The document stopped being about the person.
Hiring has always run on a proxy: a page of claims, written by the candidate, about the candidate. That proxy worked for decades — not because it was accurate, but because writing it well took effort that correlated with something real. In 2026 that correlation is gone.
Writing a great résumé is now free.
Ask any general-purpose model for a résumé tailored to a specific job posting and you get one in about nine seconds: correct keywords, plausible quantified achievements, a cover letter matched in tone, and rehearsed answers to the twenty questions you were most likely to ask. It costs the applicant nothing and requires no particular skill.
This is not a story about dishonest candidates. Most of what these documents say is broadly true. The problem is subtler and worse: the signal that used to separate applicants has been flattened. When one candidate wrote a sharp résumé and another wrote a clumsy one, the difference told you something — about care, about communication, about how much they wanted it. Now both documents are sharp. The difference has been optimised away.
A tailored application used to take an evening. It now takes one prompt. The applicant who cared most and the applicant who applied to 400 jobs produce identical artifacts.
Every ATS ranks on keywords. Every model knows this. Screening on keywords now selects for whoever used the better prompt.
One-click apply plus zero-cost tailoring means postings that drew 90 applicants now draw 900 — and reading them got less useful, not more.
Same paperwork. Different people.
Here is what the disconnect looks like in practice. Three applicants for one corporate accounts role. Their résumés differ by a single adjective and two percentage points. Read them side by side and you would rank them by whose phrasing you happened to prefer.
“Results-driven account manager. Grew portfolio revenue 40%. Exceptional communicator, trusted partner to senior stakeholders.”
Keyword match 96%Behavior and judgment land inside the target zones for this role. Written answers contain specifics a model could not have invented. All validity checks clean.
“Results-oriented account professional. Grew portfolio revenue 38%. Exceptional communicator, trusted partner to senior stakeholders.”
Keyword match 95%Genuinely strong judgment and relationship instincts. Qualification discipline sits far below what this role needs — invisible on the résumé, decisive in the job.
“Results-focused account leader. Grew portfolio revenue 42%. Exceptional communicator, trusted partner to senior stakeholders.”
Keyword match 97% — the strongest of the three41 identical answers in a row, two failed attention checks, contradictory repeated questions. By keyword rank, he was top of your pile.
The candidate with the best paperwork was the worst applicant. That inversion is the whole problem.
“Won't they just use AI on the assessment too?”
It's the first question everyone asks, and it's the right one. If measurement were as easy to fake as a résumé, we'd have moved the problem rather than solved it. The assessment is built so that outsourcing it produces a flagged profile rather than a strong one.
Behavioral items measure stable work patterns. There is no correct answer to research, and each construct is cross-checked against others for consistency.
Speed against accuracy. Honesty against comfort. A model can pick an answer, but the answer reveals a priority rather than a piece of knowledge.
The same question appears twice, reworded and re-ordered. Delegated or rehearsed profiles drift, and the drift is scored.
A social-desirability scale catches the candidate who is impossibly virtuous on every single dimension — the signature of an idealised self-portrait.
Straight-lining and impossible timing flag the profile red on the report and in the dashboard.
Scored for the generic, evidence-free register that generated text produces, and rewarded for specifics about a life a model has no access to.
A score is not a decision.
It would be easy to oversell this, so plainly: Hirovia does not tell you who to hire. It tells you which ten of a thousand applicants deserve your interview time, and hands you questions aimed at each one's weak spots. The hiring decision stays where it belongs — with you, combined with interviews, references and your own judgement about the person in the room.
Nor is any assessment perfect. Scores are one input among several, every report shows the components behind its number so a decision can be explained rather than asserted, and validity concerns are surfaced rather than silently applied. Used as the sole basis for rejecting people, any assessment — ours included — is being used wrongly.
What it does replace is the thing that genuinely stopped working: ranking human beings by how well a language model described them.
See what measurement actually returns.
A full sample report, or five of your own candidates free. No card, no demo, no sales call.