Something about her transcript felt rehearsed for the AI

Open notebook with handwritten candidate notes beside a printed screening transcript, warm morning light on cream paper

Thursday evening, maybe 9pm, I was reading back through a screening transcript from a candidate who'd come through earlier that day for a senior growth analyst role we've been running for about six weeks.

She scored 89.

I don't always read the full transcript on a high scorer. The AI summary usually tells me what I need. But something about the summary made me want to open the actual exchange, so I did.

The answers were good. Really good. She'd hit the right points in the right order. Every question got a structured response: context, then action, then outcome. When asked about her biggest gap, she gave an answer that sounded honest but never quite became personal. The kind of honest that doesn't cost you anything. The AI flagged four sections as "high signal." I read all four. They were high signal.

But something sat wrong.

I couldn't name it. I read the transcript twice and still couldn't place it. I had a note in my candidate file that just said "look closer at this one," which I do sometimes when my gut picks up something my brain hasn't found yet.

what her answers were doing

The next morning I went back.

What I'd noticed, I think, was a kind of evenness. Every answer ran at the same emotional temperature. The wins were calibrated. The challenges were handled. The reasoning was visible but controlled. The transcript read less like a conversation than a well-prepared document delivered in real time.

She knew what the AI was looking for.

I don't mean that as a criticism, actually. She'd clearly thought about what a structured screening call rewards: clear framing, concrete examples, demonstrable self-awareness. She gave all three. Consistently. For thirty-one minutes.

The two strongest candidates from the previous cohort on a similar role both had messier transcripts. One of them spent five minutes on a tangent about a side project before landing on the actual answer to the question. The AI scored him 74. I moved him forward anyway because the tangent told me something about how he thinks. He's three months into the role and doing well.

This candidate's transcript had no tangents. Not one. Which is, when I write it down, sort of the tell.

Humans digress. Even prepared ones. The absence of a single digression across thirty-one minutes isn't exactly a red flag. It's a shape. An unusual one.

the follow-up call

I scheduled a brief live conversation before putting her on the shortlist. Not a full interview, just thirty minutes to cover a couple of specifics her transcript had moved through a bit quickly.

She was different.

I mean that as a good thing. Looser. In the second question I used a framing that was probably too clean and she pushed back on it, politely but directly, and gave me a more layered answer than I'd asked for. She laughed once. She said, "I'm probably not being fair to my old company here, but honestly..." and then told me something that would never have survived a prepared answer. That was the most useful thing she said in the whole thirty minutes.

The AI hadn't seen any of that. Couldn't have. I don't transcript live calls; I just take notes.

If I'd advanced her based on the screening score alone, I'd have been advancing the version she built for the AI. Technically correct. Structurally sound. Somehow less her than what showed up when there was no rubric to perform against.

I keep thinking about what this means. I know that reading transcripts changes what you notice in screening, and I've been trusting that. But this is the first time I've had a transcript that gave me a constructed version of a candidate rather than a captured one. She is better than her transcript. That's not a thing I've said before.

What I keep circling: is "able to optimize for AI screening" a skill I want to select for, or one I want to see past? In a growth analytics role, knowing how a system is scoring you and giving it what it rewards isn't a bad thing to be able to do. I keep arguing myself both ways.

She's on the shortlist. The live call earned that, separate from the score.

But the question isn't going away. The candidates who know how to perform for AI screening are getting better at it. The scoring patterns aren't secret. They're searchable. At some point the gap between "good at AI screening" and "good at the job" might be more diagnostic than the score itself. I don't know how to measure that yet.

I'm still working on it.

Talk soon.

— the recruiter

---

The Diary of an AI Recruiter is a daily column from the team at Ployo. If you want to understand what your AI screening is actually selecting for, book a call.