A complaint that keeps surfacing from recruiters: every CV reads the same now, because they are all written by AI. The complaint is fair. The diagnosis is wrong, and the difference matters if you are the one applying.
The loop candidates are stuck in
Lay the steps out and the outcome is hard to avoid. You have to apply to dozens of roles, sometimes hundreds. Each CV has to be tailored to the posting or the screening software drops it before a person sees it. Tailoring a hundred and twenty CVs by hand is not humanly possible. So you use a tool. Then the recruiter on the other side opens their inbox and finds four hundred documents that sound like each other.
Every step in that chain is a reasonable response to the step before it. Nobody in the loop is behaving badly. That is what makes it a loop rather than a scandal.
Where the sameness actually comes from
It does not come from a machine doing the writing. It comes from what the machine is aiming at.
Almost every CV tool on the market optimises for one thing: keyword overlap with the job posting. It is easy to measure, it looks like progress, and it does move you up the ranking. But think about where that objective leads. If the target is maximum overlap with the ad, then the best possible output is the ad. Every CV drifts toward the same document, and that document is the posting itself.
Which explains the recruiter's experience precisely. Four hundred candidates with four hundred different careers submit near-identical CVs, not because they copied each other, but because they were all pulled toward the same point. The convergence is the objective working as designed.
The same automation, aimed somewhere else
Change the target and the behaviour changes with it. Optimise against the candidate's actual history, using the posting only to decide what to lead with and what to cut, and the outputs separate again, because the histories do. Two people applying to the same role should produce visibly different documents. If they do not, something upstream is broken.
This is also the only version that survives an interview. A CV written toward the ad claims whatever the ad asked for. A CV written from your history claims only what you can defend when someone asks a follow-up question, which is the moment that actually decides the outcome.
What this means for a gap you genuinely have
There will almost always be a requirement you do not meet. Optimising for overlap handles that by asserting it anyway, since the keyword needs to be present. That is how people end up in a room being asked to describe a methodology they have never used.
The alternative is to name the nearest real thing you have done and prepare the sentence you will say out loud. It scores lower. It also stops you wasting a week on a process you were never going to survive, which is worth more than the score.
The other half of the loop
It would be unfair to leave this at the candidates' feet. The same week a company asks for a CV written by a human, its first interview round is often conducted by an AI assistant. Both sides are drowning in the same volume problem and reaching for the same tool to cope with it. Neither side is being hypocritical, exactly. They are both automating the part that became impossible.
Where Job-Agent sits
This is the distinction the product is built on. Job-Agent reads the posting, compares it against what you have actually done, tells you honestly where you fit and where you do not, and rewrites your CV without claiming anything your career does not support. Several of the checks that enforce that last part are deterministic code rather than instructions to a model, because a model asked to fill a section will fill it from whatever text is nearest, and the nearest text is the job ad.
Here is what that looks like measured. A deliberately thin CV, two generic roles, no mention anywhere of Agile, Scrum, Docker, Jira or budgeting, scored against a posting demanding all of them. Before those checks existed the analysis came back at 90-95% with a recommendation to apply, citing sprint rituals and backlog prioritisation the CV never claimed. It had read the posting and written it back as the candidate's own experience. With the checks in place, identical inputs return 29-37% and a recommendation to reconsider, with every missing skill marked missing. The second number is far less flattering. It is also the one that saves you a week.
Same automation as everyone else. Different target. The model is Mistral's, European on both counts, trained by a European lab and served in Paris. Your CV never leaves Europe.