What happens when you tweet a link?

Summer 2015

Last month Ged Carroll ran "An experiment on fake Twitter followers". Over the past few months I have been running a very different Twitter experiment: "What happens when you tweet?".

Before we look at the data a bit of background.

Back in 2010 we asked the question what is the value of a Twitter follower? The answer back then was not much. We found no correlation between follower count and influence. Or at least influenced measured in page hits from Tweets. What we did find was a high probability that followers would Retweet a link, without actually looking at the page, as a signal of support for the Tweeter.

The observation being Twitter metrics are a signifier of success at the game of Twitter and little else.

That was 2010. By 2013 we observed a different trend. Seemingly a Tweet generated more traffic (i.e. page hits) but less retweets. The question being had Twitter magically pivoted into a link engine. A Google 2.0?

This then was the purpose of this experiment. An ad-hoc evaluation of that simple question: What happens when you tweet? Is all the excitement that you generate for real or is it, much like Ged's 50,000 followers, fake?

Throughout February we conducted 18 experiments of various scale and scope. The tweets linked back to posts here on the new platform and the old blogger platform.

At the end of those simple experiments we deduced that the average ratio of machine generated activity to human activity across the platform was between 6:1 to 14.5:1.

The evidence suggests the overwhelming majority of traffic generated by Twitter visiting both sites is probably Bots tracking the activity of Twitter users.

It also suggests a significant inflation of traffic numbers with each subsequent RT.

You could describe it as social inflation. Because it suggests that the social media traffic numbers of popular blogs are skewed by perhaps as much as 12x published figures.

And this before the publisher embarks on any ratings boosting activity like fake followers.

So a popular blog reaching 100,000 uniques per month, but heavily reliant on social media channels for traffic, may in fact only have a monthly audience of less than 40,000 or even 10,000.

Out on the long tail this suggests a blogger doing 100 hits a day is more likely to be doing only 40 or less than 10.

Either way the inflated number encourages the "wannabe global social influencer" to continue with their typing even though their non machine generated readership is marginal.

Essentially the premise tested by the experiment was: Is the social media landscape gamed by bots tracking the activity of real people? Does this data sample it supply a substantive proof? No. But the evidence is there to suggest further investigation of the "big data" would provide a revealing insight as to why social media is such an expensive and largely ineffective media/marketing channel for the majority of advertisers.

This is why we encourage all of you to conduct similar experiments for yourself to discover what is happening each time you Tweet.

To wet your appitite here is some raw data extracted from the experiments we have conducted over the past 30 days.

We'll begin with a cross platform experiment we conducted with @frabcus.

Here is the record of the first 24 hours of activity on Twitter

Here is the traffic generated on the site by Twitter in the 15 minutes after the original Tweet to @frabcus. Note these are events, unless indicated in orange, that did not load the whole page (i.e. Images, etc)

Here is the traffic generated by Twitter in the 15 minutes after @frabcus's retweet.

Our server records indicate over the first 24 hours we registered 12 inward arrivals from Twitter and 102 "Page Hits" at a Ratio of 8.5:1. In comparison Twitter registered only 4 outward clicks from 449 tweet views at a server page hit ratio of 25:1

Next we have the record of activity on Twitter for the blogger post.

That tweet, plus @frabcus's retweet generated 42 page hits on Blogger. 12 of which were linked to the tweets.

You will note the lack of correlation between Twitter analytics and Blogger's analytics. Hence the trigger for the question: Are social stats inflated by machine traffic?.

This second example was generated with the help of Ged Carroll.

Here is the record of activity on Twitter

Here is the traffic generated by Twitter in the 15 minutes after the original Tweet

Here is the traffic generated by Twitter in the 15 minutes after @R_C's retweet

The server records indicate the Tweets generated 38 page hits within the first 225 minutes of the original tweet. Twitter suggests there should have been only 2.

Make of this quick snap shot of data what you will but it provides a brief insight into the quality and currency of what we have come to describe as social media traffic today.

Not so much a real time marketplace of you, watching me, watching you but of machines and algorithms watching over us with... how can we best describe it?... loving grace?

Or would that be a very special interest?


Over a 3 day period we reran the experiment by repeatedly sharing this post on Twitter. The traffic generated by this activity delivered a new set of data that we can share with you. Unfortunately the amount of data generated does not permit a complete posting of the human:machine activity generated by the experiment. But here is the headline summary of the experiment.

Total number of page views: 298

Total number of page views attributed to Tweet Links: 26

Total number of page views attributed to Facebook Links: 5

Total number of page views attributed to LinkedIn Link: 1

Total number of page views attributed to other sources (e.g. Flipboard/Rebelmouse): 25

Residual page views provisionally attributed to machine activity: 241

Ratio of provisionally attributed machine generated pageviews: 5:1

The experiment was conducted in waves of tweets. Here is a breakdown of the first 15 minutes of activity generated by each new wave.

The first wave consisted of 3x targeted tweets.
The response was 1 Flipboard, 2 Rebelmouse and 38 machine events.

The second wave consisted of 3x targeted tweets + LinkedIn.
The response was 1 LinkedIn, 1 Tweet, 1 Other and 19 machine events.

The third wave consisted of 1x universal tweet.
The response was 27 machine events.

The Fourth wave consisted of 1x universal tweet.
The response was 23 machine events.

The Fifth wave consisted of 1x retweet by a 3rd party.
The response was 9 machine events.

The Sixth wave consisted of 1x retweet by a 3rd party.
The response was 1 Tweet and 4 machine events.

The Seventh wave consisted of a complex sharing event by 3rd parties the sample being extended by 20 minutes to accomodate the initial burst in activity and the creation 2 new tweet links.
The response was 5 Tweet, 3 Facebook, 8 Other and 57 machine events.

The Eighth wave consisted of retweeting the 2 new links generated during the complex event.
The response was 12 machine events.

The obvious question is what is a machine event? The broad definition here is a partial page view. e.g. an event that pulls just the page header or gets the page sans graphics.

What does the data prove? That tweeting a link triggers multiple machine events. These machine events are unique to the individual account sharing the link. Some of these machine events leak into social media statistical data. How do we know this? Because each time we tweet a link to a post on our Blogger platform it generates between 9 to 13 "page views" and yet a simultaneous tweet to our server here generates just machine events.

Are the results a substantive proof. Hardly. The sample here is far too small. Are they worthy of further analysis? Probably. Indeed for vested interests the answer would be a definite yes. But we will leave it up to others to qualify and quantify our test results by running their own experiments.

Copyright 2015 Digital Partners Pty Limited. All Rights Reserved.