Readers have gotten fast at spotting AI. On Reddit and X, an obviously ChatGPT comment gets downvoted, ignored, or removed — and called out in the replies. The tells below rarely appear alone; it's the cluster that gives it away. Here are 17, with the human fix for each.
Punctuation & structure
- Em dashes everywhere. The #1 tell — several per paragraph. (Full breakdown here.) Fix: one per paragraph, max.
- "It's not just X — it's Y." The antithesis ChatGPT can't resist. Fix: just say Y.
- Perfectly balanced lists. Three items, each the same length, each starting the same way. Fix: uneven, real examples.
- Bold-label bullet points in a casual reply nobody asked to be formatted. Fix: plain sentences.
- A neat summary sentence that restates the point at the end. Fix: end on the last real thought.
- Flawless grammar and spelling in an offhand comment. Fix: contractions, lowercase, the occasional fragment.
Phrases & sign-offs
- "I hope this helps!" / "Hope that helps." Fix: delete it.
- "Great question!" as an opener. Fix: just answer.
- "It's important to note that…" Fix: cut — note it by saying it.
- "In today's fast-paced world…" Fix: delete the whole runway.
- "When it comes to…" / "Navigating the complexities of…" Fix: name the thing directly.
- "A testament to…", "delve," "tapestry," "robust," "leverage." Fix: normal words. (The full list.)
Voice & stance
- It starts with a verdict, not nuance. Human comments open with "it depends" and then name the dependency; bots open with the conclusion. Fix: lead with the caveat.
- No first-person experience. AI describes; humans say "I tried this and…" Fix: add a real anecdote.
- Relentlessly balanced. Every point gets a counterpoint. Real people have opinions. Fix: take a side.
- Over-hedging — "may," "might," "could potentially," "in some cases." Fix: commit.
- Zero personality. No slang, no humor, no edges. Fix: sound like you, not a press release.
Why it matters on Reddit and X
These platforms don't just downvote AI comments — communities actively ban accounts that post them, and the anti-AI-slop sentiment is strong. The goal isn't to "beat a detector," it's to genuinely contribute. A human-sounding comment that adds value is welcome; a polished bot comment is removed.
The fast fix
Reading every draft for 17 tells is slow. The free Prowlo Comment Humanizer rewrites your AI comment in a real Reddit or X voice, strips the tells, and shows you which ones were in your original. Paste, pick a voice, post.
Related reading
- The em dash: the #1 sign of AI writing
- The words and phrases that scream "AI wrote this"
- How to make ChatGPT sound human
- Humanize an AI comment now →
FAQ
How can you tell if a comment was written by AI? Look for a cluster of tells — em dashes, "not X — but Y," "I hope this helps," perfect balanced structure, over-hedging, flawless grammar in a casual context. One is fine; several together is the giveaway.
What words are most common in AI writing? 'Delve,' 'tapestry,' 'navigate the complexities,' 'it's important to note,' 'a testament to,' 'when it comes to.' See the full list.
How do I make AI writing sound human? Cut the tells, write first-person and casual, vary sentence length, and have a human review it — or run it through the free humanizer.