The platform’s newly exposed recommendation mechanics make one thing unusually clear: a Like is cheap. A share is a signal.
For marketers, one of the great mysteries of social media is figuring out the formula platforms use to decide what to show in users’ timelines. While it’s unlikely that any social media platform will offer complete algorithmic transparency, X (formerly Twitter) has revealed much more of those secrets in the last couple of weeks.
On August 13, as part of an ongoing initiative by Elon Musk to be more transparent about the X algorithm, X made public a large part of the source code for Phoenix, one of two algorithms that determine what appears in the “For You” feed in an X user’s account.
That revealed two key points. One, shadowbanning, at least on X, is a real thing. Users can now check, at least in the aggregate, but not on a post-by-post basis, whether their content has been shadowbanned and why.
The other main takeaway is that the X “For You” algorithm is prioritizing shares far more than other user actions. For example, in the Phoenix algorithm’s internal scoring, copying a post’s link and sharing it is 40X as valuable as a Like. What that means in plain English is that sharing content from a user makes it 40X more likely you’ll get more content from that person in your feed, compared to liking a post from them.
Here’s what marketers need to know about what’s been newly revealed about how the X Phoenix algorithm works.
The X “For You” Feed Actually Uses Two Algorithms
How the X “For You” feed works is based on the ranking of two Grok-powered algorithms. The X Thunder algorithm handles in-network content, meaning it shares original posts, replies, and original X accounts you already follow. The Phoenix algorithm handles out-of-network content, suggesting posts based on past history that X predicts the user is likely to interact with.
Since Thunder is based on who a user follows and Phoenix on predictions tied to the user’s history, the specific combination of posts that appears in a user’s feed is unique to them.
Checking For Shadowbanning On X
As part of the expansion of the open-source codebase, X added a feature so users can check if their posts or account were tagged with visibility-limiting labels, on a new ‘Under the Hood’ page.
The feature is currently only available to a randomly selected test group of users who have had their X accounts for at least a year. However, it works as follows: users who have posted at least 10 times in the last month can download a JSON file containing aggregated statistics that explicitly show whether any visibility-limiting labels have been applied to the account or to specific posts. Another option is to feed the file into an LLM like ChatGPT, along with the GitHub repository code, to understand how the recommendation engine sorts your content.
The 10-post minimum, along with sharing results in the aggregate, is intended to prevent users from receiving feedback on specific posts. However, you will learn how many posts total were tagged with visibility-limiting labels, and what they were.
Brands typically aren’t publishing edgy or controversial content on X, so this feature isn’t going to be used as frequently by them as, say, by a stand-up comedian or rage-baiting influencer. Still, crisis management moments or engagement anomalies do happen, and access to this information will give marketers more actionable context into what’s happening on social media and why.
Identifying The User Actions That Matter The Most On X
There are at least 18 user actions that the Phoenix ranking system uses in determining what content to place in a user’s “For You” feed. The positive user actions are Share via Link Copy, Reply, Share Via DM, Quote, Follow the author, Share, Repost, Like, Click into the Post, Open a link, Expand Photo, Open Video, and per second of Dwell time. When a user takes positive actions around an X account, they’re more likely to see additional content from that account or about similar topics in their “For You”.
The 5 negative actions being tracked are Not interested, Block, Mute, Report, and Didn’t Dwell. Users are less likely to see similar content in their feed when they take those actions.
What’s new is that X has just revealed the specific number of points assigned to a user’s action in the Phoenix ranking system.
In terms of positive actions, Share via copy link is worth 20 points, while Share via DM, Quote, and Reply are each worth 5 points. Meanwhile, Follow the author is worth 4 points, Share is worth 2 points, and Repost is worth 1 point. A Like is worth only 0.5 points, with every other positive action worth even less.
For marketers who want to optimize for reach on X, Shares via the copy link are by far the most valuable metric, while Quotes and Replies should also be encouraged. That encouragement can include interacting with commenters on a post thread. On the other hand, asking users to like your post on X, aside from potentially triggering spam filters, has almost no tactical value at this point.
In terms of negative actions, reporting a post has a weight of -234 points, muting the author has a weight of -58 points, being marked “not interested” has a weight of -43.2 points, and blocking the author has a weight of -31.2 points.
In general, given that reporting a post has the most negative impact on reach, it’s an indicator that rage-baiting isn’t an effective strategy on Twitter. Marketers aren’t going to engage in that behavior anyway, but if for whatever reason a brand post gets reported, that content should definitely be reviewed and framed differently in the future.
One last thing to keep in mind is that each Twitter user’s algorithm is a closed ecosystem around positive and negative actions. In Phoenix, a post being reported by one user has no bearing on whether another user can see it.
Final Thought:
For years, ListenFirst has been preaching that vanity metrics, such as Likes, don’t matter nearly as much to brands as many people think. By revealing much of the code behind the Phoenix ranking system, X laid that out in the clearest and most mathematical terms; that’s very much the case. Getting links copied and shared, driving replies, and growing the online conversation through positive interactions are being prioritized on X and are what marketers should focus on.
I hope going forward, more social media platforms will share this level of information about optimizing content for their feed algorithms. It’s a huge benefit for brands and for social media networks. When marketers have a clearer roadmap for succeeding on a social media network, there’s more incentive to devote more resources to sharing content on that platform.
