I’ve been trying to get interviews for new jobs for over a year now, and just recently I finally got a couple of pings on my resume where the companies wanted me to talk to their talent acquisition team. I fully admit I use AI (Claude, for preference) to generate resumes that will get past ATSes and into the hands of humans. If I didn’t, I’d be left behind, because literally everyone else does the same thing.

For the second interview I had, I asked Claude to help me prepare. I gave it the job description and the resume it generated for me, and here is what it said:

Interview’s today, so let me skip the pep talk and go straight to where you’ll get hurt.

Your resume is a mirror of the job description, and that’s a liability, not just an asset.

Compare line by line. The JD says “Manage the [x], providing line management, mentoring and engagement.” Your bullet says “Manage [x] day-to-day, providing line management, mentoring, and engagement.” This is true of nearly every bullet. It’ll sail past the recruiter, but a competent hiring manager reads a resume that echoes their own posting and does one of two things: assumes you tailored it (fine) or starts pressure-testing whether there’s substance behind the matched keywords. Assume the latter. Every claim on that page is now a question they can ask, and if the answer is thin, the mirroring makes it look worse because it reads as coached rather than lived. Your job today is to have a concrete, specific, ideally-numbered story behind each claim they poke at.

Okay, fine, be prepared. I can be prepared. I’m always prepared. I just wanted to make sure I didn’t miss anything. But this is a lot bigger than just “making sure I didn’t miss anything”. This is Claude generating a resume for me that will end up getting me in trouble.

Tell me something, Claude: if these were going to be issues, then why didn’t you call them out when you wrote the resume? If I didn’t have measurable metrics, why didn’t you warn me in advance? You just gave me a vaguely positive answer when I asked if the JD matched my qualifications and then output something I could use.

So I called it out. Its response, in part:

Now the argument you’re actually making, which is the part worth spending words on. You’re treating “liability” and “asset” as if I have to pick one, and catching me in a contradiction. There isn’t one. A tailored resume and the interview it earns you are two different jobs. The resume’s job is to get you into the room, and mirroring the JD is the correct way to do that: it clears keyword filters and signals fit to a recruiter skimming for thirty seconds. A resume that didn’t echo the posting would be a worse asset and would still leave you owing substance in the interview. So mirroring dominates. It’s the right call whether I made it or you did.

What I called a “liability” isn’t a defect in the document. It’s the obligation that every well-tailored resume creates: once the page claims you did X, the interview is where you prove X. That obligation exists no matter how the resume is written. Mirroring just makes the claims land closer together, so a sharp interviewer can probe more of them faster. That’s not me sabotaging you. That’s the normal cost of the thing working.

I will say this: I’ve put some language into my Claude preferences that is intended to weed out the mealy-mouthed overly-positive crap most AI spits out and instead be more honest and blunt, so Claude certainly did that this time around. I’m just frustrated, because I’m going to go into this interview without the metrics that they’re going to want, so either I make some up or I talk around them. The specific metrics I include are around cost savings — policies and systems I put in place saved the company X amount of dollars.

Oh, and one other thing it called out: why did I put my freelance job so far down on the page, instead of in date order? Well, guess what, Claude: you did that when you generated my resume. Sure, I should’ve caught it and fixed it, but I’m applying to dozens of jobs each day; mistakes are going to be made. I guess I’ll have to update the language I use when I ask it to generate resumes to ensure that my previous roles are in date order.

This whole thing is a mess. And the worst part is: the job I’m talking about in this post is one that I actually want, that I would be great at. But if I don’t have measurable metrics to share, then I’m sunk. Here’s hoping I can tread water.