcome of a piece of content significantly better.
• Introducing biometrics such as GSR, eye tracking, and EEG (electroencephalogram) to a market research repertoire may yield insights previously unattainable to marketers.

NOTE

1. 1. Dominik R. Bach, Karl J. Friston, and Raymond J. Dolan, “Analytic Measures for Quantification of Arousal from Spontaneous Skin Conductance Fluctuations,” International Journal of Psychophysiology 76, no. 1 (April 2010): 52–55, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2877802/; Kerry Jones, Kelsey Libert, and Kristin Tynski, “The Emotional Combinations That Make Stories Go Viral,” hbr.org, May 23, 2016, https://hbr.org/2016/05/research-the-link-between-feeling-in-control-and-viral-content.

Adapted from “Can Biometrics Predict a Viral Marketing Campaign?” on hbr.org, January 10, 2019 (product #H04OYU).