What makes content go viral?
Studies of news articles, an artificial music market and a billion diffusion events give three partial answers to what makes content go viral: the message, the first audience and chance.
Answer to the question
What makes content go viral?
Content goes viral through a mix of its own properties, the size of its first audience and chance. Research on news articles finds that positive, high-arousal and useful pieces are shared more, while a study of about a billion Twitter events found that popularity depended mainly on the largest single broadcast. Early reshare patterns help predict how large a cascade becomes.
Key takeaways
- A 2016 Management Science study of about a billion Twitter diffusion events found structural virality generally low, so popularity depended mainly on the size of the largest broadcast.
- Berger and Milkman (2012) analysed 6,956 New York Times articles and found that positive content and high-arousal emotions such as awe, anger and anxiety were more likely to reach the most-emailed list, while low-arousal sadness was less likely to.
- Their pattern held after controls for surprise, interest, practical value and home-page placement, and their three experiments (49, 45 and 47 participants) relied on self-reported intentions to share.
- Salganik, Dodds and Watts (2006) gave 14,341 participants songs in eight parallel worlds and found that visible earlier choices raised both inequality and unpredictability between hits and the rest.
- A 2014 study of photo reshare cascades found that temporal and structural features were the most useful for predicting final size, and that early breadth was a better signal than depth.
Two posts can each reach 1 million people and share little else. One is broadcast by a single large account. The other passes hand to hand. Both go viral in the ordinary sense.
How does a post spread?
Structural virality, a measure introduced in 2016, runs from a single broadcast at one end to long chains of person-to-person sharing at the other. Its authors, who included Duncan Watts, applied the measure to about a billion diffusion events on Twitter, covering news stories, videos, images and petitions.
They found that popular events spread through a mix of broadcast and viral mechanisms, in every domain and at every size. Structural virality was generally low and stayed low as events grew, so popularity depended mainly on the size of the largest broadcast. A model of low-rate contagion on a scale-free network matched several of the findings but did not reproduce the variety of structural virality in the data.
That moves the target.
In these data, the largest events were mostly shaped by their biggest single broadcast. Whether a post then passes onward through many hands is a separate event from that first broadcast, and chains of that kind stayed short on average as events grew.
Do content traits matter?
Content traits act on an audience that was set by other means.
Jonah Berger's book Contagious makes the case on the content side. He argues that ideas and products that go viral catch on largely because of properties they carry, and sets out six of them, from making the sharer look good to being practically useful. The best-evidenced part of that case is an observational field study that Berger ran with Katherine Milkman in 2012, and the same properties turn up in [[The Boring Product People Talk About]].
They analysed 6,956 New York Times articles from a three-month period and asked which reached the most-emailed list. Positive articles were more likely to get there than negative ones. Articles that evoked high-arousal emotions such as awe, anger or anxiety were more likely to get there than articles that evoked low-arousal sadness, the subject of [[Why Sad Posts Go Viral Less Often]], and practically useful articles were also more likely to make the list.
The pattern held after the authors controlled for surprise, interest, practical value and external attention such as home-page placement. Three laboratory experiments, with 49, 45 and 47 participants, pointed to a causal role for arousal, using self-reported intentions to share rather than actual sharing.
Exposure comes before sharing.
The home-page control deserves attention. An article that sits at the top of a front page is seen by many more people than one that does not, and a larger audience produces more shares whatever the article says. Any study of why a piece can go viral has to separate its traits from the attention the piece was handed.
The billion-event study points the same way. If the largest broadcast carries most of the weight, then which account posted, and who saw it first, matters alongside what was posted.
What do early choices do?
Matthew Salganik, Peter Dodds and Duncan Watts built an artificial music market for a 2006 test. They gave 14,341 participants songs by bands they had never heard of and varied whether participants could see how many times each song had already been downloaded. The social-influence condition ran as eight parallel worlds, so the same songs met different crowds from the same starting point.
Identical content can finish at very different heights depending on early, visible accidents.
Where participants could see earlier choices, the gap between hits and the rest widened, and which song won differed from one world to the next. Stronger social influence raised both inequality and unpredictability. Quality still counted: the best songs rarely did poorly and the worst rarely did well, but between those bounds many outcomes were possible. The mechanism is called cumulative advantage, which [[Why the Best Song Does Not Win]] follows through.
Because the market was made up and the measure was downloads, it cannot time any particular post, but it shows how visible popularity can steer later choices: identical content can finish at very different heights depending on early, visible accidents.
Whether the final size of a cascade could be foreseen was the question behind a 2014 study of photo reshare cascades on a large social network. Its models predicted with good accuracy whether a cascade would keep growing, and temporal and structural features were the most useful for predicting final size. Early in a cascade, breadth was a better signal of eventual size than depth.
Prediction improved as more reshares were observed, which means the models describe what becomes visible once a post is under way and not what could be known before it. For several cascades of the same photo that ended up at different sizes, the models could still pick out which would become the largest.
None of this shows that content can be built to go viral on demand, and craft in the message comes after the matter of who sees it first. A share is left to chance.
Common questions
What makes content go viral?
Content tends to go viral through a mix of its own properties, the size of its first audience and chance. A study of about 7,000 news articles found that positive, high-arousal and useful pieces were listed as most emailed more often, and a study of about a billion Twitter events found that popularity depended mainly on the largest single broadcast.
Is content that goes viral mostly luck?
The research does not say so. In an artificial music market, stronger social influence made outcomes more unequal and less predictable across eight parallel worlds, yet the best songs rarely did poorly and the worst rarely did well. Chance has a large role inside limits set by the quality of what is offered.
Can you predict which posts will go viral?
Partly, and only once a post is under way. A 2014 study of photo reshare cascades found that early reshare patterns, especially their timing and structure, predicted whether a cascade would keep growing and how large it would become. Nothing in these studies predicts the outcome before a post is published.
Sources
Berger, J. (2013). Contagious: Why Things Catch On. Simon & Schuster. Seed book.
Goel, S., Anderson, A., Hofman, J., & Watts, D. J. (2016). “The structural virality of online diffusion.” Management Science, 62(1), 180-196. doi:10.1287/mnsc.2015.2158 https://doi.org/10.1287/mnsc.2015.2158
Berger, J., & Milkman, K. L. (2012). “What makes online content viral?” Journal of Marketing Research, 49(2), 192-205. doi:10.1509/jmr.10.0353 https://doi.org/10.1509/jmr.10.0353
Salganik, M. J., Dodds, P. S., & Watts, D. J. (2006). “Experimental study of inequality and unpredictability in an artificial cultural market.” Science, 311(5762), 854-856. doi:10.1126/science.1121066 https://doi.org/10.1126/science.1121066
Cheng, J., Adamic, L. A., Dow, P. A., Kleinberg, J. M., & Leskovec, J. (2014). “Can cascades be predicted?” Proceedings of the 23rd International Conference on World Wide Web, 925-936. doi:10.1145/2566486.2567997 https://doi.org/10.1145/2566486.2567997