The worst moment is not writing a bad reply. It is writing a reply that is technically fine and still sounds like it came from nowhere.
You read it back and can feel the problem immediately: too balanced, too complete, too pleased with itself. It does not sound wrong. It sounds unearned.
The giveaway is distance
A reply sounds fake when it stands outside the conversation and explains the situation back to everyone. Real replies usually stay closer to the room.
Someone posts, “finally shipped the thing I’ve been avoiding for two months.” The fake version says, “That’s such an important milestone and a testament to your perseverance.”
The sendable version is closer to: “the two-month avoidance final boss is defeated. huge.”
It notices the actual sentence. It does not make a speech about productivity.
Do not add quirks. Remove the performance.
A common fix is to sprinkle in slang, lowercase, jokes, or a typo. That can make the reply worse, because now it sounds like a polished answer wearing a fake hoodie.
The better edit is smaller:
- delete the warm-up sentence
- delete the recap of what everyone already read
- keep one specific observation
- end before it turns into a conclusion
Most social replies do not need a beginning, middle, and end. They need one useful beat.
The simple test
Ask: would this reply still make sense if it was pasted under 50 different posts?
If yes, it is probably too generic. A good reply should depend on the message in front of it. Even a short one should carry some evidence that you were actually there.
That is the job rawreply is built around: not making replies longer, not making them more impressive, just making them easier to send without the assistant voice leaking through.
If a draft already has the fake feeling, start with sounds like ChatGPT. For broader rewrites, use make my reply sound human.