X has not published its own ranking weights, but three independent analyses of its now-open-sourced algorithm agree on the shape of it: a reply counts for roughly 27 times what a like counts for, a repost for about 20 times, and a post that does not earn real engagement in its first half hour effectively disappears. None of this is officially confirmed by X. All of it is consistent enough across sources to change how you should spend your ten minutes a day on the platform, and none of it requires an audience, an ad budget, or a viral post of your own to put to use — which is the part that matters if you are starting from zero followers.
Replies outrank posts, and the numbers say so
Third-party trackers that studied X's open-sourced ranking model report that a reply is weighted at roughly 27 times a like, and a repost at roughly 20 times. X has not confirmed either number in an official statement — the figures come from analysts reading the released code, not from a company blog post — but the direction is consistent across the sources that report it.
Source: Aggregated third-party reporting (autotweet.io, teract.ai, posteverywhere.ai), 2026.
If that ratio is even roughly right, a strategy built around getting five genuine replies a day does more for your visibility than a strategy built around getting fifty likes. It also means the same ten minutes you would spend writing your own post might do more for you spent replying to someone else's.
This runs against most growth advice aimed at indie hackers, which is almost entirely about what to post: threads, hooks, a content calendar. Posting is still worth doing. But posting assumes an audience already exists to see it, and a small account has no such audience yet. Replying does not have that requirement — it borrows the audience already gathered around someone else's post, and if the reported weighting is anywhere close to accurate, it does so at a higher rate per unit of effort than almost anything else on the platform.
The three replies that actually work
The three shapes below are not ours. They are the standard formulation in this corner of the internet, and the clearest statement of where it comes from is a guide published by one of the scheduling tools that sells into it.
Source: tweethunter.io/blog/twitter-reply-strategy — captured 2026-08-20
Its opening line makes from intuition the argument the weights above make from the data: "Most people treat Twitter as a broadcast channel. They post, wait, and move on. The fastest-growing accounts on the platform do something different. They reply." We are not claiming to have discovered any of this. What is worth adding is where the numbers behind it come from and how far they can actually be trusted, which is what the rest of this piece does.
Not every reply earns its weight. The reply strategies that keep showing up across the SERP for this topic collapse into three repeatable shapes:
- The value-add replyYou answer the actual question, or add a fact the post did not have. This is the one that gets you found by people who were not following you a minute ago.
- The story replyYou share the moment you had the same problem, in one or two sentences, before saying anything about what you built. It reads as a person, not a pitch.
- The honest disagreementYou push back on something in the post, specifically and without contempt. This is the one to use sparingly — it works when you actually disagree, and reads as bait the moment you fake it.
All three share one property: they say something that could not be pasted, unchanged, into a different thread. A reply that would work on any post in the timeline is the reply that gets skipped, by the algorithm and by the person reading it.
Consider the difference on a concrete post. Someone writes: "Spent all morning exporting data between two tools that should just talk to each other. Why is this still a manual process in 2026." A generic reply — "totally agree, this space needs more automation!" — could sit under a thousand different posts unchanged. A value-add reply names the actual tools and a workaround. A story reply says "I lost an entire Saturday to this exact export before I built something for it." An honest disagreement might push back: "Most of the tools I've tried do talk to each other now — the two you're using probably don't support the standard everyone else settled on." Each of the three says something the generic reply cannot, and each invites the original poster to answer back, which is the entire point of replying at all.
Reply fast, not clever
Third-party trackers report that a reply sent within 15 minutes of the original post earns roughly 3 to 5 times more visibility than the same reply sent two hours later. This has not been confirmed by X directly, but it fits the reported mechanics: the ranking model is described as scoring a post's fate largely in its first 30 minutes, based on whether it clears an early engagement threshold. A reply that arrives inside that window is riding the post while it is still being decided; a reply that arrives after is answering a conversation that has already been scored and is fading.
The same third-party analysis describes a post's visibility as halving roughly every six hours after it goes up. We plot that decay curve, and the caveats attached to it, in our piece on posting time rather than repeating it here. The shape is the useful part: whatever the true half-life is, a post loses most of its reach early, and a reply written for a post that is already six or twelve hours old is competing for a much smaller remaining audience than one written in the first few minutes.
Third-party analysis reports that including a link in the body of a post cuts its reach by roughly half. The practical response, repeated across the reply-strategy articles we reviewed for this piece, is to post the thought without the link, then add the link in your own reply underneath — or leave it out of the reply entirely and let an interested reader ask for it. Either way, the first message a stranger sees should not read as a pitch with a URL attached.
This matters even more when you are replying to someone else's post rather than writing your own. A reply that opens with a link reads as an ad the instant it appears, regardless of how the ranking system happens to score it — the person who wrote the original post did not ask for a product recommendation, and most readers can tell the difference between an answer and a pitch within the first half sentence. Save the link for a follow-up, or for the moment someone in the thread actually asks what you built.
What the open-sourced algorithm confirmed, and what it did not
On January 20, 2026, X published the code for its Grok-based recommendation model on GitHub, under xai-org/x-algorithm, replacing a set of hand-written ranking rules with a model trained directly on user behavior — a confirmed, dated event, not a rumor. A further large update followed in May 2026, and a "mutual-follower" change arrived in July 2026 that is reported to favor visibility between accounts that already follow each other.
What the code release did not do is hand anyone an official weighting table. The specific multipliers — 27x, 20x, the 15-minute window, the six-hour half-life — are third-party readings of that code and of observed behavior around it, aggregated across several analyst blogs that largely agree with each other but were not published by X. Treat the release as confirmation that the underlying model exists and is now inspectable, and treat the specific numbers pulled from it as informed estimates rather than settled facts.
This is worth sitting with for a moment, because it cuts against how most reply-strategy content presents these figures. An article that states "a reply is worth 27 likes" as flatly as it states "the sky is blue" is doing something the source material itself does not do — the analysts who produced these numbers describe them as readings of released code and observed outcomes, not as an official specification. Nothing in that makes the numbers useless. It does mean the honest way to hold them is as a strong, repeated signal about direction rather than a guaranteed constant that will hold in every subsequent version of the model.
Reply deboosting: the other half of the story
There is a second, less flattering mechanic that the visibility numbers above do not capture: several third-party sources describe reply visibility being suppressed for accounts outside the original poster's follow graph, sometimes called reply deboosting or, more informally, shadowbanning. Under this description, a reply from someone the poster does not follow can be folded behind a "Show more replies" toggle even when it would otherwise rank well on engagement alone. X has not confirmed the specific mechanics of this in an official policy document, so treat the reports as informed, not verified.
What makes this relevant to a reply strategy is the July 2026 "mutual-follower" update reported alongside the broader algorithm changes, which is described as further favoring visibility between accounts that already follow each other. If accurate, it means a reply strategy aimed entirely at strangers is swimming against a current that rewards existing relationships more than it used to. The practical adjustment is not to give up on replying to people who do not follow you back — a genuinely useful reply can still surface — but to also spend some of the same ten minutes replying inside the small circle of accounts you already follow and who follow you, where the reported suppression does not apply.
Finding tweets worth replying to
The same discipline that makes a Reddit reply land applies here: search for the plain words a frustrated person actually types, not the tidy phrase you would use to describe your own product. "Any recommendations for," "wish there was," "just switched from," "tired of doing this by hand" — these phrases catch people mid-frustration, which is precisely the moment a reply from a stranger reads as useful rather than intrusive. If you are building the list of who to reply to from scratch, our companion piece on reply outreach walks through the search patterns in full — the mechanics transfer directly, even though X's ranking behavior (fast decay, reply-weighted scoring) rewards a different rhythm than Reddit's moderation-driven risk does.
One practical difference on X: because search only reliably covers roughly the last seven days of posts under the platform's current API limits, a search built for X needs to run more often than one built for a slower-moving forum. A phrase that returns nothing today can return three matches tomorrow — the daily habit matters here even more than it does on Reddit, precisely because the fast-decaying nature of the ranking system means a three-day-old post has already lost almost all its remaining reach by the time you find it. Running that search every day by hand is the part most people quietly stop doing. WarmList is the paid version of it (disclosure: we build it): one run reads the public timeline and scores every post it reads, so the shortlist arrives already ranked.
Mistakes that bury a good reply
Most of what buries a reply comes down to one habit: writing it for the algorithm instead of for the person who will read it. A reply optimized purely for the reported weighting — short, fast, generic, sent to as many posts as possible — starts to look exactly like the automated replies that both platforms and readers are already trained to filter out. The reports above describe what the ranking system rewards; they do not describe what a stranger wants to read, and the two overlap only when the reply is actually specific.
- A reply written inside the first 15 minutes, about something specific in the post
- A link added afterward, in your own reply, not the original post
- A short disagreement or addition that only fits this exact thread
- The same sentence, reused across many unrelated posts
- A link sitting in the first line of the post itself
- A reply arriving hours after the conversation has already been decided
A ten-minute daily routine
Set aside ten minutes, once a day, at a time you can actually keep. Scan for two or three fresh posts — under an hour old, ideally under fifteen minutes — where you have something specific to say. Write the value-add reply, the story reply, or the honest disagreement, whichever actually fits. Skip the ones where your only reply would be generic. Do this daily rather than in one long weekly batch: the 15-minute window that third-party trackers describe cannot be caught in a batch written after the fact.
Track one number from this routine, and only one: how many of your replies get a response from the original poster. Not likes on the reply, not your own follower count that week — whether the person you replied to wrote back. A reply that earns a response has done the actual job, regardless of what the ranking algorithm did with it behind the scenes. A week where three of your ten replies get an answer is a better week than a week where all ten get likes and none get an answer, because a conversation is the only outcome that can turn into anything past the timeline.
It also helps to keep the routine boring on purpose. The moment replying starts to feel like a performance — checking impression counts, refreshing to see if a reply "did numbers" — is usually the moment the replies themselves start reading like a performance too, generic enough to fit anywhere so they can be sent everywhere. Ten minutes, two or three posts, then close the tab.
What tools built for this actually automate
A cluster of paid tools exists specifically to help with parts of this routine. Tweet Hunter builds a CRM and an AI-driven "Lead Finder" on top of scheduling and a viral-tweet library. Hypefury automates a keyword-triggered "Engagement Builder" feed and can auto-DM people who reply to your posts with a trigger word. Volumn.ai — one of the few tools in this category that names "indie hackers" as a target audience directly — sells semantic search across replies and retweets to surface prospects. All three are solving the same first problem this article covers: finding the right post to reply to, faster than scrolling for it by hand.
None of them, as far as their own marketing describes, automate the second half — writing and sending the reply itself as a real answer to what the post said. That part stays manual by design in most of these products, and for good reason: a templated or fully automated reply is exactly the kind of generic content the ranking system's own emphasis on genuine engagement is reported to work against, and it is the kind of reply a human reader recognizes as a pitch on sight. Whatever tool you use for finding posts, the fifteen-minute window and the specific, unpasteable reply still have to come from you. (We went through what is left of this market, and who owns whom now, in our X monitoring comparison; the wider set is in our comparisons.)
What none of this tells you
Every number in this piece is a third-party estimate read out of a model X has not fully explained in its own words. That is worth repeating plainly, because most articles on this topic state the multipliers as settled fact. They are consistent enough across independent sources to act on, and specific enough to be useful — but they are not the same thing as X publishing "a reply is worth 27 likes" in an engineering blog post. Build your reply habit around the shape of the evidence — reply fast, reply specifically, keep the link out of the first line — and hold the exact multipliers loosely.
What none of the third-party data can tell you is which posts, in your own niche, are worth replying to at all. That part has no shortcut: it takes reading the timeline, or a search for it, with the same attention you would want someone to give your own post. The algorithm can only reward a reply that already exists. Writing one that a stranger wants to answer is still, after all of this, a human problem — not one a released model, however carefully studied, can solve for you. More of what we have measured on this is on our blog.
The search, if you would rather not run it yourself
The section above says the hard part is deciding which posts are worth replying to at all, and that there is no shortcut except reading the timeline or searching it. That search is the entire product we sell: paste a product URL, and one WarmList run reads the public timeline, scores 500+ posts, and hands back each candidate with the post, its score, and the reason it scored that way. $9 a run, once, no subscription. It does not write the reply, and after everything above about speed, specificity, and keeping the link out of the first line, you would not want it to. It only decides which twenty posts are worth the ten minutes.
