I’ve been running LiSeller for about three weeks now, and I’m seeing the personalization tokens work—names, company details, all that stuff. But when I read the actual messages going out, they still feel… stiff. Like someone ran them through a filter.
Here’s what I’m doing: I’m pulling prospect data (company size, industry, recent news), feeding it into the AI prompt, and letting it generate the message. The tokens populate fine. But the actual tone? It still reads like an AI wrote it, even though I’m supposed to be getting “human-like” messaging.
I suspect it’s not the personalization tokens themselves—those are working. It feels more like the hook or the overall message structure is still too formulaic. Like I’m using the tool correctly but the strategy behind what I’m asking LiSeller to write is off.
So here’s my question: when you’re crafting the AI prompt or the message template that LiSeller uses to generate outreach, how much of the “human” quality comes from what you tell the AI to write versus how specific your personalization data is? Am I spending time optimizing the wrong thing?
This is the copywriting equivalent of debugging. You’ve nailed the diagnosis—it’s not the data, it’s the strategy. Here’s the brutal truth: most people feed LiSeller generic prompts. “Write a personalized message about their company” is way too broad.
What actually works: start with a specific hook—something that proves you’ve actually looked them up. Not “I saw you work at Company X” but something like “I noticed your team just launched Y feature on your pricing page last month—that’s a smart move because…” Then follow with why you care, not a pitch.
The AI will only sound human if you give it human instructions. Try rewriting your prompt to include 2–3 specific observation types, then let the AI pick one based on what it finds. That constraint actually forces better, less robotic output.
Also—are you varying your sentence structure in the prompt itself? If your template always starts the same way, the AI will too. Mix it up.
One more thing: read your generated messages out loud. If you wouldn’t say it in a coffee chat, nobody else will believe it either. That’s your baseline test for whether the personalization is actually creating human tone or just filling in blanks.
Have you considered pulling in richer data sources before the AI actually generates the message? I pipe in Crunchbase data, recent LinkedIn post sentiment, and even their job posting details into a custom webhook that feeds into LiSeller.
The more context the AI has, the better the output. It’s like giving a copywriter a brief versus a novel—they’ll write better copy with the novel. Try building a simple data layer (Google Sheets + Zapier works) that enriches each prospect record before LiSeller’s AI sees it. Game changer.
Also, test this: run A/B variants where one uses minimal data (just name + company) and one uses rich data. I bet the rich-data version reads way more natural. If that’s true, your “robotic” feeling is just starving the AI of context it needs to write conversationally.
I deal with this constantly in recruiting. High-level candidates smell generic instantly, even with personalization tokens.
What changed for me: I stopped thinking about personalization as “filling in blanks” and started thinking about it as showing genuine interest in their world.
So when LiSeller generates a message, I make sure the prompt tells it to reference something about the prospect’s role or pain point—not just their company name. “You’re managing a team of X engineers dealing with Y problem because of Z industry shift” feels human. “I saw you work at TechCorp” feels like spam with a name in it.
Also: are you A/B testing message tone? Conversational + casual beats formal + buttoned-up almost every time, especially in early outreach.
One last thought: the best personalization isn’t what you write—it’s what you don’t write. Generic fluff about your product or service instantly kills credibility. Focus the message entirely on them: their need, their world, their context. Make the ask tiny. That’s where the humanity lives.
Also: are you using the platform’s built-in follow-up sequences, or are you manually sending everything? Automated sequences often read more robotic because they’re less varied. If you’re using automation, test adding random wait times and slight message variations between variants to keep things feeling fresh.
Real talk: I hit this wall hard before I understood what was happening. Turns out I was overthinking it.
Took one of my top-performing reps, watched them write five messages manually to different prospects, then reverse-engineered the feeling of their approach. The prompt I built from that manual analysis outperforms my templated stuff by literal 2–3x.
My advice: do a few outreaches manually (yes, painful) and note exactly what you say that feels natural. Then codify that tone into your LiSeller prompt. It’s not about the data layer—it’s about matching how you or your best sales person actually communicates.
What’s your current conversion rate, just out of curiosity?
Also: are you personalizing just the first message, or the follow-ups too? I found that keeping follow-ups ultra-simple and context-aware (like “saw you didn’t respond—totally fine, just wanted to make sure this landed”) actually reads way more human than continuing to personalize.
Great question. The human-like quality comes from both the prompt and the data, but here’s what most people miss:
LiSeller’s AI is trained on real human conversations, so it responds best to prompts that describe intent, not just facts. Instead of “write a message about their company growth,” try: “What would you say to someone on your team to get them interested in a quick call without being salesy?”
Also, the personalization tokens by themselves don’t create tone—they create specificity. To add human tone, make sure your prompt asks the AI to vary sentence structure, use contractions, and maybe even throw in a casual word. Play with prompt engineering. Try different phrasings and A/B test the outputs.
Do you want to share an example of a prompt you’re currently using? I bet we can tweak it to sound way less robotic.
Last thing: run a manual audit. Pull 10 messages LiSeller generated over the last week. Read them as if you’re the recipient. If you think “this person doesn’t know me,” that’s your problem. If you think “okay, they clearly researched this,” you’re close. That feedback loop is more valuable than any tool setting.