Every Application Is AI-Written Now. Here Is How to Screen Anyway

Somewhere in the last two years, the balance tipped: the typical application you receive was written, or at least heavily polished, by AI. The cover letter is fluent, the CV is keyword-perfect, and none of it tells you anything - because the same tools produced your other two hundred applications the same evening.

This is the loudest complaint in recruiting right now, and most of the proposed cures are worse than the disease. AI detectors misfire constantly and punish non-native speakers hardest. Honesty pledges are unenforceable. Banning AI penalises exactly the conscientious candidates who would obey. The productive question is different: what signals survive in a world where written self-presentation is free?

What exactly broke?

• The CV as effort signal. A tidy, tailored CV used to prove somebody cared about your role. Now it proves they have a browser.
• The cover letter as writing sample. It is now a prompt output, indistinguishable across candidates.
• Volume as interest. Auto-apply tools submit to hundreds of postings; an application no longer implies the person even read your ad.
• Keyword screening. If your filter looks for words, AI puts the words in. The arms race is over and the filters lost.

Which signals still work?

The ones generated live, under constraint, in the candidate's own head.

1. Specific, situational questions. "A guest insists they booked a sea view; the system disagrees - what do you do before calling a manager?" A generic tool produces a generic answer; the experienced candidate produces a telling one. Question design is covered in candidate screening questions.
2. Time as a signal. How long an answer took says things the text cannot. Confident domain knowledge moves at a different tempo than paste-and-edit. Qwiza records per-question timing and folds it into the score.
3. Consistency across questions. Fabricated competence drifts; real experience stays coherent when the same theme returns from a different angle.
4. Constrained formats. Choosing between four defensible trade-offs cannot be outsourced meaningfully - the choice itself is the data.
5. The interview, arriving earlier. When written signals die, conversation weight rises. Structured scoring gets the right people into that conversation faster.

Should I try to detect AI use?

No. Detection tools produce false accusations at rates no fair process can absorb, and they systematically flag fluent non-native writers. Worse, detection answers the wrong question: you do not actually care whether a tool touched the text - you care whether the person can do the job. Measure that directly and the detection problem evaporates.

There is also a fairness upside worth naming: when everyone's written materials look equally good, the playing field tilts back toward substance. A structured quiz scored on predefined criteria - the approach in blind screening - is fairer than the CV pile ever was.

What about the volume?

Structured screening absorbs it. Two hundred auto-applications into an inbox is a crisis; two hundred into a scored quiz is a Tuesday - the mismatched ones score low on the availability and situational questions they auto-answered, and your attention goes to the top of a ranked list. Volume stops being your problem and becomes your funnel.

The takeaway

Stop reading documents AI writes and start collecting signals AI cannot fake: live reasoning, tempo, consistency, constrained choices. The employers who adapt screen better than they did before the flood - because the old signals were weaker than everyone pretended.

This is precisely what Qwiza builds: a two-minute structured quiz in the ad itself, scored on answers and timing against criteria you set, with every answer attached so a human makes the call.