Two years ago, a Reddit comment with perfect grammar and an em dash was just a well-written reply. Now it's a tell.
Across technical forums, Q&A sites, and social media platforms, there's a visible shift in how people engage with discussions. Replies are longer, more structured, and suspiciously well-punctuated. The pattern is consistent enough that experienced community members have started treating certain formatting choices - numbered lists, diplomatic hedging, that distinctive dash usage - as signals that the response was generated by ChatGPT or Claude rather than typed by a human.
The concern isn't about grammar quality. It's about what happens to community knowledge when a growing share of answers come from people who pasted a question into an AI, copied the response, and posted it without verifying whether it's correct. On highly technical subreddits and Stack Overflow threads, this creates a specific problem: the responses look authoritative - polished prose, confident tone, clear structure - but may contain subtle errors that only someone with actual domain expertise would catch.
This connects directly to the "cognitive surrender" problem identified in recent research from the University of Pennsylvania, which found that people follow AI-generated advice 79.8% of the time even when it's wrong. If the person posting the AI reply didn't verify it, and the person reading it trusts the polished formatting, bad information gets laundered through two layers of uncritical acceptance.
There's no clean solution here. Platforms can't reliably detect AI-written text, and banning it would be unenforceable. The practical response for anyone relying on community forums for technical answers is the same boring advice that's always applied: verify claims against primary sources, be skeptical of replies that sound too smooth, and weight answers from accounts with established track records over anonymous perfect paragraphs.