Readers on X have become quick at spotting a reply that a language model wrote and nobody edited. They rarely stop to explain how they know. They just scroll past, and sometimes they mute.
The signs are habits of the models, and every one of them can be fixed in a few seconds. This list is for people who use AI drafts and want their replies to read as their own.
A caution before the list: these are patterns, not proof. People wrote in tidy threes and used "delve" long before language models did. Do not use this list to accuse anyone. Use it on your own drafts.
1. It opens with praise
"Great point!", "Love this perspective", "Such an insightful take". Models are trained to be agreeable, so they start by agreeing.
Fix: delete the first sentence. The reply almost always survives, and starts where it should have.
2. It repeats the post back
"You're absolutely right that onboarding is crucial for retention." The author knows what they wrote.
Fix: remove the restatement and keep only what is new.
3. It could sit under any post
"Consistency really is the key to success." No detail from the post, so no evidence anyone read it.
Fix: add one specific thing from the post: a number, a name, the exact claim.
4. The vocabulary of nobody
Game changer, unlock, leverage, delve, landscape, journey, elevate, robust, seamless, "in today's fast-paced world". These words are common in model output and rare in conversation.
Fix: replace each with the word you would say out loud. "Use", "big", "works", "helps".
5. Everything comes in threes
"Clear, concise and compelling." "It saves time, reduces errors and improves morale." Models love a list of three because it sounds finished.
Fix: keep the one item you can stand behind and drop the other two.
6. "It's not just X, it's Y"
"This isn't just a product update, it's a shift in how teams work." A dramatic contrast that says little.
Fix: say the second half plainly, if it is true. Often it is not, and the sentence can go.
7. Long dashes and perfect punctuation
A reply with an em dash in the middle, a semicolon and flawless commas looks typeset. People on X write fragments, skip the full stop and start with "and".
Fix: break one sentence. Use a full stop or a comma where the dash was.
8. It hedges everything
"This can often be a potentially useful approach in many cases." A model avoids being wrong by avoiding saying anything.
Fix: commit. "This works." If you are not sure enough to commit, ask a question instead.
9. It ends with a question nobody asked
"What has your experience been?" bolted onto the end, to invite engagement.
Fix: delete it, or replace it with a question about a specific detail of the post.
10. Emojis and hashtags from nowhere
A rocket at the end. A hashtag in a reply. Nobody writes replies with hashtags.
Fix: remove them, unless that is truly how you write.
11. Every reply has the same shape
One reply looks fine. Ten side by side, each with a compliment, a point and a question in the same rhythm, look like a machine. Visitors to your profile see them side by side.
Fix: vary the form. A one-liner, then a question, then a disagreement. See nine types of replies for a set to choose from.
12. It is too complete
A model answers the whole post, covers both sides and wraps up. A person picks one thing and reacts to it.
Fix: choose the one sentence that matters and delete the rest.
The quick version
If you only have time for three checks:
- Delete the first sentence if it is praise.
- Add one detail from the post.
- Change one word to a word you actually use.
That takes under ten seconds and removes most of what readers notice.
Why the tool matters as much as the edit
How much editing a draft needs depends on the instructions the model was given. A tool that sends "write a reply to this post" will give you most of the twelve signs every time. A tool that tells the model what to avoid gives you a draft that starts much closer to done.
XAICG is built around this list. Its drafts are written to skip generic praise, to leave out hype words, long dashes and hashtags you did not ask for, and to react to one concrete detail in the post. It keeps your recent drafts in mind so that the next one has a different opening and shape. You describe your own voice once in Settings, which is the single biggest difference. How to make AI replies sound like you shows what to write there.
No tool removes the last step. Read the draft, add the thing only you know, and post it yourself. A reply that has been through your hands is yours, whatever wrote the first version.