The Polish Trap: Screening for Substance When AI Writes a Better Résumé Than the Candidate
Photo: The Library of Congress, Public domain, via Wikimedia Commons
A decade ago, a well-formatted résumé was a mild signal of professionalism. Today it is close to noise. Within minutes, any applicant can generate a posting-aligned summary, a coherent career narrative, and a cover letter that echoes the company's own language back at it. None of that constitutes fraud. It is the new baseline, and treating it as evidence is the screening error of the moment.
The practical consequence for specialized hiring is uncomfortable: the documents in the pile have never looked better, and they tell you less than they used to. Recruiters who once read for clarity now find themselves reading for authenticity, which is a far harder judgment to make from a PDF.
What AI has flattened and what it hasn't
Flattened: formatting, phrasing, keyword alignment, structure, cover letters, thank-you notes, and the general polish of self-presentation.
Untouched: judgment under ambiguity, recall of operational detail, physical and technical execution, the ability to explain a trade-off to a skeptical stakeholder, and the tacit feel for when a process is about to fail. These remain expensive to fake and easy to verify — provided the screening process actually asks for them.
The governing principle is straightforward. Test the work, not the writing.
Move one: a targeted work sample
Replace one interview round with a ninety-minute exercise built from a real, sanitized piece of the job. A messy dataset, a contract clause with an ambiguous term, a support ticket with an unclear root cause, a wiring diagram with a fault. Score it against a rubric written before the candidates arrive, with dimensions for accuracy, approach, and how the work was communicated.
Ninety minutes of realistic work reveals more than three conversational rounds, and it costs less than a bad hire.
Move two: the specificity probe
Polished narratives survive broad questions. They rarely survive particulars. Ask for the number, the disagreement, the failure, and the consequence. What was the defect rate before and after? Who opposed the change, and what was their objection? What broke in the second week?
Follow with two questions that are difficult to prepare for: describe a time you were wrong about something technical, and walk through how you discovered it. Fabricated depth tends to dissolve at the second or third follow-up, not the first. Take notes on specificity, not confidence.
Move three: portfolio review with a decision trace
Ask for the intermediate work, not only the final artifact. The rejected draft, the version that failed review, the timeline of revisions. A finished deliverable can be described by anyone; the sequence of decisions behind it belongs to the person who made them.
This is especially useful in design, engineering, marketing analytics, and technical writing, where a polished final output is the easiest thing to borrow and the least informative.
Move four: tool-in-hand troubleshooting
For roles where execution matters, put the candidate in the environment. A live system with an introduced fault, a spreadsheet with a broken formula chain, a piece of equipment that will not calibrate. Observe the sequence: what they check first, how they isolate variables, whether they ask for help at a sensible moment. Skill shows in method, not in vocabulary.
Move five: triangulate timelines and references
Verify dates, titles, and scope against public sources — state licensing boards, professional certifications, published work, conference records. For the reference calls, skip the general question about strengths and ask about a specific deliverable the candidate described. Ask whether the referee recalls the project at all, and what the candidate's contribution actually was.
Move six: blind first-pass review
Strip names, schools, and formatting before the initial screen. Score against a structured rubric, independently, before any group discussion. Polish creates a halo that contaminates consensus, and independent scoring is the cheapest available correction.
Move seven: anchored rating scales
Use the same questions for every candidate, with behavioral anchors describing what strong, adequate, and weak answers look like. Score individually before the debrief. Group discussion should reconcile differences, not generate impressions.
Guardrails worth observing
The methods above are more defensible than unstructured conversation, but only if they are applied consistently. Keep exercises job-related, document the criteria before candidates arrive, and apply the same process to every applicant for the role. A growing patchwork of state and local rules governs automated tools used in hiring, including disclosure and audit obligations in some jurisdictions; employers deploying AI-assisted screening or assessment platforms should confirm their obligations with counsel rather than assume the vendor has handled it.
The economics favor the harder process
A ninety-minute work sample and a structured panel add perhaps four hours of internal time per finalist. A mis-hire in a specialized role costs a multiple of annual salary once recruiting, onboarding, lost productivity, and replacement are counted. The arithmetic is not close.
Keep the résumé, change its job
None of this argues for abandoning the résumé. It remains an efficient map of where someone has been and what they claim to have done. It simply cannot serve as a verdict. Used as a map, it tells you which work sample to build. Used as a verdict, it rewards the candidate with the best tools rather than the best capability — and in specialized hiring, that distinction is the whole game.