Are you wondering about the Twitter ranking algorithm for top tweets in hashtag searches? ZoomOwl does a detailed analysis of various ranking factors related to individual tweets, tweeting accounts, geography, embedded items, engagement metrics, etc., and gives clear recommendations.
Is adding a popular hashtag to your tweet enough to reach a lot more people than your followers and get retweets, likes, and eventually followers? Well if that were true, everyone would have thousands of followers, wouldn’t they?
Go to Twitter, do a random hashtag search, and you are by default shown the Top tab.
However, all the tweets that use the hashtag go by default to the Latest tab. Only a few good ones are hand-picked and ranked by Twitter on the Top tab.
It got me thinking, what are the parameters with which Twitter ranks a tweet. I wanted to get to the bottom of it.
The important thing is, if your tweet doesn’t make it to the Top tab, it’s probably not even gonna be seen. So, you need the recipe to get to the Top tab. You can see tweets that were made hours or days ago in this tab, still attracting attention and engagement. So, it’s clear that hashtag alone won’t help your tweet rank.
In this article, I am gonna do an experiment to identify how exactly Twitter is ranking your tweets.
We are going to see extensive data that could corroborate our theories on ranking.
To identify the ranking parameters, I’ve looked at not one or two, but tens of different hashtags and calculated various metrics around them. The following Table 9.1 shows a list of the top 20 results for the hashtag “#IoT” that we will be using as a sample for analysis.
So then what might those ranking parameters be to get to the Top tab? The real algorithm behind Twitter’s ranking is protected like the recipe of the coke, unlike the timeline algorithm, but it must be dependent on a few things.
We believe the ranking of a tweet depends on the tweeter himself, the tweet, and the various parameters of the searcher.
We can look closely at each possible ranking factor:
- The tweeter: The level of influence
- The searcher: Geography, people following, etc.
- The tweet: Engagement metrics (likes, retweets, etc.), time of tweet, embedded elements (images, videos, links, etc.)
A tweet is a 280-character message that can include embedded items such as links, images, infographics, videos, and GIFs. Embedded items may improve the possibility of engagements, which include replies, retweets, and likes.
We will see whether these embedded items and engagement metrics fit in the ranking algorithm.
I believe the account of the tweeter could have an effect on the ranking. For instance, if the account is an influential one with a lot of followers and tweets, it could have an edge in ranking.
There are a few parameters of the account that could have a potential effect on the ranking of its tweets. Geography, description, follower/following counts, number of tweets, contents on other related tweets and moments, etc.
Everyone sees different search results on Google. It depends on a lot of factors, such as whether you are logged in, whether Google can identify your location, etc.
Twitter may be no different. Twitter could potentially show you different results depending on your location, which it may be able to identify.
Similarly, your top tweets could be influenced by the people you follow or those that follow you.
In the analysis, we will first try to eliminate some of these variables by testing their dependence on ranking. This way, we will be able to eliminate extraneous variables and concentrate only on those that actually contribute to ranking.
So, first I did a normal search of a hashtag, using my personal Twitter handle @leninnair. Then I compared the results against the results from a couple of other Twitter handles that I manage (one of which is ZoomOwl’s).
For this, I opened all three accounts in separate browsers: Chrome, Chrome Incognito, and Firefox.
The following Table 9.2 shows top 10 tweet handles that appeared for the hashtag “#IoT” when searched from three separate Twitter accounts.
As you can see, all three accounts showed the same results at the top 3 positions. Fourth position onwards, you can see slight changes in ranking across the accounts: While the tweet by @yogeshchauhan09 came fourth in the first account, it came fifth on the second and third accounts.
So, it looks like Twitter’s ranking doesn’t care whether you are logged in or not.
I did search of a number of other hashtags completely unrelated to my logged-in accounts. All the searches turned in almost the same results for all three accounts.
In addition, I confirmed this by searching without logging into any account. This can be done using the URL “https://twitter.com/hashtag/iot” (replace iot with the keyword you are searching.
So, it makes sense to conclude that whether you are logged in or not doesn’t influence Twitter’s ranking process. So, by extension, your tweet can reach people despite whether they follow you or not.
To gather information about whether Twitter ranks tweets based on geography, we will do a search from multiple locations. We are achieving this by using a proxy server, given by ZenMate VPN.
I did a search for “#lego” from the United States, Germany, and Romania. I did this without logging into Twitter (https://twitter.com/hashtag/lego). Here are the results.
As you can see from Table 9.3, there is no difference in the results. The same was the case for all other hashtags that I searched. Some returned insignificant position changes, though. Even using TOR, I wasn’t able to get any difference in results.
So, we can conclude safely that there is no dependence on the geography for the tweet ranking. This is good news because your tweets can potentially go global, if done right.
Now, we may think that Twitter ranks tweets by influential people higher up. There is nothing wrong in thinking that way. It’s completely logical.
Influential people are those who have a pretty high number of followers along with a high follower to following ratio. High follower to following ratio signifies that their following others is not a tactic to win more followers.
So, when an influential Twitter user tweets something, does Twitter place a weightage on their tweet and make it rank high? Let’s see.
The Table 9.1 above has the top 20 accounts ranking for the hashtag “#IoT”.
We use the following parameters to estimate the influence level of each account.
- Klout: Klout is probably the most widely used platform to analyze social media influence. It takes into account a large number of factors.
- RetweetRank: This ranks your Twitter account based on recent retweets, followers, etc. The lower, the better.
- Number of followers: Another intelligent way to rank each account is by the number of followers. The higher, the better.
- Followers/Following Ratio (Ratio 1): Some people using following as a tactic to gain more followers. This tactic doesn’t essentially increase the person’s Klout score or influence in general. So, in order to more accurately measure the influence, we take the ratio of followers/following. The higher, the better.
- Followers*Followers/Following (Ratio 2): In order to improve the Ratio 1 even better, we can take the square of followers, so that it accounts for the number of followers as well.
- Webfluential: This is another way by which we are estimating the influence of an account. Webfluential shows how much a tweet by an account is worth. Webfluential provides a range for the value of the tweet. So, for the ease of analysis, we will take the average of this.
The #IoT accounts, ranked on the basis of the above metrics, are shown in the following table.
You can see that most of the influence metrics do reasonably correlate with each other. A sample correlation chart is also shown below (Klout score vs. follower count).
Other influence metrics do correlate in almost the same way. So, we can reasonably assume that any of these influence metrics should adequately represent the influence of the accounts.
Unfortunately, it’s extremely difficult to identify if the influence metrics affect ranking, because we need to find identical, simultaneous tweets by accounts of different influence levels. This is practically impossible.
But we were able to almost rule out the effect of influence in ranking. Here is the Klout score vs. ranking position table that shows clearly that the Klout score didn’t really influence ranking. However, since these tweets are not identical and they have different tweet engagement metrics (likes, retweets, etc.,) we can’t really be sure.
We use the correlation as the statistical measure of whether the score has any influence on ranking. +1 signifies a strong correlation between the data, while -1 shows opposite correlation (ranking decreases for increase in Klout score). Any correlation value between -0.5 and +0.5 signifies very weak or no correlation.
As you can see the coefficient of correlation is -0.183, which means there is no correlation between the Klout score and the tweet positions.
Then we analyzed the tweet position vs. Ratio 2 (followers^2/following). This also tells us that there is no correlation (R value of 0.1982) as you can see below.
So, we will conclude that the tweeter’s influence may not have an effect on the ranking of the tweet.
Let’s look at the ranking of the hashtag “#IoT”. The below Table 9.5 shows what is present in the tweet itself. There are various elements that can be part of a tweet, such as images, infographics, videos, text, link, GIF, etc.
The following table shows the engagement metrics around each ranked tweet above. They include replies, retweets, likes, etc.
I am also mentioning the time elapsed since the tweet has gone live and a couple of ratios: retweet rate (retweets per hour), like rate (likes per hour). These ratios may give Twitter an idea of whether the tweet will attract more attention from users.
As you can see the retweet rate and like rates were very high for the first four tweets. It’s perfectly correlated to the position as well. Retweets actually make a tweet visible to more people. So, Twitter might as well just rank it.
So, it seems that Twitter ranks tweets based on whether they will attract a lot of future engagement.
However, the fourteenth tweet by @Cisco_Jasper has a very high engagement rate, but a low position. Why could it be like that? One reason I think is this tweet has an unnatural difference between retweet rate and like rate. The like rate is significantly higher than the retweet rate.
This may be an indicator that the tweet is unnaturally promoted. That in turn could be a reason why it is ranked lower.
These are only speculations I have.
So, we conclude thus:
If your tweet has what it takes to go viral, it has the potential to rank well.
You may have noticed that the embedded infographics were present on most of the top-ranked tweets for “#IoT”. I decided to test it out with other hashtags.
Several types of embedded items could be part of a tweet, such as the following.
- Infographics: These are rich images with text, icons, graphics, etc. They convey a complete message, in general, with a lot of visual appeal.
- Images: Images could be plain photographs or graphics designed using a software tool such as Adobe Illustrator. They improve the visual appeal of a tweet.
- GIFs: These are animated images that loop that are mostly used for memes.
- Videos: In general, Twitter allows you to share short videos of a few seconds that don’t loop.
- Links: Twitter lets you share your links and creates small snippets out of them, with a featured image and a short description.
We decided to see which of these elements really rank well on Twitter for hashtag searches. The following table shows the embedded items on the first 10 results for various hashtags (#Lego, #InfinityWar, #MachineLearning, #iPhone, #KoningsDag, #ArtificialIntelligence, #CryptoCurrency, #LeBron).
It’s clear, isn’t it? Twitter likes to rank tweets that have infographics, images, short video clips, etc., more often than plain text or links.
This may be because engagement is better for tweets that provide full information within itself rather than prompt users to visit a link.
In addition, another thing noticed from the tweets ranking for #IoT is that they have quite a bit of related hashtags. Some of those tweets have more than 20 related hashtags on them. This abundance may help the tweet rank on other hashtag searches, and that could be a reason why they have better engagement metrics.
I set my alarm for every hour and checked the ranking of tweets for the same hashtag (#AI). Here are the results in Table 9.8. I’m giving only the account handles from which the tweet is done.
The observation is clear that Twitter changes the position of the ranked tweets based on time. If it finds that a particular tweet’s level of engagement wanes over time, its rank is also lowered.
Also, like Google, Twitter has an average ranking position for each tweet and it changes during the course of the day. So, the same tweet appears in different positions at different times. Also, during the course of the day, positions are adjusted based on other factors.
Also, if a newer tweet is gaining in engagement, it is ranked among the top tweets, although not necessarily at the very top positions. This is done as a way of showing users what they may have missed.
So, based on the study above, here is what we conclude. Please don’t assume that everything we identified is true. The algorithm must use several unknown or hard-to-analyze factors.
- Useful infographics & videos: Yes, use an infographic, image, or a video on your tweet. Twitter has patent to recognize text from infographics, and it is probably using it to rank relevant infographics. Anyway, try to provide maximum data on your tweet, so it improves the engagement metrics.
- Get initial engagements: Engagement metrics are extremely important. Organizations are at an advantage in this case: They can request employees having Twitter handles to like and retweet, thereby increasing the initial engagement.
- Timing factor: Try to understand the time when your followers are most active. This can have a huge effect on the engagement you receive for your tweets.
- Put additional hashtags: Yes, try and identify additional hashtags you can include on your tweet. This is extremely good for ranking. This in turn positions your tweet to receive good engagement from the feed of those additional hashtags as well. But just don’t overdo it like some of the top ranking tweets in the list.
- Mention people: Even though this is not fully evidenced in the analysis above, there is potential merit in mentioning more people. It could take your tweet to their streams and could make them like or retweet you, thus boosting engagement further.
- Gun for engagement metrics: The retweet rate and like rate are very important. They tell Twitter whether a tweet has potential to be liked by more people. So, in the first few hours of tweeting, aim to maximize these ratios. You’ll maximize the potential of being ranked.
It took a while to gather all this information to generate a detailed analysis. I hope this will be helpful in your Twitter marketing efforts. As always, please subscribe to ZoomOwl and enjoy our future content.