A Site That Went Viral and The Numbers Behind It

We engineered a viral site to understand how memes spread online. This is the full report.

It is an unfortunate but understandable reality that while we often marvel at digital projects that spread like wildfire across the web, we rarely get a chance to look at the numbers behind them. Between January and May of 2010, we built, launched and monitored Jerzify Yourself to get just such a glimpse into the dynamics of spreadable content.

This is going to be a long post, so if you are in a hurry, check the summary published in AdAge or scroll all the way down to the list of the most important things we’ve learned from this experiment.

The Setup

Jerzify Yourself was created in January a week after the first season finale of a certain MTV show that had attracted an audience of 4.8 million. The site, written in a few days in Flash, provides a familiar attraction of uploading one’s headshot onto a stylized body, and is packed with references recognizable by the show’s viewers. Or, in the much more lively words of Village Voice: ”The gist is Snooki-grade simple: upload a medium-sized jpg, scale the image to fit, choose your spraytan shade, pick your pose — and holy Freckles McGee, you’re magically recast as a human meatball.”

Tools
To monitor traffic and site activity, we used Google Analytics and GetClicky with its real-time traffic statistics and a smartphone-friendly interface. We tracked pass-alongs with a tool created by our friends at Meteor Solutions. We used an assortment of free counters to look at tweets, bit.ly links and the like. We have also built a Jerzification Live tool that showed a stream of pictures uploaded by users in real time.

Seeding
The site went live on Thursday afternoon, January 28, 2010. We had not bought any media to drive traffic to it, but we did spend some time to alert others about the project. Most significantly, we created an “event” on Facebook inviting some 800 of our personal friends to jerzify themselves, and posted a link on BuzzFeed that has since been viewed 3,150 times. We have also set up accounts on Tumblr and Twitter; we used the latter to retweet mentions, share the more amusing jerzifications, and reach out to fans of the show, producing a somewhat moderate 190 tweets in two months and attracting 83 followers.

The Pick-Up
Jerzify Yourself was picked by some of the bigger online outlets roughly in this order:

College Candy (Jan 29)
@petewenz (Jan 29)
Huffington Post (Jan 29)
NY Mag (Jan 29)
Glamour (Jan 29)
Gorrilamask.net (Jan 29)
@PerezHilton (Jan 31)
TV Guide (Feb 1)
Village Voice (Feb 2)
MSNBC (Jan 31)
Comedy Central (Feb 1, Tosh.0 via BuzzFeed)
College Humor (Feb 2)
LemonDrop (Feb 5)
Radio shows KISS 92.5 in Toronto (Feb 8.) and GNI 102.7 (undated) in Wilmington, NC
Maxim (Feb 8, via LemonDrop)

The First Week: Celebrities
In the 18 weeks between January 28 and May 19, Google Analytics registered 158,575 unique visits (85.73% were new visits), 242,514 page views, and 02:04 average time on site.

The visits graph looks like a typical “long tail:”

The traffic peaked on Sunday, the fourth day after going live, with 22,717 visits. This, to me, came as a surprise; I was expecting more of a hockey-stick graph in the beginning with a more gradual build-up.

The week between the launch on Thursday (1/28) and the following Wednesday (2/3) accounted for 58% of the visits (93,069). These visits were referred by 99 sources who drove the bulk of the site’s traffic; the total number of referrers for the entire period was 367. The referral stats are somewhat under-reported, as some sites published the site’s URL without actually linking to it. Others, most notably Maxim.com, linked to their “via” source of the news instead of the site itself (that is, Maxim linked to a lemondrop.com post about Jerzify instead of linking directly to the site).

The top six referring domains during the first week were:

Collegehumor.com (12,160 visits)
Twitter.com (10,510)
Gorillamask.net (5,797)
Facebook.com (5,535)
Buzzfeed.com (1,536)
Huffingtonpost.com (1,415)

Click-Through Rates on Celebrity Tweets
The statistics give us some insight into the intriguing topic of Twitter celebrity endorsements. Assuming, conservatively, the combined Twitter audience of @petewenz and @perezhilton was at 2.5 million at the time, and assuming for simplicity that no other account has tweeted the link, the 10,267 visits from Twitter amount to 0.4% click-through. Assuming that only 20% of all Twitter visits come from the website (as opposed to desktop and mobile clients), the total Twitter celebrity click-through hovers around 2%.

Of course, the celebrities’ influence isn’t limited to the immediate clicks they spur; it would be equally interesting to calculate the pass-along value of such endorsements to account for all visits from the sites of the celeberities’ followers. While we don’t have the exact data just for the celebrity-inspired pass-alongs, we do share our thoughts on this topic further below.

Search Engines
The traffic sent by search engines amounted to 5,955 visits from Google and 255 from Yahoo — about 4% of all visits — with the bulk of search words being different variations of the URL. We haven’t been tracking the site’s SERP rankings, but it appears that the site was indexed and ranked sufficiently high for “jerzify yourself” on the day after the launch.

The Long Tail and Conversions
I found it interesting that once the traffic hit the tail part of the Long Tail, it remained pretty steady. In the most recent month between April 18 and May 19, the site averaged 142 daily visits with only a slight decline over time with no effort to attract attention to the site on our part.

This is the site’s “conversion” ratio — the number of picture uploads (from server logs) to site visits. The conversions go up even as the site traffic trends down. If photo uploads were our metric of traffic quality, we could argue that the massive early traffic from the first week of heavy linking is actually of lower quality than the “tail” traffic that followed.

The Shelf Life of a Link In Social Networks
The heaviest referrers during the April-May period were Facebook (657 visits) and Google (~400), with Lemondrop.com (147) and Huffingtonpost.com (79) trailing far behind. Twitter referred nine visitors.

One possible conclusion after comparing the referrers during the first week and the last month is this: Links have a much longer gestation period, but also a longer shelf life on Facebook than on Twitter. For the entire January-May period, Facebook has referred 12,789 visitors, 83% of them after the first week. Twitter has referred 10,549 visitors altogether, 97% of them during the first week.

Measuring Virality
To understand how the word about Jerzify Yourself was spread around the web, we partnered with a Seattle company called Meteor Solutions. Meteor’s technology creates a unique URL for each person who visits the site by appending a string of characters at the end so that jerzifyyourself.com becomes http://jerzifyyourself.com/?fbid=zu1hD2nJUMK, for example. If this unique URL gets reposted somewhere online and subsequently clicked on, Meteor takes notice. When people click on this tagged URL and come to the site, they in turn receive their own unique URL with a different string of characters. Meteor follows the entire sharing chain.

One important thing that Meteor can tell us is the total impact of having a link posted on a popular website. The example below shows the number of visitors to Jerzify that can be attributed to LemonDrop.com.

The 4,051 “Direct Visitors” are the people who arrived from LemonDrop.com during the 18 weeks of the project but did not click on a tagged link. No wonder they didn’t — LemonDrop has stripped the Meteor tag off the link before posting it. This resulted in zero “WOM Visitors” — people who arrived (or, rather, didn’t) to Jerzify via a tagged link.

The “Yield” number is the one we are really interested in. It shows the number of people who clicked on all those unique URLs that Meteor had generated for the 4,051 original visitors from LemonDrop who then shared them elsewhere online. This “yield” for LemonDrop is 3,773. In other words, people who were sent to Jerzify by LemonDrop have brought almost as many people to Jerzify by reposting the link elsewhere. The “Total Impact” number is the sum of direct visitors, WOM visitors and yield, and LemonDrop can be credited with sending 7,824 visitors to the website, first directly and then via subsequent re-sharing by the original visitors.

We used the Meteor data to look at the first week’s top referrers from another angle, ranking them not only by the direct visits they referred, but also by the sharing activity they sparked. In the table below, “Visitors From Site” are direct visits plus WOM visits, to use Meteor’s nomenclature. “Visitors From Sharing” are the yield, and “boost” is the size of yield relative to the total impact.

What We Have Learned

While Jerzify Yourself has not reached the stratospheric heights of the meme pantheon, as an experiment it was successful in adding the much-needed texture to our knowledge of how content gets passed along online. Here’s what caught our attention.

The Invisible Impact. If you find yourself measuring the value of referral sources for your campaign, consider their total impact via re-shares in addition to the direct traffic they send your way. Counting only the direct clicks from any site is likely to underestimate the site’s total value; five out of six sites on our top-referrers list sent almost as much traffic through re-shares as through direct clicks. It would make for an interesting follow-up experiment to see if this difference holds up as solidly for paid campaigns as it does for “organic” content. If it does — and this difference is measured — it would have important implications on how we plan media buys.

2. If It Doesn’t Spread, It’s Halfdead. Dr. Henry Jenkins once made this now-famous remark about the destiny of content in the age of social media: “If It Doesn’t Spread, It’s Dead“. Having seen the data, we can now say with a degree of confidence that you’ll still get viewers if your link gets picked up by major online publications, but content that’s designed to be spreadable can nearly double the referred traffic through re-shares.

3. Some Sites Are Read By More Active Spreaders Than Others. Some sites on our top list turned out to be a lot more spreadful (for lack of a better word) than others. Buzzfeed in particular sent more traffic — twice as much! — via re-shares than through direct clicks. In fact, the number one direct referrer, collegehumor.com, will come last if we re-rank the sites by “boost”. These observations along with our understanding of the reasons behind the differences will influence the way we design online properties meant to encourage content sharing.

It would also be interesting to see if there is a difference in the length of pass-along chains between different sites. On average, the link traveled down two or three generations of users before the chain broke and we’ve seen chains as long as seven users, but we couldn’t produce a more precise analysis.

4. The Speed of Content Depends on The Medium Through Which It Travels. You know how the speed of sound depends on the medium through which it passes? It’s like that with Twitter and Facebook, and probably other social networks. To repeat an observation made earlier: in our experiment, shared links had a much longer gestation period but also a longer shelf life on Facebook than on Twitter.

For the entire January-May period, Facebook has referred 12,789 visitors, 83% of them after the first week. Twitter has referred 10,549 visitors altogether, 97% of them during the first week. This difference probably has to do with how people access the news feeds on these sites. On Twitter, the single stream of news quickly washes away older items. On Facebook, older news can still be front-page material on the individual, slower-moving walls. If you find yourself choosing between the two sites for your next campaign, be aware of this difference.

5. Don’t Reach For the Off Switch. As the 404 errors on formerly popular viral branded destinations demonstrate, it might be tempting to kill the destination site some time after the traffic has peaked. I’ve argued elsewhere that abandoning old microsites in their Long Tail phase means leaving money on the table, and our experiment has demonstrated that not only do off-peak sites attract healthy traffic, these visitors can also be more valuable than the rush-hour crowd.

Finally, we are well aware of the argument in favor of using “spreadable” instead of “viral” when describing projects of this nature. The irony is that, in our experience, blog posts about viral projects tend to be more spreadable than posts that use the correct terminology. How’s that for a social media Moebius strip?

Credits

The Jerzify Yourself site was created at Hill Holliday by Bob Gates, Rick McHugh (CD); Frank Cartagena (CW); Carissa DiCenzo (AD); Paul Lenzi, Lisa Belden, Maggie Foley, Kim Ryan (production); Dan Pearce, Greg DeMelo, Jad Mintun, Kelsey Meuse, Nathan Vey, Fred LeBlanc, Dominic Giangiobbe, Moya Hynes (development); Johnny Won and Steve Bagdasarian (seeding).