I’ve been running outreach for about three weeks now, and I’m hitting a wall. My reply rates are sitting around 2-3%, which is honestly frustrating because I thought personalization was supposed to fix this. The thing is, I’m using LiSeller’s AI to generate my messages, and they’re definitely tailored to each person—I’m pulling in their job title, company details, recent posts, all that stuff. But when I read them back, something feels off. They still sound… polished? Corporate? Not like something a real person would write.
I’ve tested a bunch of variations—shorter messages, longer ones, different hooks—but the underlying problem feels like the tone. It’s like the AI is writing in this “professional” voice that screams automation. I’m wondering if I’m just not setting up the personalization right, or if there’s a fundamental issue with how I’m structuring my messages.
The crazy part is that when I manually write outreach (which I do for like 5 people to test), those get way better response rates. So it’s definitely the message tone, not my targeting. I’m thinking maybe I need to dial back the formality, add more conversational elements, or maybe even inject some casual language?
Has anyone here actually cracked the code on making AI-generated messages sound genuinely human? What’s the actual difference between a 2% reply rate message and one that converts at 8-10%? I’m guessing it’s not just about personalization depth—it’s about sounding like you actually know the person, not like you just ran their LinkedIn profile through a template.
This is the exact problem I see with most personalized outreach—people think adding their company name = personalization, but that’s not the hook. The hook is the reason they should care about talking to you right now.
Here’s what I’d test: instead of leading with who they are, lead with what you observed about them. Like, don’t say “Hi [Name], I saw you work at [Company].” Say something like, “I noticed you just published that post about [specific topic]—your take on [specific detail] is exactly what I’m seeing with our clients in [industry].”
The difference? The first is flattery. The second is proof you actually read their stuff. AI struggles with this because it defaults to safe, professional language. You need to tell it to be specific and opinionated.
Also, cut the corporate fluff entirely. Real humans write short. They use fragments. They repeat words. They say “hey” not “greetings.” The more conversational and slightly imperfect your message feels, the better it performs. Try rewriting one of your best-converting manual messages and feed that as a style example to your AI prompt in LiSeller. That might bridge the gap.
You’re probably looking at this wrong—the message tone issue often comes down to how you’re structuring your AI prompt. Most people just drop the variables in and let it go, but LiSeller’s AI is actually pretty customizable if you dig into the persona section.
What I do is create a custom persona that sounds like me—casual, a bit sarcastic, uses contractions, throws in emojis sparingly. Then I feed in a few examples of messages that actually worked for me and tell the AI to match that style. It’s a prompt engineering thing more than a platform limitation.
That said, if you’re not already doing this, you might want to A/B test your messaging at the sequence level too. Run 50% with your high-personalization AI approach and 50% with shorter, simpler messages (still personalized, but way less embellished). My gut is that shorter wins. The more you write, the more chances you have to sound robotic.
I deal with this constantly in recruiting. High-level talent can smell automation from a mile away, and they’re already skeptical of outreach. What changed my game was realizing that specificity in the personalization actually makes it feel less robotic, not more.
Instead of generic observations, I reference something that took effort to find. Like, I’ll mention a specific project they led, or a technical decision I disagreed with (in a friendly way), or a hire they made that I thought was smart. This shows I didn’t just run a scraper—I actually looked at their profile.
For your AI prompts, try including instructions like “sound like you’re texting a colleague, not writing an email” or “use first-person casual tone.” And honestly? A 2-3% reply rate from AI feels in line with most platforms. The 8-10% rates I see usually come from either (a) incredibly niche targeting with warm intros, or (b) a LOT of manual refinement on the hook. It’s not just the tone—it’s usually the message + the targeting combo.
Before you keep tweaking tone, I want to flag something: sometimes messages feel robotic because of how they’re being sent, not just what they say. If your proxy is rotating too frequently or your account is sending 100+ messages a day from a brand-new account, LinkedIn’s filters might already be ghosting your messages—meaning good tone won’t matter.
Make sure your account is warmed up properly. Spend 3-5 days doing normal LinkedIn activity (liking posts, commenting, connecting organically) before you fire up the AI sequences. That alone can bump your rates up. Then, dial back your daily send limit to like 20-30 personalized messages instead of blasting. Quality over volume.
After you’ve confirmed your account health is solid, then focus on the tone. But start with the foundational stuff first—account safety and pacing. A perfectly written message from a flagged account still won’t convert.
Great question, and this is something a bunch of users have mentioned. Here’s what I’d suggest: in the AI Message Generator, there’s a “Tone & Style” section that most people skip. That’s where the magic happens. Instead of just picking “Professional,” try “Conversational” or “Casual colleague.”
Also, the persona prompt you feed the AI matters way more than people think. Instead of just listing variables like [Name] and [Company], write out a mini-scenario like “You’re reaching out to [Name] because they just published a post about [topic] and you have a strong perspective on it. You want to start a genuine conversation, not pitch anything yet.”
The more context you give the AI about why you’re reaching out, the less generic it sounds. It’s not just about personalization depth—it’s about giving the AI a reason to write naturally instead of Safe Corporate Mode. Try that and report back on your reply rates?
You’ve identified the core conversion challenge correctly: tone. Here’s the data: prospects respond to messages that feel like they’re from someone who knows them, not someone running a sequence. A 2-3% baseline is actually realistic for cold outreach, but 8-10% is achievable if you’re hitting three things simultaneously: precision targeting, genuine personalization (not just variable insertion), and conversational copy.
The mistake I see most is that personalization usually means “I mentioned your company name.” That’s not personalization—that’s just variable replacement. Real personalization means you reference something that would take 5-10 minutes to research. A recent post, a specific hire, a company milestone.
For your AI setup, I’d run a controlled test: take your best-performing manual messages and analyze them for tone. What words do you use? Sentence length? Level of formality? Then create an AI prompt that explicitly mimics that style. Feed it examples. The AI will learn the pattern.
Second thing: A/B test at scale. Split your list 50/50 between your current AI tone and a experimental “hyper-casual” variant. Run both for a week, measure reply rates side-by-side, and go with the winner. That removes opinion and gives you data.