Job applications have quietly become one of the strangest places AI has taken over. Candidates use it to rewrite resumes, mirror the wording in job descriptions and produce tailored cover letters. Employers then use screening software and AI to work through the growing pile of increasingly polished applications and decide which people are worth speaking to. We have somehow built a system where one machine helps a person convince another machine that the person is a person.
The strange part is that nobody is behaving irrationally. Candidates need to get through automated screening, so they are rewarded for using the right keywords and describing their experience in language the system recognises. Employers receive more applications than they can reasonably assess by hand, so they introduce stronger filters. Candidates then optimise harder for those filters, which makes applications more similar and gives employers even less useful information to work with. Everyone is solving the immediate problem in front of them, while collectively making the whole process worse.
The cover letter might be the clearest example. Companies still ask why someone wants the role, even though everyone knows an AI can produce a competent answer in seconds. The new test is not whether the letter is good, but whether it sounds sufficiently unlike something written by AI. Add a personal observation. Make it less polished. Remove the em dash. Introduce the occasional awkward phrase so a human definitely appears to live there. We are now deliberately adding imperfections to machine-generated writing to prove it was not machine-generated. Fantastic system. No notes.
I do like applications that ask one or two questions in a text box, even when they are fairly standard ones such as why you want the job or why you want to work for that particular company. They at least make the candidate engage with the role instead of uploading the same documents and moving on, and a vague or generic answer is easier to spot when it sits underneath a specific question. It probably filters out some of the lowest-effort applications, but it is not foolproof. AI can answer those questions too, often quite convincingly, especially when the job ad and company website already contain everything needed to manufacture a respectable response. The better questions ask for an opinion, a trade-off or a reaction to something real, because they reveal how someone thinks rather than how well they can assemble a professional-sounding paragraph.
The useful questions are still fairly simple. Can this person solve the problems we have? Have they done anything similar before? Can they explain their thinking without turning it into corporate soup? Would I want to work with them? You would probably learn more by showing someone a real problem and asking what they notice than by making them write another paragraph about how excited they are to join a collaborative, fast-paced environment.
AI did not create the problem, but it has made the absurdity much easier to see. The application process was already built around vague job ads, rehearsed answers and documents written for screening systems. Now the machinery is simply talking to itself faster.