The professionals pulling ahead aren't working more. They're using Claude.

Our free guide will show you how to:

Transform your workflow with AI and stay ahead of the curve with this comprehensive guide to using Claude at work.

Get The Free Guide

You already know everyone is using AI to write resumes. You can feel it in the advice you get, which has quietly become a loop: use the tools, but not too obviously, and make sure you sound like a real person while also being perfectly optimized.

Recruiters are living the other side of that loop. In a report published in mid-October, the chief executive of Employ, which owns several of the systems your applications land in, described the problem in a way that should stop you cold. Candidates have gotten good at prompting. The resumes come back overly polished. They are so perfect, he said, that they are literally identical.

Here is where it gets more interesting than another "be more human" lecture. Researchers at the University of Maryland's Smith School, working with colleagues at the National University of Singapore and Ohio State, ran a controlled experiment on 2,245 real resumes written by actual job seekers before generative AI existed. They had language models rewrite those resumes, then had language models screen them. The finding: models systematically preferred resumes written by themselves, over human-written versions of the exact same candidate, at rates between 67 and 82 percent for the larger models, even after controlling for writing quality.

But the most useful thing in that paper is not the headline number. It is the study design. To run the experiment cleanly, the researchers had to pick one section of the resume to rewrite and hold everything else fixed. Which section they picked, and why, is a map of where this entire fight is actually happening. Let's read it.

The researchers replaced the executive summary at the top of each resume and left the work experience, skills, education, titles, and dates completely untouched. Their stated reason: those other sections "largely report verifiable facts," while the summary is "inherently subjective and stylistically flexible."

Read that again as a candidate. A team of researchers studying resume manipulation concluded that there is exactly one part of your resume where wording can swing the outcome, and it is the paragraph most people spend the most time agonizing over. Everything below it is a record. Records do not get better when you rephrase them.

So the summary is the one section where you are competing purely on style against a machine with style preferences you cannot see. That is a bad game. Cut yours to two lines, or cut it entirely, and move the real estate down the page.

Your first instinct on reading this research is probably the obvious one: find out which model the company screens with and write your resume in that model. Kill that instinct now.

The study's own numbers show why it fails. Model-versus-model preference was inconsistent and sometimes backwards. GPT-4o actually preferred DeepSeek's writing over its own by a wide margin in one comparison. LLaMA showed almost no preference in either direction. Meanwhile, no employer on earth publishes which model is inside their screening stack, and the models tested have already been replaced twice over.

The honest read of this research is defensive, not offensive. You cannot win the style game, so stop funding it. What you can do is reduce how much of your outcome depends on a coin flip you are not allowed to watch.

The sections the researchers held constant are the sections where you still have leverage, because facts survive paraphrase.

Go line by line through your work experience and ask whether a stranger could restate the line without losing anything. "Managed vendor relationships" survives rewriting intact because it contains no information. "Ran 14 vendor contracts worth $4.2M, consolidated to six, cut spend 23 percent" cannot be rewritten into something else, because the facts are the content. Numbers, named systems, headcount, dollar figures, cycle times, and the specific scope of what you owned are all immune to the thing this study measured.

One more that candidates skip constantly: if your employer is not a household name, add a six-word descriptor next to it. "Series B logistics software, 80 people." That line costs you nothing and does more work than any summary you will ever write.

Get $5 off Lovable

_{^{Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.}}

This is the part of the study that should genuinely change your behavior. The researchers brought in human annotators to judge which resume summary was actually better, blind to the source. In some cases, the humans judged the human-written summary to be clearer, more coherent, and more effective. The model picked its own version anyway.

Sit with that. A person wrote a demonstrably better paragraph and the screener still passed them over, not because the writing was worse, but because it did not sound like the machine. Hours spent perfecting that paragraph bought nothing.

That is your argument for reallocating your week. If polish has an unreliable payoff, spend the time on the things with a reliable one: a referral, a direct note to the hiring manager, a portfolio link, or one more fact added to a bullet that currently has none.

The effect was not uniform across jobs, and the pattern matters for how seriously you take all of this.

When the researchers simulated real shortlisting, candidates whose resumes matched the screening model were 23 to 60 percent more likely to make the cut. The damage was worst in business-facing occupations, with sales and accounting at the top. It was mildest in fields like agriculture, the arts, and automotive.

The likely reason is that business resumes are written almost entirely in the register models are best at imitating: polished, confident, achievement-framed prose. If you are in sales, finance, accounting, marketing, or operations, assume your written positioning is doing less for you than you think and your verifiable record is doing more.

Open your resume and do this in one sitting.

Delete your summary. Read the page without it and notice that you lost nothing. Now take the three strongest claims you buried in that paragraph and relocate them into your experience section as facts, each with a number or a name attached to it.

Then, for every bullet on the page, apply one test: could a model have written this line about someone who never did the work? If yes, it is decoration. Rewrite it or delete it.

If you want to keep a top section, make it a scope line rather than a self-description. Something like this:

No adjectives. Nothing to rephrase. Nothing a screener can prefer a shinier version of, because there is no shinier version.

Everyone's words now sound the same. Your record does not.

Cool air, colorful leaves or a whole new city skyline: There's still time to book your fall escape with Alaska Airlines. Watch the aspens turn golden in Denver, catch a set on Broadway in Nashville, or trade sweaters for sunshine in Palm Springs.

Book Now

_{^{Subject to availability. Terms and conditions apply.}}