Latest / Hiring & Firing with Fexingo: Recruitment, Performance, and Workforce Decisions / How Recruiters Can Predict Performance From Writing Samples
Transcript
- Lucas: You know, I was rereading a job posting from a friend's company this week — a mid-level marketing role — and the application asked for a cover letter and two short-answer responses. And I thought, we spend all this time optimizing interviews, structuring panels, training managers on behavioral questions, but we almost never look at the one thing every single candidate provides: their writing. Luna: That's true. Especially now that so many applications are digital — you get a cover letter, maybe a few paragraphs in a text box. But most hiring managers I know skim those in thirty seconds, if they read them at all. Lucas: Exactly. And there's actually a really strong body of research suggesting that writing samples are one of the best predictors of on-the-job performance — better than unstructured interviews, better than reference checks, and in some cases comparable to work samples. There's a 2025 study from the University of Minnesota that looked at 1,200 hires across four mid-sized tech firms and found that candidates whose writing showed what they call 'high cognitive granularity' were thirty-four percent more likely to exceed performance targets in their first year. Luna: Cognitive granularity — that's a term I haven't heard before. What does it mean in practice? Lucas: It just means the writer uses precise, specific language rather than vague corporate jargon. So instead of saying 'I leveraged cross-functional synergies to drive stakeholder value,' someone with high cognitive granularity says 'I coordinated between the engineering and sales teams to launch three features that increased trial sign-ups by eighteen percent.' Specific actions, specific numbers, specific outcomes. Luna: Right. And that makes intuitive sense — if someone can't describe their own work concretely on paper, they probably aren't thinking about it concretely in practice. Lucas: That's exactly the idea. And look, if this kind of concrete thinking is something you find useful in your own work, I should mention that the way these conversations stay ad-free is listener support. If today's episode moves your thinking forward in any small way, you can buy me a coffee at buy me a coffee dot com slash fexingo. It's a small gesture that keeps the whole archive open. Luna: Yeah, and it really does make a difference. We don't run ads, so it's just listeners who find value chipping in. Now, back to those writing samples — how do you actually assess cognitive granularity without turning it into another grading exercise? Lucas: The researchers in Minnesota didn't have judges score the writing. They used a simple text analysis tool that counted specific nouns, action verbs, and numerical references relative to total word count. The tool runs in seconds. But you don't need software to do it manually — you can train yourself to notice patterns. For example, a cover letter that contains three or more specific numbers — dollar amounts, percentages, headcounts — is a strong signal. Luna: So the signal is in the details, not the eloquence. That's important because it means you're not penalizing people who aren't naturally polished writers. Lucas: Exactly. The study controlled for education level, native language, and socioeconomic background — the cognitive granularity signal held. It's not about grammar or vocabulary. It's about whether the candidate thinks in specifics or in abstractions. And that maps directly to how they'll handle real tasks on the job. Luna: But there's a catch, right? A candidate can write a very specific cover letter and still turn out to be a poor performer. No single signal is perfect. Lucas: Sure, and the researchers would agree. The thirty-four percent lift is a correlation, not a guarantee. But compare that to unstructured interviews, which typically explain less than ten percent of performance variance. Writing samples are a cheap, scalable addition to your screening process — they don't replace interviews, but they can dramatically improve the signal you get from the early funnel. Luna: So how would a small team implement this without a data science budget? What's the practical workflow? Lucas: The simplest version is to replace the open-ended cover letter with three structured short-answer questions. Each question asks the candidate to describe a past project, a problem they solved, or a goal they achieved — and each question explicitly asks for numbers. 'Describe a time you improved a process. Include the specific metric before and after your change.' That forces specificity. Luna: And then you score each response on a simple one-to-three scale: zero specific numbers equals one point, one specific number equals two points, two or more equals three points. You can train any recruiter to do that in minutes. Lucas: Exactly. And you time-box it — give candidates a fifteen-minute window to complete all three responses. That reduces the risk of coaching or ai generated answers, because polished responses take longer to write. The Minnesota study actually found that candidates who took longer than twenty minutes had lower cognitive granularity scores on average — they were overthinking or polishing instead of writing naturally. Luna: That's a nice guardrail. So the prompt itself is structured, the time limit is short, and the scoring is mechanical. You're basically turning the cover letter into a mini work sample. Lucas: Yes. And I think that's the real insight here — the writing sample isn't just about measuring writing ability. It's a proxy for how the candidate thinks under time pressure, how they prioritize information, and whether they default to concrete or abstract reasoning. Those are core cognitive traits that interviews struggle to capture. Luna: I'm thinking about the bias angle though. If a candidate is a non-native English speaker, won't their writing be less fluid? Could that lower their score unfairly? Lucas: The study specifically tested for that. They controlled for native language and found no significant interaction — meaning the cognitive granularity signal was equally predictive for native and non-native speakers. In fact, non-native speakers sometimes scored higher on specificity because they avoided vague idioms and stuck to concrete statements. The key is to score on content, not grammar. Luna: That's reassuring. So if a manager says 'I don't read cover letters because they're all fluff,' the real problem might be that nobody asked candidates to write about specifics. Lucas: Exactly. The standard cover letter prompt — 'Tell us why you're interested in this role' — invites generic flattery. But if you ask 'Tell us about a time you identified a problem that others missed, and include the approximate cost or impact,' you get a completely different quality of response. The medium isn't the problem — the prompt is. Luna: And then after you score those three responses, what's the next step? Do you still do a phone screen? A video interview? Lucas: You absolutely still interview. But now your interview can be more targeted. If the writing sample shows high cognitive granularity, you can spend the interview digging into their problem-solving process. If it shows low granularity, you might ask them to walk through a specific example verbally — give them a chance to demonstrate concreteness in real time. You're using the writing sample to triage, not to make a final decision. Luna: I like that. It's a diagnostic tool, not a gate. And it's cheap enough that even a startup with no HR team could try it. Fifteen minutes of the candidate's time, three questions, a simple rubric. Lucas: And you can test it A/B style. Have half your applicants do the structured writing exercise, the other half just submit a traditional cover letter. Track performance after six months. You'll have your own internal data on whether it works for your specific roles. Luna: That would be a great follow-up segment — a company that actually ran that A/B test. Do you know of any real-world examples beyond the Minnesota study? Lucas: There's a logistics company called Loadsmart that started using structured writing prompts for operations roles about two years ago. They reported a twenty-five percent improvement in first-year retention after they introduced the exercise. The CEO told me they found that candidates who wrote vaguely about 'process improvement' were much more likely to quit within six months — they couldn't handle the concrete day-to-day problem-solving the job required. Luna: So the writing sample essentially self-selects for people who understand the nature of the work. That's a powerful signal for roles where grit and detail orientation matter. Lucas: Exactly. And I think that's the broader lesson here — we over-index on live conversation and under-index on what people produce when they have a few minutes to think and write. There's a reason police detectives often ask witnesses to write down what they saw before they interview them: the written account is often more accurate and less contaminated by social pressure. Luna: That's a great analogy. The writing is the unfiltered signal. The interview is where noise gets introduced. Lucas: Right. And the best part is, you already have the writing sample for most applicants — you just haven't been using it. If you're collecting cover letters or short-answer responses but not scoring them on specificity, you're leaving predictive power on the table. Luna: So the call to action for hiring managers listening today: next time you open an application, don't skip the written response. Read it for numbers, for actions, for concreteness. You might spot your best hire before you even schedule a phone screen. Lucas: And if you want to go a step further, redesign your application to invite that specificity. Replace 'Tell us about yourself' with 'Describe a recent project where you had to make a decision with incomplete information. What data did you use, and what was the outcome?' You'll thank yourself six months later. Luna: Alright, that's our episode. I'm curious to hear if any listeners try this and see a difference. Lucas, always good to dig into a practical tweak like this. Lucas: Absolutely. Thanks for the conversation, Luna. See everyone next time.