People Open Up When Something Listens

HireIntel v1: A Chat Meant to Coach the Whole Search

I built HireIntel as a career coach, not just a resume tool, something meant to help job seekers see what wasn't working across the whole search: the resume, the outreach, the interview, the strategy. It started as a chat, and the real numbers say something on their own: 1,012 people used it, organically, nobody was paid to be there, and it logged nearly 3,500 messages back and forth.

What I expected to learn was something spread evenly across all of that. That's not what happened.

What the Usage Data Actually Showed

The same few things came up over and over. "I keep applying but getting no responses" appeared 106 times. Ghosting, getting somewhere in a process and then just hearing nothing, came up 164 times total. Forty-four people landed on "I don't know where to start." Some of that came from people typing freely; some came through quick-start prompts I'd written into the chat myself, so I want to be precise about what the data shows. It's not that people spontaneously confessed something a resume would never ask. It's that when given a menu of real, honest starting points, plus room to type freely, the same handful of problems dominated everything else, by a wide margin, every time. That's still a real signal. A resume doesn't offer "I don't know where to start" as an option, and it doesn't ask either.

One part of the product also answered a question I hadn't even asked yet: which part of the search people actually care about most. The CV-optimizer was, by a wide margin, the most-used feature in the whole thing. People could ask about outreach, interviews, strategy, anything, and most of them kept coming back to the resume. That's the data-driven reason the voice agent I built later narrowed to resume only. Worth being honest about what that data actually proved, though: it showed what people reached for first, the easiest, most familiar thing to ask for, not necessarily what would help them most. Demand and value aren't the same measurement, and that difference matters later in this thesis (Chapter 12).

"Cold, But It Found the Problem Fast"

The chat worked, but I started hearing the same complaint from real users: it felt cold. Methodical, even a little stressful, like a real interview, not a conversation. And in the same breath, the same people would say it asked the right questions and got to the actual problem fast. That combination stuck with me. The tool was doing something useful, extraction, structure, speed, and missing something a person clearly still wanted, warmth. I didn't want to guess at what that meant. I wanted to actually understand it.

Fifty Conversations

So I went and had the conversations myself. I wrote a LinkedIn post asking people to talk to me directly about their job search, no pitch, just a real conversation. It got over 15,000 impressions and more than a hundred people signed up. I couldn't personally talk to all of them, so I ran roughly fifty of those calls myself, twenty-five to thirty minutes each, one person, one real conversation at a time. That became what's now the Impact Page concierge process (more on that in Chapter 13). I also built a voice agent to do the same thing at scale, an AI that called people and talked with them like someone actually curious, not a script reading questions off a list.

What Twenty-Five Minutes Can Do That a Page Can't

Here's what those fifty conversations taught me: you can extract things from a person in twenty-five to thirty minutes of real conversation that never make it anywhere near a resume. People describe themselves completely differently out loud than they do in writing. On paper, everyone flattens into the same careful, formal voice, the one that's supposed to sound impressive. In conversation, with someone genuinely interested, the real person shows up: the way they light up talking about one project and go flat on another, the thing they downplay that's actually the most impressive part. This isn't just my read of fifty calls. It lines up with real research on speech versus writing, in a different context (word of mouth, not job interviews): spoken communication carries far more emotional intensity than written communication, because writing gives people time to deliberate, and deliberation dilutes emotion. I'm applying that mechanism to a hiring conversation, not claiming the study tested one. The word "amazing," said out loud, lands harder than "excellent" typed on a page, even though they mean the same thing on paper.

There's a plainer reason live conversation is worth trusting more than a page, too, separate from warmth or emotional register. Writing gives you a draft, a delete key, time to construct a version of yourself that sounds right. A real-time conversation doesn't. You can polish a resume for a week before anyone sees it. You can't polish an answer to a question you didn't know was coming, and you can't quietly delete the hesitation before it lands. That's not a claim about what predicts good hires. It's a claim about what's harder to fake, and it's the whole reason a live conversation, chat or voice, catches something a written page never will.

And there's something underneath even that. When people talked and felt genuinely heard, something shifted in how they talked about themselves. They'd start hesitant, almost apologetic about their own experience, "it's not that big a deal, but..." And by the end of a real conversation, several of them had, without anyone telling them to, started describing their own work with something closer to pride. Not exaggeration. Just... permission. Like they'd forgotten they were allowed to think their own work was good, and being asked good questions by someone who was actually listening reminded them.

That's not a small thing. People spend so long compressing themselves into a page that they start to believe the compressed version is all there is. Give them a real conversation instead, chat or voice, and they don't just reveal more information. They remember they're worth something beyond the bullet points.

HireIntel v2: Building the Voice Agent

This is exactly why the next version of HireIntel became a voice agent. Text chat had already shown me where the real appetite was (the resume) and surfaced the handful of problems that mattered most. Fifty live conversations showed me what was still missing even from a well-built chat: the warmth and the emotional register that only happens out loud. The voice agent was built to hold both at once, structured enough to extract what the chat already proved people needed help with, human enough to get the part only a real conversation ever reached. The signal was there the whole time, in the chat logs, in fifty calls, and in early tests of the agent itself. What was missing was never the story. It was something built to actually listen for it, warmly enough that people would tell it.