How to Get More Replies on X (The Only Metric That Still Matters)

Most people chase the wrong number on X. They obsess over follower count, refresh their like tally, and feel a small dopamine hit every time a post crosses 100 likes. Meanwhile their reach has flatlined for months and they can't work out why.

Here's the uncomfortable truth: likes are close to worthless in 2026. The open-source X algorithm, published on GitHub at xai-org/x-algorithm, makes this explicit. Replies aren't just "important", they're the single strongest lever you have over distribution, and the gap between replies and everything else has only widened this year.

This isn't another generic "engage with your audience" post. We're going to walk through the actual weighted scoring system, the reply-specific benchmarks nobody wants to publish because they're unflattering, the new reply-ranking overhaul X shipped in 2026, and exactly what to write to pull replies out of people who'd otherwise scroll past.

Why Replies Beat Every Other Metric on X in 2026

Twitter open-sourced part of its recommendation code back in 2023, and xAI went much further in January 2026 with a full Grok-powered replacement, followed by a major update in May. Between the two releases, one thing has stayed consistent: the relative value of a reply compared to a like.

The widely cited weights from the code show reposts worth roughly 20x the value of a like, replies worth roughly 13.5x the value of a like, bookmarks worth around 10x the value of a like, and likes sitting at the 1x baseline. That alone should change how you think about your own metrics. A post with 300 likes and 4 replies is, in the algorithm's eyes, weaker than a post with 40 likes and 15 replies.

But the real number that matters is what happens when you close the loop. The code contains a specific bonus for two-way conversation: a reply that generates a response from the author is cited at 150 times the value of a like. That's not a typo. Replying to the people who reply to you is worth more than ten times the value of a plain reply on its own.

Put those two facts together and the practical implication is stark. A post that gets 100 likes has a lower algorithmic score than a post that gets 4 replies, if you reply back to each of those 4 replies, triggering the author-replies-back signal four times. You don't need viral reach to win here. You need a handful of people who care enough to type something, and the discipline to reply back to every one of them.

What the X Algorithm's Code Actually Predicts About Your Replies

The May 2026 update to the repository (187 files changed, over 18,000 lines added) shipped a runnable inference pipeline, meaning anyone can now download a "mini Phoenix" model and see how the ranking transformer actually scores content. The README lays out exactly what the system predicts for every candidate post.

According to the repo, Phoenix predicts a set of engagement probabilities per post, including P(favorite), P(reply), P(repost), P(quote), P(click), P(profile_click), P(video_view), P(photo_expand), P(share), P(dwell), P(follow_author), and several negative actions like P(not_interested), P(block_author), P(mute_author), and P(report). Those probabilities get multiplied by fixed weights and summed into a single ranking score, described in the code as a weighted scorer that combines these into a final score.

Two things stand out here. First, reply probability is one of maybe 15 to 19 predicted actions, but it's weighted heavily enough to outweigh most of the rest combined once you factor in the author-reply-back bonus. Second, negative actions carry real weight too. Reports and mutes actively suppress future distribution, so a post that generates arguments and reports isn't automatically a winner just because the reply count is high.

If you want the full breakdown of every ranking stage, including retrieval, filtering and the diversity scorer, our complete guide to how the X algorithm works in 2026 covers that end to end. This piece is deliberately narrower: it's about the one signal that dwarfs the rest, and what you actually do about it.

X Just Rebuilt How Replies Get Ranked, Not Just Counted

Here's something most 2026 algorithm write-ups miss entirely. In mid-2026, X added a thumbs-down button on replies for Premium subscribers, and the reason behind it matters more than the feature itself. X added a thumbs-down button to post replies, available to X Premium subscribers, and when a user downvotes a reply, they choose from five reasons: not interested, incorrect or misleading, AI-generated, spam, or report post. These signals feed directly into the reply ranking algorithm, and quality, on-topic replies now surface higher in threads while AI-generated spam and misleading content get pushed down.

X's head of product, Nikita Bier, was blunt about why this was needed. He described the existing reply ranking as having "no logic, no signal, just garbage". That's a rare moment of a platform admitting its own system was broken. It also tells you something practical: getting a high raw reply count isn't enough anymore. The quality and relevance of the conversation you generate is now feeding back into how visible your replies (and by extension your account) become in other people's threads.

The Reply Benchmarks Nobody Wants to Show You

Most "engagement benchmark" articles quote a single blended number and move on. That's not useful if you're specifically trying to improve replies, because reply rates are a much smaller slice of total engagement than likes, and comparing yourself to the wrong baseline will either make you complacent or send you into a panic for no reason.

The honest numbers are low across the board. Industry data puts the average reply rate at 0.02% to 0.05% for promoted tweets, against a blended average engagement rate of 0.5% to 1%. Organic reply rates for regular creators tend to sit a bit higher than promoted content, but not by much. If your last ten posts averaged one or two replies each on a few thousand impressions, you're not failing. You're roughly in line with the platform.

That said, the platform overall is trending in the wrong direction. One 2026 study across 350,000 accounts found Twitter/X has the lowest engagement rate of any major platform at 1.11%, with a year-over-year decline of 9%, the steepest of any platform tracked. Even the most engaged accounts are seeing smaller pools to work with. That makes deliberately engineering for replies more valuable, not less, because it's one of the few signals still climbing in relative importance even as raw numbers fall.

Context by account size helps too. Smaller accounts have a structural advantage here: accounts above 200,000 followers usually settle between 0.5% and 1.5% engagement, since a fixed share of any large audience stays passive, while smaller accounts under 5,000 followers regularly see 3% to 6% or higher. If you're a smaller account and you're not converting some of that headroom into replies specifically, you're leaving the algorithm's favourite signal on the table.

For a full picture of how to read your own numbers rather than platform averages, our guide on using X analytics to grow faster walks through the exact metrics worth tracking week to week. You can also run your own numbers instantly with the Engagement Rate Calculator or check any public profile's average with the Engagement Checker.

How to Write Posts That Actually Get Replies

This is where most advice gets vague. "Ask questions" isn't wrong, but it's incomplete, and half the time it produces posts that read like a customer satisfaction survey. Here's what actually moves the needle, backed by what's been tested at scale.

Ask for an opinion, not a fact

Closed questions with a yes or no answer get skipped because there's no social cost to ignoring them. Open questions that ask for someone's take, story, or disagreement get replies because people want to be heard, not just correct. Data on this is fairly consistent: tweets that explicitly invite replies generate 2 to 4x more reply engagement than the same content without a question, and making the question specific and easy to answer immediately matters more than the topic itself.

The difference between "thoughts on remote work?" and "what's the one remote work habit you'd never give up?" is the difference between a post that gets three replies and one that gets thirty. Specificity gives people something concrete to respond to instead of a blank page.

State a position and don't hedge it

The strongest single-tweet format right now isn't a question at all. It's a confident, slightly one-sided opinion that people either agree with loudly or want to argue against. The highest-performing single-tweet format on X right now is a confident one-sided opinion, structured simply: state something many people won't fully agree with, then defend it without hedging, no "on the other hand," no qualifiers. This format dominates because it generates the reply-chain behaviour the algorithm weights most.

This is uncomfortable advice for anyone who was taught to write balanced, hedged takes. But hedged takes don't provoke a response. Nobody replies to "there are pros and cons to both sides." They reply to a claim they think is wrong.

Leave a gap, not a full answer

Incomplete structures invite completion. A numbered list that stops at "three rules" without listing all three, or a claim that implies there's more detail worth asking about, pulls people into the comments to ask "what's the third one?" This works because it converts passive readers into active participants without asking them to do anything effortful like compose an original thought.

Use polls when the question is genuinely open

Polls remove almost all friction from responding, which is why they consistently outperform plain text on raw interaction volume. Tweets with polls get 2 to 3x the engagement of standard text tweets, because tapping a poll option requires less cognitive effort than composing a reply. The catch is that poll votes and reply counts aren't the same signal, so don't rely on polls alone if replies specifically are your goal. Use a poll to warm up an audience, then follow with a reply-only question in the next post to convert that attention into actual conversation.

Ask deeper questions, not shallow ones

There's a subtler version of this worth calling out separately. Yes/no prompts get short, disposable replies that don't do much for dwell time or conversation depth. Deeper prompts do more work per reply. As one breakdown of the algorithm's dwell-time weighting puts it: instead of asking for a "yes/no" answer, ask for an opinion, which encourages longer replies and keeps people on your post longer. Longer replies mean more dwell time in the thread, which is its own separate ranking signal on top of the reply count itself.

The Reply-Back Trick Most Creators Ignore

Everyone knows they should reply to comments. Almost nobody treats it as the highest-leverage five minutes of their day. Given the 150x multiplier attached to author-reply-back chains, this should be the first thing you do after posting, not an afterthought you get to if you have spare time.

Here's the practical routine. Post, then set a timer for the next two hours and check back every 15 to 20 minutes. Reply to every single comment, even the short ones. A one-line acknowledgment still triggers the signal. If someone asks a follow-up question, answer it properly rather than with a thumbs-up emoji, because a reply that gets a reply back from them again extends the chain even further.

One detail people miss: this works even better as a deliberate strategy in your own replies. Reply to every reply you get, especially in the first 2 hours, because each volley extends the engagement window. Treat your own comment section like a live conversation you're hosting, not a mailbox you check once a day.

This is also where a tool genuinely earns its keep rather than just automating noise. Xpert's AI Reply Generator is built for exactly this moment: paste in a comment you've received, and it drafts a few reply options in your own voice so you're not staring at a blank box trying to think of something worth saying to twelve different people in a row.

Reply Strategy for Growing Small Accounts

If you're under a few thousand followers, posting original content and waiting for replies is the slow path. Your posts simply don't reach enough people to generate meaningful conversation yet. The faster route is replying to bigger accounts in your niche, and the data behind this is more compelling than most people expect.

One documented growth case tracked a jump from 500 to 50,000 followers over eight months, built almost entirely around a reply-first routine. The creator behind it described spending 70% of their time on strategic replies to high-follower accounts, and only 30% creating original content, on the logic that X rewards conversation starters, not broadcasters, and the algorithm gives replies roughly 15x more weight than likes, so when you reply thoughtfully to an account with 100,000 followers, thousands of people see your reply, and if it adds value, they check your profile.

The results in that specific case were substantial: a 10.3% engagement rate with replies driving 70% of total growth. Your mileage will vary, but the mechanism is sound and repeatable: you're borrowing an audience you haven't earned yet by adding something useful to a conversation they're already in.

Timing matters just as much here as it does for your own posts. Replying within the first 15 minutes matters because that's when a post has the most eyes on it and the fewest competing replies, and one growth breakdown found that replies posted within 15 minutes earned roughly 3 to 5 times more visibility than replies left after two hours. Getting in early beats writing the perfect reply two hours late almost every time.

A practical daily volume worth aiming for: roughly 10 to 20 quality replies per day when you're growing, enough to build momentum, few enough that each one stays thoughtful. Don't fire off "great point!" thirty times a day just to hit a number. A low-effort reply that gets flagged as spam under the new thumbs-down system will do more harm than good.

If timing is your bottleneck, cross-reference your own posting windows against your audience's active hours using the Best Time to Post tool, and see our dedicated breakdown in the best times to post on X in 2026 for the specifics by timezone and niche.

What Actively Kills Your Reply Rate

Getting more replies isn't only about doing the right things. There are several habits that quietly cap your reply count no matter how good your hooks are.

Restricting who can reply. If you set your posts to "followers only" or "mentioned only" to avoid harassment, you're also mechanically cutting off the algorithm's favourite signal. Tweet authors can restrict who replies, and while this reduces harassment, it mechanically lowers engagement rates. Only restrict replies when you genuinely need to, not as a default setting.

Obvious engagement bait. The algorithm and your audience have both got good at spotting manipulation. The algorithm and Twitter users can spot "LIKE if you agree!" tactics from a mile away, so the goal is to invite your audience to participate naturally by asking thoughtful questions that make people feel their reply matters. "Reply YES if you're still here" gets ignored or, worse, reported as spam.

Leading with a link. This hurts reach generally, but it hurts reply rate specifically because people who click away don't come back to comment. The algorithm penalizes external links, and Elon Musk confirmed that posts containing links that take users off-platform often see a reach reduction of 50-90%. Put the link in a reply to your own post, not in the body.

Ignoring the replies you already get. This is the single biggest wasted opportunity on the platform right now, because it's throwing away a signal worth up to 150x a like for free. Silence after someone comments doesn't just miss the bonus, it also signals disengagement, which drags down how likely that person is to bother replying to your next post.

Vague, AI-flavoured replies of your own. With the thumbs-down system live, low-effort or clearly templated replies get actively downranked now. If you reply to posts, make it specific and genuine, because vague or AI-generated replies now carry real ranking risk, while authentic replies are also more likely to surface higher as spam gets downranked around them. If you're using AI to help you reply faster, use it to beat writer's block, not to sound generic. A good reply generator drafts something specific enough that you'd still edit it, not something you can copy-paste unread.

How to Track Whether Any of This Is Actually Working

Vanity metrics lie. Impression counts inflate as your posts get shown to non-followers who scroll past without reading. The only honest way to know if your reply strategy is working is to track reply rate as its own line item, separate from your blended engagement rate, over a rolling window rather than post by post.

Pull your last 10 to 15 original posts (skip retweets and replies you posted, since those skew the picture) and note replies, likes and impressions for each. Calculate replies as a share of impressions specifically, not just total engagement. If that number is trending up over a month even while impressions stay flat, your content is doing its job regardless of what the raw follower count says.

Watch your reply rate the way you'd watch a fitness metric, weekly, not daily. Individual posts will spike and dip based on timing and luck. The trend over four to six weeks tells you whether your hooks, questions and reply-back discipline are actually compounding.

If you want a starting point for what "good" looks like for your specific tier before you start tracking, our guide on building a strategy around the real 2026 algorithm weights covers the wider scoring system this article zooms in on, useful if replies are one part of a broader plan rather than the whole thing.

FAQ

How many replies is considered good on X in 2026?

There's no single number that applies across account sizes, but as a rough guide, five to fifteen replies on a post with a few thousand impressions is solid for a smaller account, while larger accounts should expect a smaller percentage of a much bigger audience to comment. Rather than chasing an absolute count, track replies as a percentage of impressions over several weeks and look for that percentage climbing, since the average reply rate sits around 0.02% to 0.05% for promoted content and organic content from an engaged niche account should comfortably beat that.

Do replies really matter more than likes on X's algorithm?

Yes, and the gap is large. The open-source code puts replies at roughly 13.5x the value of a like, with reposts around 20x and bookmarks around 10x, against a baseline of 1x for a plain like. The bigger multiplier comes when you reply back to a comment on your own post, which the code values at roughly 150 times the value of a single like.

Why did my reply count drop even though my content hasn't changed?

Two 2026-specific changes are worth checking. First, X's reply-ranking overhaul, tied to the new thumbs-down feature, now actively demotes vague or off-topic replies in threads, which can quietly suppress lower-quality conversation even if the raw count doesn't show it immediately. Second, Twitter/X engagement fell 9% year over year in 2026 platform-wide, so a decline that feels personal might partly be a platform-wide trend affecting everyone in your niche.

Should I reply to every comment I get, even short or low-effort ones?

Generally, yes. Even a brief acknowledgment triggers the author-reply-back signal, and replying to every reply you get, especially in the first two hours, extends the engagement window. The exception is obvious spam or bot replies, where engaging does nothing for the signal and just wastes your time.

Do polls count as replies for algorithm purposes?

No, poll votes and text replies are tracked as separate actions in the scoring system, though polls are still useful for building momentum because tweets with polls get 2 to 3x the engagement of standard text tweets since tapping an option requires less effort than composing a reply. If replies specifically are your goal, follow a poll with a direct, open-ended question in a later post rather than relying on the poll alone.

Is it better to post original content or reply to bigger accounts if I'm just starting out?

For accounts under a few thousand followers, replying to larger accounts in your niche tends to compound faster because it puts you in front of an audience you haven't earned yet. One documented case built 70% of their growth time around strategic replies to high-follower accounts, with replies within the first 15 minutes earning 3 to 5x more visibility than late ones. Keep posting original content too, but don't expect it to carry your growth alone while your following is still small.