How to write good replies on X that get noticed
A good reply adds one specific thing the post didn't have: a fact from your experience, a counterexample, a useful question or a fix. Keep it short enough to read in a few seconds, and choose posts where you know the subject. Generic praise gets ignored because it gives nobody a reason to tap your name.
Who sees your reply
This part changes how you should write. A filter in X's feed code removes replies from accounts the viewer doesn't follow, under the default settings. Your followers may still see them, at a discount. Strangers mostly find your reply under the original post.
So write for someone reading that thread. They already care about the topic. Your reply has to make sense to them and make them curious about you.
X's conversations help page says reply ranking considers whether the original author replied, whether the reader follows the replier, and whether the replier has X Premium. X's code also has a reply scoring system that uses language models, but X keeps those prompts private. So nobody outside X can list exactly what it rewards. Treat any guide that claims to know the formula with care.
You can't control most of that. You can write something the author wants to answer.
Five patterns, with before and after
All of these are invented examples. Swap in details that are true for you, and never make up a result.
1. Add your own number. Post: "Cold email is dead."
- Before: "So true, nobody reads them anymore."
- After: "Mine still work when they're under 60 words and mention one thing from the person's site. Long templates get nothing back."
2. Give a counterexample. Post: "Never launch without a waitlist."
- Before: "Disagree."
- After: "A waitlist helps if people are already asking. For a small tool I'd ship in a week and ask ten users directly. A waitlist of strangers tells you little."
3. Ask a question only a practitioner would ask. Post: "We cut our meetings by half this quarter."
- Before: "Love this! How did you do it?"
- After: "What replaced the status meeting? We tried written updates and people stopped reading them by week three."
4. Finish the job. Post: "Anyone know why my Postgres query got slow after adding an index?"
- Before: "Have you tried checking the query plan?"
- After: "Run EXPLAIN ANALYZE. If the planner picked the new index and got slower, stale statistics are a common cause, so try ANALYZE on that table."
5. Be funny only if you're funny. A joke that lands can travel. A forced one reads like a brand account. If you wouldn't say it to a friend, skip it.
Choosing posts to reply to
Pages on this topic often tell you to reply to big accounts. Big threads can bury you, so choose by fit first.
Common practitioner advice, not a rule from X:
- Pick accounts a bit larger than yours whose replies still get read.
- Favor posts that ask something or make a claim you can test against your experience.
- Reply early when you can. One guide recommends replying early in the thread. Early replies face less competition, but we found no data that proves a specific cutoff.
- Skip posts where your only thought is "agreed". Like them instead.
If the author replies to you, answer back. The help page above names an author's reply as a ranking factor, and the conversation may pull in other readers too.
A quick check before you post
Read your draft and ask four questions:
- Would this still make sense if the author's name were hidden?
- Does it contain something the post didn't say?
- Could I copy and paste it under a hundred other posts? If yes, delete it.
- Is it free of links and pitches? Save those for when someone asks.
The same guide favors quality over volume, and we agree. One sentence can be enough, as long as it's a specific sentence.
Habits that hurt you
Repeating yourself. X has a duplicate-text rule for replies that can label matching replies as copypasta spam. Reusing a template could trip it even if you type each one by hand.
Annoying people. X's README describes jobs that label accounts from how others respond, including blocks, reports and spam reports relative to likes. Picking fights or pitching in every thread invites those signals.
Automating it. X's developer guidelines say AI-generated replies need prior approval, and that deploying them without approval is a violation. They also say keyword-triggered replies get rejected. Reply tools that post by themselves can put you on the wrong side of this. Drafting with help and posting by hand is a different activity, though X doesn't publish a ruling on every case.
Joining reply pods. The authenticity policy bars coordinating to exchange engagement. X's code comments also say that going straight to a post from a group chat has no ranking impact.
Make the click worth it
A good reply earns a look at your profile. Then your profile has to close the deal. Someone who just read your Postgres answer should see a bio that says you work on databases. This bio guide has rewrites you can copy. If your reply drafts come out sounding canned, these fixes for AI-sounding text apply to replies too.
Repli suggests posts worth replying to each morning and drafts replies in your voice, and you decide what gets posted.
Paste any post into the reply generator on our free tools page to get three starting points.