I’ve been using LiSeller for about three weeks now, and I’m seeing decent reply rates on my first messages—around 4-5%—but something feels off. When I read back through my sent messages, they technically mention the prospect’s company or recent news, but they still feel… templated. Like, the personalization is there, but it doesn’t feel like I’m actually talking to them.
I think the issue is that I’m relying too heavily on the AI to sound human, when really it’s just inserting variables into a structure that’s still fundamentally the same for everyone. The hook changes, sure, but the cadence of the message feels identical across prospects.
I’ve started experimenting with actually rewriting some of the AI suggestions before sending, injecting my own tone and references that feel more natural to how I actually talk. My reply rates bumped up slightly when I did this, which makes me wonder: am I overthinking this, or is there actually a threshold where hyper-personalization stops being about data insertion and starts being about actual conversation style?
How are you all handling this? Are you editing the AI’s work before sending, or trusting it to nail the tone on the first pass?
You’ve just identified the biggest trap in AI-powered outreach. The platform can insert facts, but it can’t replicate your voice—which is actually your biggest asset. Here’s the shift I made: instead of using the AI message as the final product, I use it as a first draft. Then I strip out the overly “polished” language and inject one or two conversational elements that only I would write.
Example: instead of “I noticed your company recently expanded into the California market,” I rewrote mine to “saw you guys just landed in California—that’s a huge move.” Same fact, completely different feel. My hook also changed: I stopped using “I think you’d benefit from…” and started with a genuine observation or question that showed I actually cared about their situation, not just their lead score.
The AI is great at data synthesis. You’re the one who makes it feel like a real person.
Also—and this is critical—test this hypothesis properly. Send 20 messages using the AI output as-is, then send 20 with your rewrites. Track reply rate on both. I did this, and my rewritten messages got a 6.2% reply rate versus 3.8% on the pure AI versions. Once you see that data, you’ll stop second-guessing yourself. The personalization depth matters way less than the authenticity of the delivery.
This is actually a prompt engineering problem, not a platform problem. The AI is only as good as the instructions you give it. I built a custom prompt in LiSeller’s settings that includes my personal communication style—casual, direct, no corporate fluff. I feed it examples of messages I’ve personally written that got responses, and the AI started matching my voice much better.
If you’re not customizing the system prompt, you’re basically running the default settings. Most users don’t realize they can do this. Set up a prompt template that reflects how you actually talk, and the output changes dramatically. I can share mine if you want to adapt it.
In recruiting, this issue is even more pronounced because high-level candidates smell inauthenticity from a mile away. I started adding one micro-personalization that isn’t data-driven at all—just a genuine reaction to something I found interesting about their profile or work. Like, “saw your talk on [specific topic] last year—that was a perspective I hadn’t considered.” Not templatable, not scalable, but human.
I still use the AI for the research and structure, but the actual message always has one element that proves I spent real time on them. My response rates jumped from 12% to 18% when I implemented this for senior roles. It’s the difference between automation and actually reconnecting with someone.
I’ve been scaling outreach across 3 client accounts, and here’s what I’ve learned: the best-performing messages are the ones that sound like they came from an actual human who did 30 seconds of research, not 5 minutes. The AI messages that try to be too detailed actually underperform.
I now tell my team to use LiSeller’s AI for the heavy lifting—finding the hook, structuring the ask—but then we layer in one authentic detail that comes from our actual research. When we do that, reply rates are consistently 5-7%. When we don’t, it drops to 2-3%. The math is clear.
Great question, and you’re thinking about this the right way. The AI generates messages based on the data and tone you configure, so if you’re getting generic-sounding output, it might be worth revisiting your message template and tone settings in LiSeller. Try adjusting the tone slider toward “conversational” or “casual” and see if the output shifts.
Also, the quality of your input data matters. If you’re pulling generic info about the prospect, the AI will generate generic personalization. The richer your research segment (their recent posts, specific job changes, industry news), the more textured the message becomes. Experiment with feeding the AI more specific details about why you’re reaching out, not just who they are.
You’re experiencing what I call the “data vs. voice” gap. Most AI tools excel at inserting relevant facts, but they struggle with tone because tone is context-dependent. A message that sounds warm to a startup founder might sound sloppy to an enterprise executive in the same industry.
Here’s my framework: use the AI for personalization (facts, research, structure), but make sure your tone template—the voice you want to project—is set correctly before the AI even generates a draft. If your tone is set to “professional but approachable,” versus “enthusiastic,” you’ll get entirely different outputs. The AI responds to these instructions. Make sure you’re actually adjusting them per campaign or per segment.