What Signals From Twitter Does Google Care About?
Earlier this year, Google and Twitter struck a deal which once again gave Google access to Twitter's data stream. Columnist Miriam Hirschman explores how this might play out.
Ever since news of the Google-Twitter deal broke in early February 2015, there has essentially been radio silence. We know that the deal gives the search giant access to Twitter’s “firehose” of live tweet data — but, thus far, there has been no information released on what search results will look like once the deal is implemented “in the first half of this year“(which itself is already half over).
To fill the information void, there has been plenty of speculation. But perhaps a better question than, “What will search results look like?” is, “How might Google use all the data that will come through that firehose?”
Google already uses a highly sophisticated algorithm to determine which of millions of sites to serve up as results for any given search query. According to Google, there are over 200 “signals” — such as content freshness, region, and terms on site — that help determine each search’s unique results set.
With its new access to Twitter’s data feed, Google will not only gain direct access to the tweets themselves, but could also benefit from all sorts of metrics they can likely collect from the volume of tweets and users’ interactions with them. Presumably, Google will use these metrics to determine how (and what) tweets will show in search. But it seems possible that they could incorporate at least some of these signals into their regular search algorithm as well.
What Exactly Are These Twitter Signals?
The universe of Twitter signals breaks up fairly neatly into four categories, where the category is the best indicator of how the signal is likely to be used.
Those categories break down as follows:
|Trend Strength||How popular a trend is at any given time; how many people are talking about it||Hashtag usage, Keyword usage (frequencies)|
|Tweet Strength / Engagement||The popular value of any specific tweet, how many people have seen or interacted with it||Impressions, Favorites, Retweets, Link clicks, Video plays|
|User Influence||The strength of any particular Twitter user, particularly as it relates to their follower network||Followers, Follower/Following ratio, Lists included in, @mentions|
|Link/Page Strength||The only dimension to move beyond Twitter itself, measures traffic to external sites from links within Twitter||# Links to a specific page, Link Clicks|
It is important to keep in mind that part of the significance of the Google-Twitter deal lies in its bringing (back) the real-time web to search results. Beyond just the possibility of having tweets included in a results set, the Twitter firehose will provide Google with real-time and highly recent data for these very metrics — e.g., what is trending now, what links have garnered the most clicks in the last 24 hours, etc.
Thus, it may be more accurate to consider time as a fifth category in the table above, but one that is layered over the other four categories.
Hashtag (#) Usage; Keyword Usage
These two are effectively the same. For any given period of time, they are an excellent indicator of how popular or talked-about an issue is. This can signal to Google the existence of the breaking news event.
Additionally, the pattern and velocity of change in usage — in other words, how quickly a hashtag increases or decreases in usage — may indicate the level of importance of the issue or event. A topic that has slow-growing chatter for a few days or weeks followed by a spike (such as the Oscars or the Super Bowl) represents a different type of event from one that goes from no chatter to everyone talking about it in a matter of hours (a plane crash or earthquake).
As a result, beyond just affecting results and news rankings in the organic search results, it might also find a place on Google Trends that incorporates the timeliness of topics trending on Twitter.
Tweet Strength / Engagement
Favoriting a tweet is the virtual equivalent of calling out “Amen, yeah sister.” It indicates that the tweet resonated personally with the person who favorited it.
Nonetheless, the person chose to favorite it, but not necessarily to share it with their social network — indicating limited perceived broader value or relevance. As such, this metric would be useful to Google as part of an algorithm that assesses tweet relevance, but likely doesn’t carry as much weight as another engagement factor, retweets.
Retweeting enhances a tweet’s visibility by extending it to social circles outside of those of the initial tweeter. When someone chooses to retweet, they are saying “I believe this tweet is valuable and important enough to my social network to pass it on to them.” Therefore, this metric — more so than Favorites and probably most strongly of all the metrics — may be an indicator of the perception of relevance of the tweet’s content.
If this does turn out to be the case — that RT counts will be the most influential aspect of Twitter data — then RTs themselves will become more powerful in their impact. Directly, they extend the network of people who are likely to see the tweet; indirectly, by signaling to Google that this is a tweet worth greater exposure, they have the potential to raise the tweet’s profile even further if it leads to it ranking on a search page.
Link clicks, Video plays
These two metrics reflect a different type of engagement with a tweet compared to Favorite/RTing and aren’t evaluative in the way those engagements are. However, clicks on these links can provide additional data to Google as to the interest/popularity of a particular site when traffic to that site is coming from Twitter rather than from Google. These data are likely to be combined with what Google already knows about a site or link such that it can impact the rankings in search of the site linked in the tweet. (See also “Number of Links to a Specific Page,” below.)
Impressions (potential, actual)
These are passive measures and therefore likely to be less important than metrics which are indicators of involvement. Moreover, this is more important as an effect rather than as an indicator of inherent tweet popularity or importance. However, it may be important as a basis for comparison of the other metrics, where the starting point is assessing how strong those other measures are based on how many people saw the tweet in the first place.
As with all the metrics here, in a general sense, the more the better; but at some point, increasing impressions stops leading to increasing tweet engagement numbers in that the tweet is reaching people to whom it is not at all relevant.
Followers (including network size)
This is a direct measure of the immediate sphere of influence of a Twitter user. However, like all user influence metrics, it’s a passive measure, one of potential exposure, and doesn’t relate to the specific relevance of any tweet.
Follower/Following ratio, Lists included in, @ mentions
These, too, measure various aspects of the sphere of influence of a Twitter user. Because they don’t relate to a specific tweet, these cannot indicate whether a topic is trending or provide insight into the relevance and value of any content or link in particular.
However, they may be used as criteria in determining which of many tweets on a topic to show on a search page, on the assumption that a tweet crafted by a user with a large follower base or many @ mentions is likely to be of broader interest or relevance, given the large number of votes of confidence in/opt-ins to the tweeter.
Number of links to a specific page
The key to this metric is that it is measured in aggregate across the entire Twitter platform and considers all forms of link shorteners (Twitter-shortened, bit.ly, or any other format). Coupled with a time factor (how many links tweeted in the past 24 hours? past 7 days?), Google might use these links to affect rankings of the linked sites themselves in organic search results, especially on a short term basis as a topic or particular link is trending.
At a most basic level, this metric might be considered a Twitter-based version of Google’s famed PageRank.
Although this metric came up earlier in the Tweet Strength/Engagement category, here its sphere goes beyond that of a single tweet. As part of a link/page strength measure, it measures in aggregate the number of clicks across the entire Twitter platform to a destination page off Twitter. Similar to the previous metric, this one goes a step beyond in considering how many people actually clicked through the link — a measure of interest/relevance.
Where Do All These Signals Lead?
Ultimately, without more information from the Google or Twitter teams as to how their union will be implemented, we can’t say with certainty which of these metrics Google will use, how important they will be in impacting tweets in search, or how they will be integrated into algorithms already in place.
As of now, still prior to the opening of the Twitter firehose, Google indexes a relatively small percentage of Tweets — roughly 7%, according to an excellent study by Stone Temple Consulting. One correlation that emerged from their research is that the more followers a Twitter account had, the more likely its tweets were to be indexed. Likewise with the presence of images and hashtags, themselves indicators of stronger engagement.
Once the firehose does open and all tweets become indexed (or at least index-able), how will these factors influence search engine rankings (if at all)? That’s something we can only make a reasonable guess about. But what is clear is that it will be ever more important for both companies and individuals to invest in their social engagement.
We believe it is worth thinking not just about how the deal will impact the face of search but also its workings under the hood. Looking at what the metrics signify is a first step in helping those with any online presence decide which areas they are best equipped to focus on to better prepare them for whenever the firehose does open.
Opinions expressed in this article are those of the guest author and not necessarily Search Engine Land. Staff authors are listed here.