Crypto Twitter marketing in 2026 runs on a ranking system anyone can read line by line, and most teams still plan their week as if it were a mystery. X published the code behind its For You feed on 20 January 2026 under an open-source licence, then added the actual ranking weights on 13 August, with 18 releases by early September. Those weights say a reply is worth ten likes and a repost is worth two.

The same year closed every cheap way of faking that conversation. X banned apps that pay people to post on 15 January, days after CryptoQuant's chief executive counted 7,754,367 bot-written crypto posts in one day. By the end of April X's head of product said crypto had become the most-muted topic on the platform, ahead of politics.

Most crypto teams post for volume, and the system they are posting into pays for conversation and charges for everything else.

How does the X algorithm rank crypto posts in 2026?

X ranks posts by predicting how likely each viewer is to take about twenty different actions on a post, multiplying each probability by a fixed weight and adding the results. In the published code a reply carries a weight of 5.0, a repost 1.0 and a like 0.5, while a report carries minus 234.

The system has three parts, described in X's public repository. Thunder collects recent posts from accounts a viewer follows, Phoenix finds and scores posts from accounts they don't, and Home Mixer applies the weights and builds the feed. The formula in the README is one line, "Final Score = Σ (weight × P(action))", and the same page notes that the weights scale predicted probabilities rather than raw engagement counts. The model is guessing what this particular viewer will do, so a thousand likes from strangers do not transfer to a reader who never engages with posts like yours.

The weights below are the defaults in the Home Mixer parameter file as we read it on 3 October 2026. X changes them, and the "not interested" penalty was listed at minus 43.2 in an independent walkthrough of the August release, so the ratios matter more than the decimals.

A worked example shows how much this changes the maths. Take two posts from the same project, and assume the model predicts that a given viewer has a 4% chance of liking the first, a 0.5% chance of reposting it and a 0.2% chance of replying. The score is 0.02 plus 0.005 plus 0.01, or 0.035. The second post asks a specific question about a design decision and the model predicts a 2% chance of a like, a 0.3% chance of a repost, a 1.5% chance of a reply and a 0.1% chance that the viewer copies the link to send to someone. That scores 0.01 plus 0.003 plus 0.075 plus 0.02, or 0.108, roughly three times the first post on half the likes. These probabilities are invented for illustration, but the weights are X's, and a 0.1% predicted chance of a mute would subtract 0.059 and wipe out the first post's entire score.

Three adjustments sit on top of the score, and all of them work against volume. The README says each post after an author's first in a feed "is multiplied by a decaying factor, down to a floor", that posts from accounts the viewer doesn't follow "are multiplied by a factor below 1", and that posts from authors with very few impressions are lifted toward a target position. Nikita Bier, X's head of product at the time, put the human version bluntly in January. He said the average user sees only 20 to 30 posts a day and that every post uses up part of an account's daily reach, which is why accounts replying "gm" hundreds of times had none left for their announcements. He deleted the posts after a backlash, but the code published nine days later described the same mechanism.

Does X still penalise links, and does Premium buy reach?

X says it no longer penalises links and the published code has no link flag, but posts with links still tend to earn the least engagement. Premium is different. Buffer's analysis of 18.8 million posts found Premium accounts received roughly ten times the median impressions of free accounts, so for a crypto project the subscription is close to mandatory.

On links, the official position has been consistent for a year. In October 2025 Bier explained that link posts got lower reach because the browser covered the post and people forgot to like or reply, and X began keeping the engagement bar on screen while a link is open. In July 2026 he told Mark Zuckerberg publicly that links no longer need to go in the replies, and Elon Musk added that X hadn't penalised them for over a year. The August code supports that, since the ranking model receives only six yes-or-no facts about a post and none of them is whether it contains a link.

The behaviour data is less kind. Buffer's study, which covered 71,000 accounts from August 2024 to August 2025, found text posts from Premium accounts earned about 0.9% median engagement, video just over 0.7%, images 0.4% to 0.5% and link posts 0.25% to 0.3%, and for free accounts the median link post earned nothing at all. That data ends before the in-app browser change, but the weights point the same way, because a link opened is worth only 0.2, less than half a like. Put the argument in the post, put the link at the end, and let the replies do the ranking.

On Premium, the same Buffer study found a median of about 600 impressions per post for Premium accounts and more than 1,550 for Premium+, against fewer than 100 for unpaid accounts. X sells the Basic tier at $3 a month and Premium+ at $40, and a company gold checkmark through Premium Business starts at $200 a month, with a $1,000 tier that adds affiliate badges for team accounts. We could not find a Premium multiplier in the published weights file, so the lift appears to come from elsewhere in the system. Either way it is the cheapest reach a team will buy all year, and the founder and the brand account should both have it.

What did the January 2026 InfoFi ban change for crypto Twitter marketing?

The InfoFi ban ended the practice of paying strangers per post to talk about a crypto project. On 15 January 2026 X revoked API access for apps that reward posting, Kaito closed its yaps leaderboards the same day and its token fell about 17%. X has since restricted automated replies and removed 42,000 AI reply accounts.

Bier's announcement was two sentences long. X would "no longer allow apps that reward users for posting on X (aka 'infofi')" because it had "led to a tremendous amount of AI slop & reply spam". Kaito's founder Yu Hu conceded that fully permissionless, incentive-driven distribution was "no longer viable", and Cookie, a rival, shut its own product as its token dropped 20%. We covered that model in our piece on Kaito yaps and InfoFi. The point here is that the ban was one move in a year-long sequence, and every move removed a way of manufacturing volume.

How much of crypto X is bots?

Nobody outside X has a clean number, but the scale is visible in the enforcement figures. Ki Young Ju's count of 7.75 million bot posts on 9 January was a 1,224% jump on the days before, and in April Bier said X was identifying and suspending 208 bots a minute. In one case reported in September, 34 of the 59 replies under a single Paul Graham post were hidden as possible spam, and Bier said about 95% of those were language-model bots. X's August and September code releases went further, adding an LLM-based reply-spam classifier on 4 September.

For an honest project this means crypto posts start with a trust deficit. Ju's complaint that X would rather ban crypto than improve its bot detection is fair, and it doesn't change what a team has to do. It also means anything resembling the banned behaviour is now a liability. X's stated line is that drafting a post with AI is acceptable while replying to people automatically without a human in the loop is not, and a team that hands its replies to a script is betting the account on a classifier whose false-positive rate X has never published.

Founder account or brand account: which should carry crypto X marketing?

The founder's account should carry the conversation and the brand account should carry the record. X's weights give a reply 20 points instead of 5 when the author and the replier follow each other, and people follow people back far more readily than they follow logos, so a founder's replies and the replies a founder receives are worth more to the ranking model.

This is a mechanical argument, separate from the trust argument in our piece on founder-led marketing. A founder with 3,000 followers who is followed back by 300 builders, investors and users in the same niche has an asset a 60,000-follower brand account usually lacks, because very few of those 60,000 are accounts the brand follows and talks to.

The brand account still has work to do. It is the place for releases, incident notices and documentation, and it should hold the gold checkmark. What it should stop doing is posting four times a day to fill a calendar, since the diversity decay discounts every post after the first in a viewer's feed. Bier's own advice on growing an account was to post one unique insight a day, under five sentences, for six months, which is a smaller and harder brief than most content calendars.

Which formats work on crypto Twitter in 2026?

The formats that work on crypto Twitter in 2026 are replies, single posts that invite a reply, long-form posts and Spaces that produce clips. Communities no longer exist, video carries no direct weight for quality views in the current parameter file, and threads work only when each post could stand alone.

Long-form is the format X has pushed hardest. It opened Articles to all Premium subscribers on 7 January 2026 and offered a $1 million prize for the best one, saying its payout formula "may weight formats like Articles more heavily than short-form content". By March Bier said the number of long-form posts had grown 18 times in three months. That statement is about creator revenue and not feed ranking, so we wouldn't promise a reach boost. The better case for long posts is that a crypto project usually has something technical to explain, and a post that someone copies the link to and sends to a colleague earns the heaviest positive weight in the table.

Video needs a caveat. In the parameter file as we read it, the weight for a video quality view is zero and opening a video counts for 0.07, so a clip has to earn its reach through replies and shares like any other post. Clips still matter for distribution across many accounts, which is the subject of our piece on crypto clipping campaigns.

Spaces are worth running when they are built to be cut up afterwards. One practitioner's analysis in HackerNoon argues that peak listener count is the wrong target and that replays and clips shared in the first day matter more, and it recommends 30-minute sessions over 90-minute ones. Communities are gone entirely. X shut the feature because fewer than 0.4% of users touched it while it produced 80% of spam reports, financial scams and malware, and pointed admins to group chats it said would grow to 1,000 members. A project that built its holder community there needs a chat it controls.

How do you measure crypto Twitter marketing?

Measure crypto Twitter marketing by the signals X weights and by what happens after the click. Track replies per post, the share of replies from accounts you follow back, shares, follows per thousand impressions and negative labels, then tie posts to tracked links, sign-ups or wallets. Impressions and follower counts tell you very little.

X now hands every active account a diagnostic. The August release shipped a tool called Under the Hood, at x.com/i/under_the_hood, which lets accounts with at least ten posts a month download the visibility labels applied to them, with per-post safety labels and percentile comparisons against accounts in the same follower band. Since 18 September those reports also show whether visibility was limited for legal compliance. A team should pull that report monthly, because a spam label explains a reach collapse faster than guessing about shadowbans.

• Replies per post and reply depth. Count replies, then count how many of them the author answered and how many of those were answered again.

• Mutual-reply share. The fraction of replies that come from accounts you follow back, since those carry four times the weight.

• Shares and copied links. A post that gets sent privately is doing what the weights reward most.

• Follows per 1,000 impressions. A follow is weighted at 4.0, and it is the only metric that compounds.

• Negative signals. Under the Hood labels, plus any post that draws visible complaints or blocks.

• Downstream actions. Tracked link opens, sign-ups, deposits or wallet connections within a week of the post.

The last line is the one that keeps the rest honest. X is a discovery channel, and the evidence from the InfoFi era, set out in our Kaito yaps analysis, is that engagement on its own did not predict how tokens performed. The feed rewards conversation, but a project is paid in users, and a week with fewer impressions and more qualified replies is a better week.

What we'd tell a founder about crypto Twitter marketing: a weekly operating plan

Cut the posting volume before adding anything. If the brand account posts more than once or twice a day, the diversity decay and the 20 to 30 posts a viewer sees are already working against you. Drop the "gm" posts and the engagement bait, and keep the posts a stranger would reply to.

Spend the saved time in other people's replies. A thoughtful reply under a larger account in your niche is the cheapest out-of-network reach available, and when the author answers, both sides earn the reply weight. It has to be a named person typing, because automated replies are what X spent 2026 removing, and the habit pairs with the outbound system in our piece on web3 lead generation.

Answer every real reply within the hour. A reply you answer becomes a conversation the model can predict next time, and a mutual follow turns each future reply from 5 points into 20. Follow back the builders and users who talk to you.

Treat mutes and reports as the real budget. One predicted report cancels 468 likes and one mute cancels about 118, so a giveaway that brings 2,000 likes and annoys a few hundred people can leave the account worse off. Label paid posts, because X now requires it and suspends repeat offenders.

A five-hour weekly crypto Twitter strategy

That plan is about five hours a week and five to seven original posts across both accounts, which is fewer than many crypto teams publish in a day. The time goes into replies because that is where the weights are.

The feed pays for conversation

Crypto spent two years treating X as a place where attention could be bought by the post, and X spent 2026 dismantling that one mechanism at a time. Paid leaderboards, reply bots, reply farming for payouts, Communities and undisclosed promotions are all gone or labelled, and what the platform published in their place is a table that says a reply between two people who follow each other is worth forty likes. Bier, who made most of those calls, moved to an advisory role on 6 August, and the weights were published a week later anyway. A team that posts less, replies more and answers its own replies is doing what the code rewards, and it is also doing the only version of crypto X marketing that an exhausted audience hasn't already muted.

Frequently asked questions

What is crypto Twitter marketing?

Crypto Twitter marketing is the use of X, formerly Twitter, to build awareness and users for a crypto project through posts, replies, long-form writing, Spaces and paid partnerships. In 2026 it is shaped by X's published ranking weights, which score a predicted reply at 5.0 against 0.5 for a like, and by the January 2026 ban on reward-for-posting apps.

How does the X algorithm work for crypto accounts?

The X algorithm scores every candidate post by predicting the viewer's chance of about twenty actions and multiplying each by a weight. In the published code a reply is weighted 5.0, a repost 1.0, a like 0.5 and a report minus 234. Crypto accounts face the same weights as everyone else, plus heavier spam filtering after 7.75 million bot-written crypto posts appeared in one day in January 2026.

Is InfoFi banned on X?

Yes. On 15 January 2026 X said it would no longer allow apps that reward users for posting and revoked their API access, citing AI slop and reply spam. Kaito closed its yaps leaderboards, which had about 157,000 participants, and its token fell roughly 17%. Paid partnerships with crypto projects are still allowed on X if they carry a Paid Partnership label.

Does X penalise posts with links in 2026?

X says it does not. Nikita Bier, then head of product, confirmed in July 2026 that links no longer need to go in replies, and the published ranking model has no link flag. Opening a link is weighted at only 0.2, though, and Buffer's study of 18.8 million posts found link posts had the lowest engagement of any format.

Is X Premium worth it for a crypto project?

For most crypto projects it is. Buffer's analysis of 71,000 accounts found Premium accounts received about 600 median impressions per post and Premium+ accounts more than 1,550, against fewer than 100 for free accounts. Individual tiers run from $3 to $40 a month, and a company gold checkmark through Premium Business starts at $200 a month.

Should a crypto founder post from a personal account or the company account?

A crypto founder should hold the conversation on a personal account and leave announcements and documentation to the company account. X's code adds a 15-point boost to replies between accounts that follow each other, which lifts a reply from 5 to 20, and individuals collect mutual follows far more easily than brand accounts do.

How do you grow a crypto project on X in 2026?

To grow a crypto project on X, post less and reply more. X's former head of product said the average user sees 20 to 30 posts a day, and the ranking code discounts each extra post from the same author in a feed. One strong post a day from the founder, plus 10 to 15 real replies, beats a calendar of five scheduled posts.