How do you screen candidates when every resume is AI-written?
Move the evaluation off the resume. LinkedIn recorded a 45 percent jump in applications in a year, roughly 11,000 per minute, and recruiters report that AI-polished resumes read identically. Interview-first screening tests something a template can't produce, which is how a candidate handles structured, role-specific questions with follow-ups. Anna interviews each applicant at the moment they apply, so the resume becomes context instead of the gate.
The flood, in numbers
The New York Times reported that applications on LinkedIn surged more than 45 percent in a year, to roughly 11,000 per minute, with generative AI driving the deluge. One HR consultant in the piece posted a remote role and hit 1,200 applications within days. Ashby's analysis of 109 million applications found applications per hire tripled from 2021 to 2024 and stayed above 300 through 2025. Greenhouse's benchmarks across 640 million applications show a 111 percent increase in applications per job between 2022 and 2025.
Volume is only half the problem. When candidates prompt an AI with the job description, the output is a resume containing every keyword the employer listed, and recruiters report the results read almost identically across hundreds of applicants.
Why keyword filters fail now
Keyword matching worked as a rough filter when resumes varied. It stops discriminating when every resume is generated against the posting it answers: the keywords are present by construction, so the filter passes the best prompters rather than the best candidates.
The failure cuts both ways. Harvard Business School's Hidden Workers research found 88 percent of employers acknowledge their screening software vets out qualified high-skills candidates, and 94 percent say the same for middle-skills roles. So resume-stage filters now admit polished fabrications while excluding real talent whose documents don't match the template. Tightening the filter makes the second problem worse without fixing the first.
Interview-first screening
The fix is to evaluate a layer the template can't fake cheaply: a structured conversation. Anna interviews every applicant the moment they apply, in a text conversation with role-specific questions and follow-ups that probe the answer just given. Knockout questions resolve pass/fail. Behavioral answers are scored 0 to 100 against the employer's rubric, with written reasoning per score, and candidates slot into a live ranked shortlist.
The resume still matters as context and supporting evidence, but it stops being the gate. Recruiters open the pipeline to ranked interview transcripts instead of 500 near-identical documents, which is the difference between reading claims and reading answers. That holds whether a role draws 50 applicants or 500. How the interview works end to end is on the How Anna Works page.
Common follow-up questions
Won't candidates just use AI in the interview too?
Some try, and a static form is easy to game. A conversation with adaptive follow-ups is harder, because the next question depends on the previous answer. More in our answer on cheating AI interviews.
Should we penalize candidates for using AI on their resume?
Employers increasingly treat AI-assisted writing as normal, and detection is unreliable. The practical move is to weight evidence the candidate produces live, in their own words, over the polish of the document.
Does this punish candidates with strong resumes?
No. Strong experience surfaces in interview answers, and resume evidence still supports scoring. What changes is that a strong resume alone no longer substitutes for demonstrating the experience it claims.
