A first-hand story about losing a role to a candidate whose CV was a raw copy-paste of their conversation with a chatbot, why the keyword-scoring layer rewarded exactly that, and why being early is the only defence that scales.
A while back, I applied for a role at a major employer. Plenty of relevant experience on paper, the job description read like it had been written for me, and the listing still had weeks left before closure. I sent the application in, waited the polite three weeks, and got the polite no-thanks email back.
I have a friend who works at the company, so I asked. His answer was, of all things, that I'd left it too late.
Weeks left before closure, mind. Plenty of theoretical runway. But the company's applicant tracking system pre-sorts incoming CVs against the role's keyword profile and the recruiters work from that pre-sorted list almost exclusively. By the time my application landed, a viable shortlist had already been built from earlier applicants, and my CV had gone into the maybe-pile. Maybe-piles never come back.
Then he told me what happened when he sat down to interview one of the candidates the system had picked.
The candidate had submitted, as their CV, the complete unedited transcript of their conversation with ChatGPT. Question, answer, question, answer. The prompt was visible on the page. The model's responses were visible. The candidate's actual experience, if any, was not.
That CV got the interview. Mine, with whatever relevant experience I'd earned over a real career, didn't.
It's worth understanding why this works, because it explains a lot about how the current employment landscape feels.
Where an applicant tracking system runs a keyword or AI-scoring layer over inbound CVs, that layer is grading on density and terminology breadth. A real human CV has the right keywords spread across a finite number of bullet points, maybe 30 to 60 of them, written in compact resume English. A chatbot transcript on the same topic has hundreds of paragraphs of verbose explanation, every sub-concept named, every synonym surfaced, every adjacent skill referenced (and a few unrelated ones for good measure). To a scoring layer that measures overlap with the job description, it looks like the strongest match in the pile.
To be fair, not every applicant tracking system filters this way. The widely-repeated claim that ATSes silently reject every CV missing specific keywords is overstated, and we've written about that here. Workday, Greenhouse, Lever, Ashby and SmartRecruiters default to surfacing every application to a human recruiter, no auto-rejection involved. Mostly is the operative word though. Plenty of large employers run AI-scoring layers, knockout questions, or third-party screening services on top of their stack, and where those exist, the LLM-stuffed CV game absolutely works.
That's the bit that bothers me.
There is no signal in a pasted chatbot transcript that the candidate did anything more than type a prompt and copy the answer. They probably did not read the job description carefully. Probably did not write a tailored cover letter. Probably did not rehearse an answer to a single interview question. None of that matters at the sift, because the sift is grading text density, not effort. Recruiters at the other end of the pipeline then sit through interviews with people who pasted prompts at them, which can't be a great use of anyone's time.
The candidates losing to that, predictably, are the ones writing careful, accurate, hand-written CVs.
You cannot out-stuff an LLM with a hand-written CV. You shouldn't try. The defence is to be in the application pile so early that the scoring layer is largely irrelevant, because the recruiter is still picking from a small batch of date-ordered applications, before the AI-sifted shortlist gets locked in.
Apply on day one and you're applicant number seven. Apply on day twelve and you're applicant 240, sorted somewhere underneath a copy-pasted chatbot conversation. Recruiters close requisitions when they have enough viable candidates to interview, not when the queue is empty. How that funnel actually works is the same regardless of how clever the filtering layer at the top is.
This is also why the first-mover effect on application response rates is so robust across industries, and why the roles that fill before they hit LinkedIn are the ones worth chasing. The systems that decide who gets interviewed all work in your favour when you're early. Late, you're at the mercy of whichever filter the employer happened to bolt on, and increasingly that filter rewards machine-generated noise.
After the loss, I changed my routine. Specifically:
Our complete guide to applying early walks through the daily routine in more detail. Most of it is fifteen minutes, not three hours.
None of this means the keyword-stuffed transcript wins every time. Lots of recruiters, in lots of pipelines, would skim that CV for five seconds, recognise it as garbage, and bin it. Lots of ATSes would never have surfaced it as a top match in the first place. The story above is one role at one employer, told second-hand. I cannot prove the candidate had no qualifications either.
What I can say is that the structural conditions for this kind of gaming exist, and they're spreading. There is no version of the next few years where the AI-stuffed CV problem gets less common. There is, however, a version where you take the positional defence seriously and stop trying to beat the bots at their own game.
Be first. Write the real thing. Let the wall-of-text candidates fight it out in row 240 of the queue.