Why do AI-written CVs get rejected?

Because a general-purpose model writing a CV has almost nothing true to work from. Given a job description and a thin prompt it produces fluent, generic text that describes a plausible candidate rather than a real one — no specific projects, no numbers that came from anywhere, and the same phrasing thousands of other applicants submitted that week. The problem is not that AI wrote it. It is that the AI was not grounded in anything.

What does a reader actually notice?

Specificity, or its absence. A real career contains odd, particular details — the constraint that made a project hard, the decision that turned out to be wrong, the number that was surprising. Generated text has none of these, because they were never in the prompt.

Consistency is the other tell. When each document is generated separately, the CV, the cover letter and the profile drift: different dates, different framing, different claims about the same role.

Is the answer to stop using AI?

No — the answer is to change what it is working from. A model grounded in a structured, verified record of your actual work writes about that work. The same tool that produces generic text from a job description produces specific text from a Professional DNA, because specificity was available to it.

That is also what keeps the outputs consistent: they are all generated from one record rather than from each other.

What about applicant tracking systems?

They reward the same thing for a different reason. A tracking system matches structure and keywords against a role, and a document assembled from a structured record maps onto it cleanly, because the underlying facts are already fields rather than prose.