I’ve written previously about the difficulties (impossibilities) of predicting whether content will go viral or not.  Nevertheless, the prospect of your brand unleashing a Harlem Shake remains tantalizing enough for many to succumb to the charms of marketing agencies that guarantee viral results.

With an hour of video uploaded to YouTube every second though, the success to failure ratio is quite incredibly small, and it’s very likely that your viral masterpiece will languish in the depths, with a relative minority of people tuning in to watch.

Obviously a major part of predicting the virality of content is understanding why some content trends whilst other content tanks.  Research led by Jed Hofman from Microsoft Research set out to measure the virulence of online content.

Hofman and his team recorded every tweet containing a link to the 40 biggest websites over an 18 month period.  To put that into perspective, that represents over 1 billion pieces of content across sites such as YouTube and the BBC.  They then fine tuned that list down to only the content that was linked to by at least 100 different feeds.  That narrowed things down to 300,000 bits of content, which between them generated over 1.4 billion tweets.

By then analysing this content in detail it allowed them to build a picture of how content spread from person to person.  How closely connected were people when they shared content for instance.

The end result was that each piece of content was rated according to their virality, with a final score awarded out of 100.  To give this some context, only a handful of the billion or so pages they originally analysed scored 100 points.

Viral trends

So what interesting trends emerged from the report?  Is it possible to predict whether something will start trending?

One interesting finding was that most content only survives for a few ‘hops’ in the network.  In other words, content might be shared by some people, and some in their networks might share it too, but after that it tends to peter out.  Content that persists through 20 generations is literally one in a million.

They also confirmed something I wrote about last year, ie that a lot of people are all too happy to share something on Twitter without actually clicking through to read/watch what it is they’re sharing.  An example they provide is that of a YouTube video that went viral in terms of Twitter mentions, but the video itself only generated tens of thousands of views.

The research forms part of a new application by Microsoft called ViralSearch.  The app aims to visualise the virulence of content in a couple of ways.  The first is in the form of a family tree of all those who retweeted the content.  The second is in circular form and shows the number of people reached on Twitter.  Users can then drill down to compare the popularity and virulence of different tweets linking to the same content, say, or focus on various tweets posted by a single user.

It seems logical that such technology will eventually become part of Bing, allowing users to search for what’s hot online.  Suffice to say though, whilst the research was able to showcase what content was going viral, they weren’t able to pin down exactly what it was that sent content viral.  For the time being, that will remain very much a mystery.

I’m very much in the sceptical camp when it comes to viral marketing.  That’s not to say that I believe that the concept of an idea spreading in a viral way is wrong, merely that I don’t believe it’s something that we can predict or control.  I posted a piece back in 2011 about some Harvard research that cast doubt on the traditional notions of how ideas spread through a social network.

Some new research by Sharad Goel adds further weight to our understanding of just how ideas spread.  He kinda suggests that it doesn’t happen at all.

Goel and his colleagues studied seven different online scenarios to see how they spread:

1.  Yahoo! Voice, an online phone service started in 2004;
2. Zync, a Yahoo! Instant Messenger video-sharing application;
3. Friend Sense, a Facebook app introduced in 2009;
4. “The Secretary Game,” the online  version of a classic hiring test devised by psychologists;
5. Yahoo! Kindness, a charitable website launched in 2010;
6. News stories sent via Twitter in November 2011;
7. and Youtube links diffused through Twitter in November 2011.

The traditional thinking of viral memes is that they spread from friend to friend, which is known as multistep diffusion.  Goel’s data reveals that to be a complete myth.

“What we see is something qualitatively different. Most of the time it adopts and dies out within one generation,” Goel says.

For each of the seven scenarios examined, just 6% of them managed to get passed along more than once.  In other words, this notion that our content will magically spread throughout the six degrees of seperation that ties together the entire world is little but a myth.  If we’re lucky it might get seen by a friend of a friend.

The data suggests that a more realistic target to aim for is 20% spread.  So if you reach 10 people with your message, it’s realistic that another two people will then get to hear of it, which isn’t bad.

Of the outliers that were uncovered through the study, all were found to achieve their success through a traditional broadcast approach than through anything approaching viral.

Goel is following up his initial study with another that will look at what makes things popular online.  It’s currently under peer review and is expected to be released in March.  For anyone looking to spread an idea, product or piece of content it should be compulsory reading.

In the meantime, the following video should be well worth watching.