What we are working from
Everything below is grounded in what xAI actually publishes. Since 2026 the For You feed has been ranked by Phoenix, a Grok-based two-tower transformer that handles retrieval and ranking in one model, replacing the Heavy Ranker, SimClusters and RealGraph stack from 2023. The repository states the team eliminated every hand-engineered feature and most heuristics.
One caveat we will repeat wherever it matters: the numeric engagement weights are not public. They sit in a params module that has never shipped, in 2026 or in 2023.
Anyone quoting you an exact multiplier is repeating a leak or an inference. The ordering is reliable; the ratios are not. Full breakdown here.
What virality actually is
Mechanically, a post going wide is a feedback loop. Phoenix scores each candidate post by predicting how likely a specific person is to react to it.
Strong early reaction is the best available evidence that the next person will react too, so the post gets shown to a wider test group, which produces more reaction, which extends it again.
That has an unintuitive consequence: the same post can go viral or die depending on nothing but its first few minutes. Content quality sets a ceiling. It does not determine whether you approach it.
It also means virality is a property of the post plus the audience plus the timing. Reposting identical content a week later routinely produces a different result, and that is not mysterious.
The early window
Phoenix scores relative to the account, so a post performing unusually well against its own baseline is the signal, which is why small accounts are not structurally excluded.
Across the launches we run, the ceiling is effectively set inside the first thirty to sixty minutes. X has never published a window and we would distrust anyone who claims a precise one; this is our own observation across campaign data, and it is consistent enough that we schedule every launch around it.
What makes a post worth reacting to
Phoenix runs nineteen prediction heads. Fifteen are positive, replies, reposts, quotes, bookmarks, profile visits, dwell time, video completion and others. Four are negative: not interested, block, mute, report.
Three things follow directly from that list:
- Write something disputable. Replies sit high in the ordering and a claim nobody can argue with generates none. Specific, checkable statements outperform agreeable ones.
- Earn the dwell. Dwell time and video completion are predicted directly. A dense post someone actually reads beats a one-liner they scroll past, even if the one-liner gets more likes.
- Do not trip the negative heads. Engagement bait produces replies and mutes together. Four of nineteen heads exist to model annoyance, and that is the expensive side of the trade.
Notice what is absent: hashtags, posting-time folklore, and follower thresholds. The hand-engineered features those tactics targeted were removed from the system.
What you cannot control
Being straight about this matters, because most advice on this topic is sold as a formula.
You cannot control whether a large account happens to see the post, what else is happening in the news cycle that hour, or whether the model has classified your audience the way you assume. Two posts of identical quality routinely differ by an order of magnitude, and no amount of optimisation closes that gap reliably.
What you can control is the floor: how many people who plausibly care see it early. That is arithmetic rather than luck, and it is the part we run for clients.
We guarantee nothing on the upside, the average launch through our network lands around 1M views and strong ones reach 2–5M over 24 to 48 hours, but the ceiling belongs to the model.
Questions we get asked
What actually makes a post go viral on X?
Early engagement velocity relative to the account's own baseline. X's Phoenix ranker predicts how likely a given person is to react to a post, and reaction already happening is the strongest early evidence available, so a strong start earns a wider test and a slow start is capped. Content quality sets the ceiling; the first stretch determines whether you approach it.
Do hashtags help you go viral on X?
There is no evidence they help under the current system. xAI's 2026 release states that hand-engineered features and most heuristics were removed from the ranker, and the model instead learns relevance from engagement sequences. Tactics aimed at those older heuristics no longer have anything to target.
How many followers do you need to go viral on X?
None in particular. Ranking is relative to the account's own baseline rather than to an absolute follower threshold, so unusual performance from a small account is a strong signal. One founder account with twenty followers reached 1.1 million views on a single amplified article.