A CV was never a perfect signal, but it used to carry real information: how someone described their own work said something about how they thought about it, and the variance in writing quality across candidates was itself a weak but genuine data point. AI writing tools have compressed that variance sharply. A strong candidate and a mediocre one can now both produce an equally polished, equally well-structured CV, because the polish is coming from the tool rather than the person.
That's not a complaint about candidates using available tools to present themselves well. It's a description of what's happened to the CV as a screening instrument. When the writing quality stops correlating with the underlying capability, screening decisions that lean on writing quality as a proxy start making worse calls, more confidently, because the CVs all look equally credible.
What still carries signal
The parts of a CV that are hardest to make equally impressive regardless of underlying substance are the specific, checkable claims: what was actually built, what scale it operated at, what the person's role was on a team versus what the team collectively did. Generic competency language ("strong communicator," "proven leader") was always weak signal and is now essentially noise. Specific, falsifiable claims remain useful precisely because they're the part a polishing tool can't manufacture without the recruiter being able to ask a follow-up question that exposes whether they're true.
Where the screening effort has to move
The practical adjustment is to weight the CV itself less and the structured follow-up more: a short screening conversation that asks a candidate to go one layer deeper on a specific claim from their CV, which a well-polished but thin CV can't survive as well as a genuinely substantive one. That's more recruiter time per candidate, not less, which cuts against the instinct to screen faster now that AI tools exist. The tools didn't make screening faster to do well. They made the old fast method less reliable.