What if we could finally know why certain posts explode on X (formerly Twitter) while others disappear almost immediately from the “For you” feed? Elon Musk’s platform has just lifted part of the veil on how its recommendation algorithm works by publishing the source code of Phoenix, the system in charge of determining which content is likely to interest each user. An initiative presented as a transparency exercise, but one that above all lets us understand very concretely what can make a post take off — or bury it — on the social network.
Like, comment and RT please
Contrary to what some might imagine, this isn’t a leak or a hack of the platform: Twitter (sorry, we still struggle to call it X even today) has voluntarily made the workings of its recommendation system public, allowing us in particular to observe how Phoenix assigns a different value to each interaction. And the key takeaway is that not all engagement is remotely equal. In fact, we learn that the system first tries to predict a user’s reaction to a post coming from an account they don’t follow. To do so, Phoenix analyses their recent browsing history and tries to anticipate the probability that they will like, share, comment on or report the content. These various actions are then weighted in order to calculate a recommendation score.
And this is where things get interesting, because according to the values found in the code, a user’s interaction with a tweet doesn’t carry the same weight for the algorithm. A simple like is worth only 0.5 points, while a repost is worth around 1 point. A reply, a quote or a share by direct message are credited with 5 points, while a new follow of the account represents 4 points. But the interaction that stands out by a long way is sharing a post by copying its URL, since that can represent around 20 points, or nearly 40 times the weight of a like. In other words, X seems to place particular value on posts that make users want to leave the platform in order to share them elsewhere. Copying a post’s link and sending it to someone, for instance, is therefore an extremely powerful signal for Phoenix. Conversely, simply piling up likes or reposts isn’t necessarily enough to push a post into the “For you” feed.
So what about ragebait?
But the most interesting part may lie on the side of negative interactions. Phoenix assigns them markedly larger coefficients than those given to positive reactions. A report can weigh around -234 points, while a hidden post represents around -59 points. The “not interested” choice sits around -43 points and a block around -31 points. In short, a report weighs several hundred times more than a simple like. The system is therefore not only trying to identify what drives engagement, but also what risks strongly displeasing the user. It also puts a very widespread idea about X into perspective: the notion that ragebait is a particularly effective strategy for going viral. In reality, the workings revealed by Phoenix show that this kind of strategy can quickly backfire. A post can generate plenty of reactions, but if those come with a significant number of reports, blocks or hide requests, its recommendation score can fall sharply.
Phoenix obviously works in a more complex way than this simple addition of points. The system combines the probabilities of each interaction with their respective coefficients before determining which posts will be shown to each user and in what order. This is therefore not a universal formula that guarantees a tweet will go viral. But this transparency nonetheless gives creators and brands some particularly valuable pointers. It shows in particular that the type of interaction a post generates sometimes counts for more than the raw volume of engagement. A post that gets 10,000 likes is therefore not necessarily better placed than another that draws far fewer reactions but does more to encourage users to share it directly or to follow the account.
Otherwise, Twitter / X has also introduced a new experimental feature called “Under the Hood”, intended to let certain users better understand any restrictions that might be affecting the visibility of their content. For now, this feature is offered only to a selected group of accounts as part of a test. X could nonetheless widen the rollout depending on the feedback it gets. Elon Musk’s platform says it wants to give users more visibility into the mechanisms that can influence their reach. An approach that fits directly with Elon Musk’s stated desire to make the algorithm more transparent and to let the public better understand the decisions the system makes.
It isn’t a miracle recipe either
Publishing Phoenix’s code obviously doesn’t mean X has just handed over a miracle recipe for going viral. But for content creators, the lesson is fairly clear: rather than simply trying to pile up likes, it is far more worthwhile to produce posts that users will want to share, comment on, quote or pass on directly. Engagement is, as ever, the heart of any post’s virality, whatever it may be.
