The AI wrote your resume. Another AI is reading it. Now what?

There's a strange loop running underneath ecommerce hiring right now. A candidate opens ChatGPT, pastes in a job description, and gets back a clean, keyword-stuffed résumé in thirty seconds. That résumé lands in an applicant tracking system, where a second model scores it, ranks it, and decides whether a human ever sees it. Machine writes, machine reads. The people on either end are mostly along for the ride.

This isn't hypothetical. LinkedIn's own research from January 2026 found that 81% of people have used or plan to use AI in their job search, while 93% of recruiters plan to lean on it harder this year. The pipeline is software on both sides.

And it gets weirder. A 2025 study out of the University of Maryland, NUS, and Ohio State, titled AI Self-preferencing in Algorithmic Hiring, found that when an LLM evaluates résumés, it systematically prefers résumés that were written by an LLM. Worse: if the company's screener happens to run the same model you used to write yours, your odds of being shortlisted jump somewhere between 23% and 60%. The researchers called it a novel form of bias. You could also just call it the machine recognizing its own handwriting.

So the obvious move is to fully automate your side too, right? Let the bot write it, pray it matches the bot reading it, roll the dice.

Don't.

The real filter isn't "AI vs. human." It's "signal vs. noise."

The number everyone quotes, that ATS software auto-rejects 75% of résumés, is a myth. It traces back to a 2012 sales pitch from a startup that went out of business in 2013. There's no primary source. Repeating it just trains people to be paranoid about the wrong thing.

Here's what's actually worth worrying about: roughly half of hiring managers say they'll bin a résumé the moment they clock it as AI-generated (Resume.io put it at 49%). They're not catching you with some forensic watermark. They're catching emptiness: the generic phrasing, the "results-driven professional with a proven track record" filler that says nothing at all. Recruiters have started seeing five different candidates submit the identical bullet point for the same role. A Boston University professor described the AI-written cover letters landing in his inbox as dead giveaways: they're "all five paragraphs long and the language is very similar." That's the failure mode, and it's a content problem, not a detection problem.

For an ecommerce engineer, this is good news. Generic is your competition's weakness, and specificity is your edge, because the work you actually do is intensely specific.

What actually gets an ecommerce engineer through

Show the stack, not just the title. Plenty of roles on this board are literally titled "Shopify Developer," but the title is the wrapper, not the match. Look at what each of those listings actually spells out underneath: Liquid, Checkout Extensibility, Shopify Functions, the Storefront API, Hydrogen. That keyword soup is what the parser scores you against, and a résumé that says "Shopify developer" and stops there matches almost none of it. If you've shipped with it, name it. The match is literal, and vague language is invisible to both the parser and the human reading after it.

Quantify in commerce terms. Every engineer on earth writes "improved performance." You happen to work in the one field where performance has a dollar sign bolted to it. Compare these two bullets:

Before: Experienced Shopify developer with a proven track record of building high-performing ecommerce solutions that drive business results.

After: Rebuilt a Shopify Plus checkout on Checkout Extensibility + Functions, killed a legacy app dependency, dropped checkout LCP from 3.8s to 1.9s, and lifted mobile conversion ~11% the following quarter.

The second one couldn't have been written by a bot that's never seen your repo, and that's the whole point. LCP, conversion rate, GMV handled, peak Black Friday traffic survived, checkout completion rate: these are numbers a generic model literally cannot invent on your behalf.

Build one public artifact. A WooCommerce plugin on GitHub. A headless Medusa storefront demo. A repo with a Shopify Functions discount you wrote. The engineers doing the hiring at places like Automattic, Lab Digital, or DigitalSuits are technical, and a link to working code beats three paragraphs of adjectives every time. It's also un-fakeable proof in a market drowning in fakeable text.

Apply direct. A big chunk of the noise in modern hiring is recruiter middlemen blasting the same résumé at forty openings. Going straight to a company's careers page, which is the only kind of listing we index here, means your application isn't buried inside someone's bulk-submit batch. Fewer layers, less noise, faster signal.

Use AI the way a senior engineer uses a linter

None of this means swearing off AI. The consensus among the people who actually study this is boring and correct: use it to edit, not to ghost-write. Let it tighten your phrasing, clean up the formatting, catch the keyword you forgot. Then put your fingerprints all over it: the specific numbers, the specific stack, the specific thing you fixed at 2 a.m. before a launch.

The model can match the screener. It can't be you. In a hiring pipeline where both ends are increasingly automated, the single most valuable thing on your résumé is the part no AI could have generated: the actual work you did.

That's the part that gets you the interview. Everything else is just getting past the bouncer.