I’ve been running campaigns for about three weeks now, and I’m hitting a wall. My reply rate is sitting around 3%, which isn’t terrible, but it’s way below what I expected when I started personalizing everything.
Here’s what’s happening: I’m using LiSeller to pull in hyper-personalized messages based on each prospect’s role and company, and on paper it looks great. The messages mention their specific industry, reference their LinkedIn activity—all that stuff. But when I actually read them back, they still feel… templated? Like, there’s a pattern to them that a human can spot from a mile away.
I’ve been thinking about this a lot, and I’m wondering if the problem is actually my AI prompt. Like, maybe I’m being too prescriptive in how I’m asking the system to write? Or maybe the issue is my targeting—I’m casting a pretty wide net (any director-level person in SaaS), so the messages are broad enough that they don’t really land.
I’ve also noticed that when I manually tweak a message after the AI generates it—just adding a personal note or changing a phrase to sound more conversational—those outreach attempts get way better responses. But obviously I can’t do that at scale.
So here’s my question: are you finding that the initial AI output needs serious editing to feel human, or is that a sign my targeting is too loose and the system can’t write something truly personalized without knowing more about the person?
This is the copywriting problem everyone hits. The issue isn’t your AI prompt—it’s that you’re leading with context instead of curiosity. You’re saying “I see you’re a director at a SaaS company” when you should be saying “I noticed you just posted about building remote teams—here’s what we found works.”
The best personalization isn’t smarter targeting. It’s a better hook. Your AI can know the prospect’s role, but it doesn’t know why they should care about talking to you today. That’s on you.
Here’s what I’d test: rewrite your AI prompt to start with a genuine observation or question, not flattery. Something like “Based on [specific thing they posted/company metric], I’m guessing you’re probably dealing with [real problem].” Then shut up and let them respond.
Your 3% reply rate screams that your opening isn’t compelling enough. The personalization is working, but the hook is weak.
Also—directness wins. Cut the fluff. Every extra sentence is a reason for them to delete. I’ve noticed the highest reply rates come from messages under 50 words that ask a specific question. Your AI is probably being too careful and formal. Tell it to be conversational, almost casual.
You’re thinking about this the right way, but there’s a technical solution too. Have you set up conditional logic in your AI prompt based on data fields? Like, if you’re pulling in their latest post or company news, you can feed that directly into the prompt as a variable. That makes the personalization actually specific, not just role-specific.
I pipe our prospect data through to Zapier, which feeds it into a custom prompt template, and then it goes back into LiSeller. Basically, the AI isn’t just writing about their role—it’s reacting to actual signal data. The outputs feel way more human because they’re addressing something real, not just surface-level company info.
Try segmenting your audience tighter and creating role-specific prompts instead of one master prompt for all directors.
In recruiting, I’ve learned that the personalization people actually respond to is when you show you’ve done your homework on them as a person, not just their job title. I spend 30 seconds looking at a candidate’s recent posts, endorsements, or projects before I message.
What if you pulled in one additional data point—like their last LinkedIn post topic or a skill they were recently endorsed for—and fed that into your prompt? It sounds like a small change, but it shifts the message from “I know you’re a director” to “I saw you working on X, thought of you.”
That personal touch is what breaks the template feeling. The AI helps you scale, but the specificity has to come from somewhere real.
One thing to watch: over-personalization can actually trigger spam filters if you’re inserting too many unique variables. LinkedIn’s algorithm is pretty smart about detecting when messages are too customized too quickly. Make sure you’re not crossing into territory that looks like mass personalization—there’s a sweet spot between generic and hyper-specific.
From a safety angle, I’d also suggest testing on a smaller segment first before scaling whatever changes you make. Your 3% baseline is actually solid—don’t burn your account trying to hit 10% overnight.
I’ve been there. Honestly, what fixed it for me was just getting more aggressive with my hooks. I stopped using LiSeller to write my entire opening message and started using it just for the body—I hand-craft the first two sentences to be a genuine question or observation.
My reply rate jumped from 2.5% to almost 5% when I did that. Takes a bit more manual work upfront, but worth it. And I batch the manual work—write 20 hooks at a time, then let the AI fill in the rest.
Have you run any A/B tests on your prompts themselves? Like testing a curiosity-based prompt against a benefit-based prompt?
Great question. The AI output quality really does depend on how specific your prompt is. Here’s what I’d suggest: when you’re setting up your personalization prompt in LiSeller, include instructions to vary sentence structure and avoid repetitive openings. Something like “Mix short and long sentences, avoid starting with ‘I noticed…’ twice in a row.”
Also, the targeting thing you mentioned—that’s real. If your audience is too broad, the AI has to stay general. Try creating 3-4 separate campaigns for different buyer personas, each with its own customized prompt. You’ll see the messages feel way more natural because they’re written for a specific person, not an average director.
And test: create two versions of your prompt (one more formal, one more casual) and run them against the same small segment. You might be surprised which one wins.
The conversion issue you’re describing is textbook personalization-without-strategy. Here’s the reality: generic personalization (using their name, title, company) doesn’t drive conversion. What drives conversion is relevance—showing them you understand their actual business problem.
Your targeting is too wide. “Director-level in SaaS” covers everyone from product directors to sales directors with completely different pain points. That’s why the AI can’t write something that lands.
Here’s what I’d do: narrow your ICP to one specific buyer persona (e.g., “VP of Sales at Series B SaaS companies with 20-50 reps”). Now your AI can write about a specific, universal problem they all face. Your reply rate will climb because the message is actually relevant, not just personalized.
Personalization at scale only works if the targeting is strategic first.