Open any LinkedIn post with 50+ comments and you can spot the AI-generated replies instantly. They start with "Great insight!" or "This really resonates with me." They use em dashes where no human would. Every sentence is perfectly parallel. The whole thing reads like a press release, not a person.
The irony is that these tools were supposed to save time while making you look good. Instead, they make you look like you stopped trying.
Two things we shipped recently change this. Here's what they are and why they work.
What Actually Makes AI Replies Obvious
It's not the ideas — AI can generate solid ideas. It's the surface patterns. Specifically:
- Em dashes everywhere. The long dash (—) is a staple of AI output. Real people use commas or just end the sentence.
- Hollow openers. "This really resonates," "I couldn't agree more," "Such an important topic." No human types these.
- Perfect parallel structure. AI loves lists of three things with identical grammatical weight. It reads like a PowerPoint slide, not a comment.
- AI vocabulary. "Delve," "leverage," "navigate," "seamlessly," "robust," "foster." These words appear in AI output at ten times the rate of human writing.
- Filler phrases. "Thank you for sharing this," "In today's landscape," "At the end of the day." Hollow, vague, and immediately recognizable.
The problem isn't that people know you used AI. It's that these patterns signal you didn't think carefully about what you were saying — you just hit a button and posted.
What We Changed: The Authenticity Pass
We went through our reply generation algorithm and explicitly trained out these patterns. The prompts now include a hard ban list: no em dashes, no filler openers, no AI vocabulary, no hollow phrases, no starting a sentence with "I."
The instruction we give the model is to write like a smart colleague responding in a Slack thread — casual and substantive, not crafted and polished. Imperfect phrasing is fine. Incomplete sentences are fine. Short and direct beats long and eloquent every time.
The result is output that reads like something you actually typed rather than something that passed through a content filter.
Brand Voice: Replies That Sound Like You Specifically
Eliminating AI patterns gets you to "sounds like a human." Brand Voice gets you to "sounds like me."
There's now a text field in the ReplyGen sidebar called About You / Brand Voice. You describe yourself there — your role, how you write, what matters to you. A few sentences is enough:
"I'm a product designer at a B2B SaaS company. I'm direct and prefer concrete examples over abstract principles. I like asking questions that challenge assumptions without being confrontational."
Every reply generated from that point draws on your description. The output reflects your perspective, not a generic professional voice that could belong to anyone on LinkedIn.
This matters most when you're building a professional presence around a specific niche or point of view. If people read ten of your comments, they should be able to tell something real about who you are. Generic AI output works against that — Brand Voice works with it.
How to Use Both Together
The authenticity pass runs on every generation automatically — you don't need to do anything. Brand Voice requires a one-time setup in the sidebar.
The combination changes how the tool feels. Instead of generating something you then have to edit to sound like yourself, you get a draft that's already close to your voice and free of the patterns that would give it away. The edit is small — adjusting a word or adding a specific reference — rather than a complete rewrite.
Over time, the replies you generate start to reinforce a consistent voice rather than average out to a generic one.