the twitter algorithm & growth guide
i’ve spent the last week going over the twitter algorithm code, line by line, to understand exactly how it works.
joke
i uploaded the entire thing to chatgpt, asked it to translate it into human, then sanity-checked it. on top of that, i read through the thousands of articles that dropped last week to see what people got right, what they got wrong, and where everyone was confidently talking nonsense.
this article is the result of pulling all of that together into one clear breakdown for creators, social media managers, and marketers who actually want to understand how growth works on twitter now.
this update changes how posts are evaluated at a structural level, moving away from surface-level factors like formats, posting times, or whether threads are back, and instead reshaping the underlying logic of how content is surfaced, ranked, and quietly killed off.
if your old playbook feels less reliable, it’s not because you suddenly forgot how to write, it’s because you’re posting into a different system now.
this is a breakdown of how that system works in plain english, and what you need to change if you want to grow under it.
twitter is no longer ranking posts, it’s predicting behaviour
the most important shift in this update is simple to explain but hard for people to internalise.
twitter is no longer asking whether a post is good.
it’s asking what a specific user is likely to do if they see it.
under the hood, the system predicts the likelihood of different actions happening when a post appears in someone’s feed. likes, replies, reposts, clicks, dwell time, profile visits, follows, and just as importantly, negative actions like not interested, mute, block, or report.
all of those predictions are combined into a single score. posts with higher expected positive outcomes rise, while posts with higher expected negative outcomes are suppressed.
this matters because it kills one of the most persistent myths on twitter, which is that any engagement is good engagement.
a post that generates replies but trains people to mute you is not winning in this system. it’s slowly teaching the algorithm to show your content to fewer people.
how the for you feed actually works
the for you feed follows a fairly consistent pipeline.
first, twitter gathers context about the user: who they follow, what they’ve engaged with recently, and how they tend to behave when content appears in their feed.
then it pulls in posts from two places:
- people they already follow.
- people they don’t follow but are algorithmically similar to what they usually engage with.
those posts are filtered, scored, and ranked together in the same system. there is no protected lane for accounts you follow and no guaranteed visibility just because someone hit the follow button.
each post is scored individually based on predicted behaviour for that user, and only the highest scoring posts survive into the feed.
this is why smaller accounts can still break through, and why following someone does not guarantee you’ll see their posts on a regular basis.
discovery is similarity based, not popularity based
one of the most misunderstood parts of this update is how discovery actually works.
twitter is not primarily looking at what is trending globally and pushing that content everywhere. instead, it looks at how users behave and tries to surface posts that resemble patterns they already engage with.
that resemblance is not just about topic, but about interaction: how people reply, how long they stay on the post, whether they follow afterwards, and whether they disengage negatively.
this means niche consistency is no longer just a branding preference, it’s an advantage.
if you post about marketing one day, crypto the next, then mindset, then memes, the system struggles to confidently place you. uncertainty reduces distribution, not as a punishment, but as a safety mechanism.
accounts that grow steadily tend to be boring in the same way every day, and that’s not an accident.
why some posts never get a chance at all
not every post even makes it to the scoring stage.
before ranking happens, posts pass through a series of filters that remove content the system considers ineligible: duplicates, very old posts, content users have already seen, posts containing muted keywords, or posts from muted accounts are often filtered out before scoring even begins.
this matters because volume does not equal opportunity.
reposting the same idea multiple times does not give you more chances, it often gives you fewer.
posting constantly does not mean you are testing more. in many cases, it means you are exhausting your own eligibility.
the system is designed to reduce repetition, not reward it.
why overposting works against you
there is also an explicit diversity mechanism built into the feed:
even if someone likes your content, twitter will not show them only you. the system intentionally reduces how often the same author appears to keep feeds varied and prevent domination by a small number of accounts.
this explains a lot of things people are often misinterpreting:
why impressions dip suddenly, why strong posts do not last forever, and why posting more does not linearly increase reach.
it's absolutely not a shadowban, it's just how the algorithm works.
the uncomfortable implication is that fewer strong posts often outperform a higher volume of average ones.
negative engagement is a silent killer
one of the quiet but important changes in this system is how seriously negative signals are taken.
the algorithm explicitly predicts whether someone is likely to mute, block, or mark a post as not interesting, and those predictions carry negative weight in ranking.
this is why rage bait is a dangerous long-term strategy: it might get you a ton of replies, but it also trains the system to suppress you with the audiences you actually want.
confusing content is another issue people underestimate: when people don't understand what you are saying, they don't always scroll past. sometimes they disengage permanently.
formats don’t rank, behaviours do
there is no special boost for threads, videos, images, or hashtags baked into the latest release of the algorithm.
formats only matter indirectly, through the behaviour they trigger.
for example: a thread works because it increases dwell or replies, a video works because it holds attention, a single tweet can outperform both if it consistently leads to follows or reposts.
chasing formats without understanding the behaviour they produce is how people end up copying tactics that do not work for their audience.
the better question is not what format the algorithm likes, but what action you want this post to trigger and how people will react to it.
what you need to do differently now
first, design posts for one primary action rather than trying to optimise everything at once. decide whether a post is meant to get people to reply, earn reposts, generate follows, or hold attention.
second, write for a specific audience every time; not everyone interested in marketing, but one type of reader.
third, reduce "noisy engagement". avoid bait that attracts people who will never follow you or care about what you post next.
fourth, post less but be more impactful with every post you do make. quality over quantity will get you seen more.
fifth, optimise for the question people subconsciously ask after reading a post, which is whether they would follow you after this. follow prediction is one of the strongest positive signals in the new alogrithm.
what didn’t change
there are still no magic posting times, but post when your audience is online.
there are still no guaranteed "viral" formats.
there is still no "hack" to unlock more reach.
and shadowbans aren't hiding in the code.
the mental model that actually works now
the easiest way to break your growth on twitter right now is to think of it like a stage:
that model assumes you’re performing outward, trying to reach as many people as possible, hoping one post travels far enough to change your trajectory.
that’s not how the system you’re posting into works anymore.
twitter is closer to a recommendation engine that’s constantly asking one quiet question over and over again: "when this post is shown to someone, does it make sense to show them another one from the same account tomorrow?"
every post you publish is less about how far it travels and more about what it teaches the system about you:
does it attract people who stick around, follow, and engage again later, or does it attract people who reply once and never want to see you again.
so a more useful way to think about posting now is this: each post is training data; you're either reducing uncertainty about who your content is for or increasing it.
when you post consistently for the same type of person, on the same set of problems, with roughly the same level of depth, the system becomes more confident about where to place you. discovery gets easier, not because you cracked a format, but because the algorithm can predict outcomes more reliably.
when you jump topics, chase formats, or optimise purely for engagement without thinking about who that engagement comes from, you slow that process down. the system gets noisier signals and responds by being more cautious with distribution.
the practical conclusion is uncomfortable but useful.