The Best Résumé Is a Red Flag

Hey everyone, Neil here. You're reading High-Signal Hiring. Hiring systems from 20+ years of global recruitment experience and 500+ technical hires. Zero noise and instantly actionable.

Last week I argued the live follow-up is the one part of your interview a candidate can't fake, because the AI tools feeding them answers lag a second or two behind an unscripted question. That was about the interview itself.

This week I want to back up one stage, to the very top of your funnel. Before the interview, before any take-home, there's the moment you decide who's even worth an hour of your time. For most founders that call still comes down to a résumé. And the résumé has quietly stopped telling you anything.

You'll learn why the document you're screening is now the same for every applicant, why buying a tool to screen it faster makes the problem worse, and how to rebuild your first filter this week around something a model can't produce.

Not a subscriber yet? Sign up here

The résumé used to do a job. It was a rough filter that let you guess, from a page, who might be worth talking to. That filter worked because writing a good one took effort, and effort was a signal.

That has gone. More than half of applicants now use AI to write their résumés and cover letters, and 81% have used or plan to use AI somewhere in their job search, according to LinkedIn's 2026 research. The polished, keyword-matched, achievement-stuffed document you were trained to look for is now the default output of a free tool. The flawless résumé that used to earn someone the top of your pile now signals nothing more than a paid subscription. Everyone clears the bar you're measuring against, so the bar tells you nothing.

The pile got deeper at the same time. Popular roles now draw hundreds of applications, most of them fired off in under a minute by someone applying to fifty others that morning. It takes far longer to get through and gives you almost nothing to go on.

The instinct is to buy your way out. Bolt an AI screener onto the top of the funnel and let it rank the flood for you.

That automates the same mistake at speed. The tool grades the identical AI-written polish, so it surfaces whoever prompted their résumé best. Whether they can do the job is a separate question it never looks at. There's a legal cost creeping in on top of that, since the June 2026 Mobley v. Workday ruling opened the door to holding the vendor, and you, liable for how an algorithm auto-rejects people. Either way you'd be paying to filter faster on a signal that already stopped working.

There's a deeper problem no tool fixes. The engineers worth hiring mostly aren't in your inbox. One hiring engineer in the Pragmatic Engineer's 2026 market read described his inbound as a wall of noise with no real signal getting through, while the strong people move through networks and never fill in your form. I made the fuller version of this case back in Issue #11. The résumé pile was never your best channel, and it is now close to your worst.

Strip it back to what you are trying to learn from that first filter. You want to know whether this person has done the thing, and whether they understood what they were doing while they did it.

A model can produce a plausible account of both in seconds. It just can't make that account true, specific, and checkable. It will happily write "led a migration that improved performance by 40%." It cannot hand you the actual pull request, or explain the one call in that migration the candidate would make differently now, in their own words, with the small details only someone who lived it would have.

That is the ground to screen on. Screen for evidence you can follow and specificity a model can't invent for someone. Polish is free now, and keywords were always gameable. It is the same principle I applied to the interview in Issues #28 and #29, judge what they own rather than what they can generate, moved right to the front of the process.

Most of the fix comes down to three moves, and none need new software.

First, move your evaluation weight off anything asynchronous. Cut the scored take-home and the one-way video. If you keep a take-home at all, treat it as a conversation starter rather than a gate, and assess it live by asking the candidate to extend it in front of you.

Second, stop running a fixed question list. Generate each question live from what the candidate just said. If they mention picking Postgres over Dynamo, your next question comes from that choice, not from row three of your sheet. A fixed script can be transcribed and pre-loaded before you finish asking it. An answer that branches off their own words can't be. This isn't about grilling them on trade-offs, I did that in Issue #28. It's about making the next question impossible to see coming, so the tool never gets its head start.

Third, go deep on something only they lived through. Pick a project from their history and branch into it live. What broke, what they'd redo, who disagreed with them and why. Nobody has a pre-generated answer for the specifics of their own project, and a model can't invent them convincingly on a one-second delay.

Three moves, and none of them need new software.

1️⃣ First, demote the résumé. Stop scoring it. Use it as a basic sanity check on whether a real person sits roughly in the range you need, and nothing more. It doesn't rank anyone.

2️⃣ Second, make your real first filter a single piece of proof of work. One question, sent async. Link one thing you've built and shipped, and tell me one decision in it you would make differently today. It costs the candidate five minutes and it is anchored to something that exists. The strong ones answer with specifics. The pile-fillers can't, because there is nothing behind the polish to draw on. This screen won't catch everyone, and it doesn't need to. It clears out the low-effort majority in one step, and the live follow-up from Issue #29 handles anyone who fakes their way past it.

3️⃣ Third, when you read the answer, follow the artifact and the reasoning, not the adjectives. You're looking for a short, specific, slightly messy account of a real thing. Fluent and generic is what you get when nobody did the work.

One caution, because this cuts both ways. Don't turn it into AI-detection. Trying to sniff out who used a model to write their application is the same losing arms race I warned about last week. Whether they used AI to draft the page is beside the point. Look for the real work and real judgement underneath it, and the tool they wrote it with stops mattering.

The résumé stopped telling you who can do the work. So stop asking it, and ask for the work instead.

Cheers
Neil