Chapter 25: Calculating and Improving Your Twitter Click-Through Rate
Editor's Note: Seeking to improve the click-through rate of tweet links, Rand walks readers of The Moz Blog through a manual process for analyzing Twitter data. He notes that while the process is interesting, it is not scalable, and challenges readers to create a robust “Twitter Optimizer” tool. That was November 2010. Flash forward to today. Marketers now have a dizzying array of Twitter Optimizers to choose from, including Buffer, HootSuite, Sprout Social, Social Bro, and Raven Tools, to name a few. Moz has even acquired a Twitter Optimizer of its own, Followerwonk, which helps you increase your Twitter ROI by analyzing your social graph. Who would have thunk?
As marketers, many of us leverage Twitter as a direct traffic tool. We use the service to share URLs to increase brand awareness, help increase site visits, and possibly drive some direct actions (for example, sign-ups, sales, or subscriptions). From what I've seen and experienced, not many of us spend time thinking about how to improve the click-through rate (CTR) of the links we tweet.
Given that I have 21K+ followers, but most of the links I tweet generate 150-250 clicks, my CTR is only averaging 1.34 percent.
As analytics junkies, we're well aware that we can only improve things that we measure, analyze, and test. So let's look at a process for measuring our tweets, analyzing the data, and testing our hypotheses about how we could improve our CTR. If we do it right, we could make Twitter a more valuable marketing and traffic channel for our brands.
First off, we're going to need some data sets that include each of the following:
Profile Data
# of followers
# following
# tweets
Avg # tweets per day
Tweet Data
# clicks
# retweets
Time of day
Tweet structure (e.g., text, URL, text vs. URL, text VS text, URL vs. text, URL hash tags)
Editor's Note: Tweet Data only applies to tweets containing a unique, trackable URL such as those produced by the URL shortening service bitly—see http://bitly.com.
This data can be time-consuming to grab. If you know how to use Twitter and Bitly's APIs, though, you can make an automated system to monitor it. Once you have your data, you'll want to build a spreadsheet that looks something like this:
I've made the version I created for my own stats public on Google docs (see http://mz.cm/ZETblg).
With the help of my Twitter history page and Bitly, I constructed a chart of my last 25 tweets containing links that I personally created. (I did not include retweets, nor tweets containing links created by others in this chart, as they were irrelevant to this particular exercise.)
Using this data, I can find the answers to some very interesting questions, discussed in the following sections.
Q: Do My Wordier Tweets Earn Higher CTR?
To answer this question, I merely need to compare the number of words per tweet against CTR, and build a graph to visually illustrate the data.
The trend lines (the straight, dashed lines) are showing me that there's a slight pattern, and Excel's correlation function returns a value of -0.262, suggesting that there's a very subtle correlation between shorter tweets and more clicks. I might try testing this in the future with particularly short tweets, since my average word count is 15.88 with a standard deviation of only 3.88 (i.e., most of my tweets are consistently lengthy).
Q: Do My Shorter Tweets Perform Better?
Let's look at the raw length of the tweet. According to HubSpot's data, shorter tweets are more likely to be retweeted (see http://www.slideshare.net/danzarrella/the-science-of-twitter).
Does a similar relationship between shorter tweets and CTR exist?
According to my analysis, the results are similar. The relationship is actually a little stronger here. The correlation is -0.335. This again, suggests that shorter tweets might be getting higher CTRs. My average tweet is 108.92 characters in length with a standard deviation of 16.94. Given this extra data point, I'm certainly tempted to focus a bit more on brevity when composing my tweets.
Q: Do On/Off-Topic Tweets Affect My CTR?
In order to find out whether the topic focus of my tweets has an impact on the CTR, I had to assign a numerical value to each degree of “on-topicness,” and then map them each URL accordingly. My Twitter profile says that I tweet about SEO, startups, and technology. Since I'm in the SEO field, and the majority of my tweets are about these subjects, I created this scale:
0 - Completely unrelated topic
1 - Topic subtly related to marketing, technology, startups, and SEO
2 – Topic is strongly related to marketing, technology, startups, and SEO
3 – Topic is specifically about SEO
I then made the following chart to compare topic focus against CTR:
The correlation function is a bit higher, 0.43. This suggests that when I tweet about the topics people expect to hear from me about, the CTR is higher. That's not unexpected. In fact, I would have predicted a higher correlation. And who knows? Across a larger dataset, it might have been stronger.
Q: Is My CTR Improving Over Time?
This is a pretty simple question to answer.
Sadly, that answer is no. I hit my CTR peak in early October with a few choice tweets, and haven't had many in the high ranges since that time. This is a good lesson for me. It shows me why it's important for me to be monitoring, testing, and working to improve my performance.
On a broader scale, we also recently analyzed 20+ Twitter accounts containing hundreds of tweeted URLs. Our raw dataset contains about 250 tweeted URLs with data for CTR and several other metrics (see https://spreadsheets.google.com/pub?key=0AjlLl1iDXp81dDU4VDVVT0pmSGUtZjlyMmtHVzVkUmc&hl=en&output=html). Our hope was to see whether any of the metrics could help predict a higher versus lower CTR.
The following chart illustrates the findings.
Overall, no single metric was particularly predictive of higher CTR, with the exception of Twitter Grader Rank. However, in this case, a higher numeric rank (meaning a “worse” rank) had a higher correlation to CTR, suggesting an awkwardly inverse relationship. I was also bummed to see that Klout scores, which I had hoped would be predictive of CTR, barely correlated.
One interesting thing I found was that the average CTR for all 250+ tweets was only 1.17% (with a 0.024 standard deviation.) Therefore, I shouldn't feel too bad that my average CTR is 1.34%.
The research, unfortunately, didn't lead to any great conclusions. The full report is available for download at http://mz.cm/ZETsVj.
While this type of analysis can be interesting, it's not a scalable or practical solution for most marketers. What we need is a tool that can analyze our Twitter accounts using more metrics in an automated fashion. That tool doesn't exist … yet.
P.S. Special thanks to Ray Illian for compiling the research and report discussed in this chapter.