When you hit publish, your post doesn't go to your followers. It goes to a prediction machine, which decides — repeatedly, in stages — whether your followers ever see it. Understanding that pipeline won't let you game it, but it will stop you from misreading your results, and it explains why some posts die quietly for reasons that have nothing to do with the writing.
Stage one: classification before anyone sees anything
Before distribution, the post itself gets processed. Automated systems classify its language, topic, and format; check it against spam and policy filters; and note structural features — does it contain a link, a video, an image, an external domain the platform has opinions about? None of this involves your audience. It's the platform building a feature set: a machine-readable description of what this post is, which downstream models will combine with what they know about each candidate viewer.
This is where some posts lose before the race starts. Content that trips a filter — or merely resembles content that historically performs poorly for the platform's own goals — begins with a handicap no amount of audience enthusiasm fully repairs.
Stage two: the audition
The post then goes to an initial slice of potential viewers — typically a subset of your followers and people who've engaged with you recently. This is the audition. The system watches what that first group does: do they pause on it, expand it, react, comment, share, click your profile? Do they scroll past at full speed? Each behavior updates a set of predictions about how the next group of viewers would respond.
Two things follow from this that most creators get wrong. First, early engagement matters not because the algorithm rewards speed for its own sake, but because early responses are the only evidence available when distribution decisions are being made. Second, who auditions your post matters enormously. If your recent posts trained the system to show your content to the wrong crowd, your best post can audition in front of an audience that was never going to care. This is one of the quiet costs of posting off-topic: you're not just wasting one post, you're miseducating the model about who your next one is for.
Stage three: expansion, or not
If the audition goes well, the candidate pool widens — more followers, then second-degree connections, then whatever discovery surfaces the platform runs. At every widening, the same evaluation repeats with fresh predictions for the new audience. Distribution isn't a single verdict; it's a sequence of small bets, each conditioned on the last. A post can pass the first round and stall in the second because the signals that excited your core followers didn't generalize.
What the model is actually optimizing
It's worth being precise about the objective. Ranking systems don't optimize for quality, truth, or your growth. They optimize for predicted engagement and time spent, tempered by whatever integrity constraints the platform imposes. Your interests and the model's overlap — genuinely useful posts tend to earn genuine engagement — but they aren't identical, and the gap is where frustration lives. A nuanced post can underperform a provocative one, not because the system misjudged it, but because the system judged it correctly against a metric you don't share.
The practical takeaways
- Judge posts against their audition conditions. A post published into a stale account, or after a string of off-topic posts, wasn't tested fairly. Don't kill the idea; retest it later.
- Consistency is model training. Every post teaches the system who your content is for. Coherent topics over weeks produce better auditions than any single-post tactic.
- Optimize for the human behaviors the model measures. Expansions, dwell, comments worth typing — these come from writing that respects readers. The honest path and the effective path are, for once, mostly the same road.