How A Twitter Reputation Algorithm Needs To Work

Big news last week: Twitter Search to dive deeper, rank results. Twitter’s @Santosh Jayaram has indicated Twitter Search, which currently searches only Twitter post text, will begin crawling links and indexing the content of pages posted in tweets. To help with search result ranking, Twitter is creating a “reputation” ranking system that among other things divines the reputation of each “tweeter.” Read more on how this might work in this great piece from Loren Baker and Danny Sullivan’s How We Search With The Twitter “Help Engine”. Then peruse the comments to these and other articles on this subject. The underlying theme to the comment flow is concern over Twitter link drop spam, as well as questions about how such a “Twitter algorithm” might work.

Since I am in the unique position of having…

A) Never worked for a search engine (although I did install SWISH in 1993 without major injury)
B) Oceans of experience with links and how they are built, attracted, baited, and shared, and
C) A Twitter account @ericward, I feel compelled to slip back into LinkMoses mode for a moment, so please indulge me. If Twitter’s reputation algorithm is to be taken seriously, it will have to consider and ignore many signals, many of which are quite different than you might expect.

What Matters, and What Doesn’t

Doesn’t matter: Follower count. First and most obviously, the number of followers anyone has is meaningless. There are many fee based services already that will get you thousands of followers automatically. So please Twitter, do not give any weight to a URL tweeted by a person with 806,000 followers, just because they have 860,000 followers. @AshtonKutcher may like a web site and tweet a link to his 1.7 million followers, but just how useful is that site from a content quality standpoint? Might not be at all. Yes, that tweet will spark a frenzy of clicks by half a million people to whatever site is tweeted, but this is a separate issue. As a lone signal, follower count is meaningless.

Does matter: Co-follower rate. If everyone from Harvard class of 2005 is on Twitter, and if every one of them follow each other and have only a couple non-Harvard classmate followers, that’s essentially a quasi-private closed Twitter loop, and while the tweets from Mimsy and Tad may be fascinating, they are useless from a reputation perspective. Caveat: I see potential for value from high co-follower numbers within certain fields. For example, if there are 3,789 audiologists on Twitter, and 75% of them follow each other, the tweets from them represent a high trust signal. Go ahead, scoff. They matter.

Doesn’t matter: When you started Tweeting. Just because you were a savvy early adopter and have been tweeting since day one, your tweets don’t automatically have extra reputational value. They may be crap. Likewise, there are tens of thousands of professionals who aren’t on Twitter yet but will be a year from now. If Warren Buffett starts tweeting, I could care less if the Buff-ster’s a newbie, I’m following. And I’m not alone. Although https://twitter.com/W_Buffett has been registered since February 20th with an “opening soon” message and nothing more, “he” already has more than 4,200 followers. Go figure.

Does matter: Who your followers follow, and what they tweet about. Probably more important than who you follow is who your followers follow, and what they tweet. It’s circular logic similar to the PageRankish notion that what links to what links to what links to what is important, but it’s a useful signal. If you have 2,000 followers, and all of them tweeted viagra and casino links in the past week, guess what?

Doesn’t matter: frequency of Tweets/updates. Less is more with most forms of online communication. Twitter’s 140 character limit alone proves this. And with tweet freaks, the more often someone tweets the more likely it is those tweets will sail right down and off your update page, never seen nor clicked. Like they never happened. But they did happen, and Twitter’s algorithm will see/crawl those links, even if your followers didn’t. So be judicious as to what you tweet, and how often. And remember there are time-delayed auto tweeting tools to factor into this mess. Those aren’t automatically evil. A smart PR firm might time delay a tweet so as to coordinate it with other announcements. That doesn’t make it spam, but it doesn’t make it gold either.

This is just a start, but think about all the other data Twitter can study that might be indicators of link spam. For example, do you follow known spammers? What about percentage of your tweets that are original, as opposed to retweets? Or frequency of link drops to the same URLs. Number of Twitter IDs on a single IP address. URL in your profile. Link profile of that URL.

I commented on a blog recently that Twitter posts can definitely be signals for Google. But they are no different than any other online medium. The signal strength will be driven by trust the search engine places in each individual Twitterer. It will now matter for those who hope to have signal influence at Google via Twitter to be cautious with follows and tweets. What nobody has mentioned yet is that Twitter pages and tweets are all still just URLs, like any other URL. That means over time a backlink profile will emerge, and combined with a follow/follower profile and a few other signals, Twitter trust might just be easier to measure than anyone thinks.

Parting note: A shout out to BuzzStream. Any of you link builders who haven’t already checked out BuzzStream should head over and have a look. I’ve been impressed enough to become an advisor to them. It’s in beta and already made my jaw drop. For years in my head I have envisioned the perfect link building management app, and in BuzzStream’s first feature set I see evidence it can be exactly what I’ve imagined. Sign up fast, because I have a feeling the demand might be about to get crazy.