We Built a Fake Resume to Break Our Own Tool. It Worked.

We built a resume with every word a job asked for and nothing real behind it. Then we ran it through our own tool. It came back Qualified, and it ranked above a real candidate who had actually done the work.

That was not a demo. That was a bug in our product, and we are going to walk you through the whole thing.

The setup

Synthalyst Match is our checker. You paste a resume and a job posting, and it gives you a straight read: Qualified, Stretch, or Not Qualified for that specific job, with the reason.

The whole promise rests on one claim: it reads deeper than keywords. That claim is easy to make and hard to check, and a tool that gets it wrong fails quietly, because nobody sees the resumes it judged badly. We went looking for ours by trying to break the tool on purpose.

So we built a fake.

The fake was a project manager with nothing behind him. The right title. Every phrase lifted straight from the job posting. Agile, stakeholder management, cross-functional, delivery. But every single line described a task he had been handed, never any result that came of it. “Responsible for managing project timelines.” “Involved in stakeholder communication.” It was a resume made entirely of vocabulary, with no evidence underneath any of it.

Then we built a second resume. A real candidate this time. Actual numbers, actual outcomes, actual results. But his job title sounded more junior than the role he was applying for, and the best part of his experience sat halfway down the second page.

We ran both.

The fake won

The fake came back Qualified. The real candidate came back Stretch. Our own tool put the empty resume above the person who had genuinely done the work.

Just to be sure the result we got was real and not an error, we rebuilt the fake resume from scratch and ran it again. New name, new company, different verbs, same nothing behind it. Qualified again.

That ruled out a fluke. It was not one badly worded resume getting a lucky pass. The tool was rewarding a pattern: keywords with nothing behind them were counting as proof.

To be clear, having the job’s words on your resume is a good thing. You need them, or a search will never surface you at all. What they cannot do is stand in for evidence that you have actually done the work. Our tool was letting them, and that is the exact thing we built it not to do.

Why it happened

The tool was treating two very different questions as if they were the same question: can a keyword search find you, and would a recruiter judge you qualified.

Question one is mechanical. Will a keyword search find you? A recruiter types a term into the search box of an ATS (the software that companies use to store and search resumes). Either that exact word is on your page or it is not. There is no judgment here. It is a yes or no about whether the word is there.

Question two is a judgment. Would a recruiter actually credit you with the skill? That is not about whether the word appears. It is about whether the resume shows evidence you can do the thing. A recruiter reads for proof, not for matching words.

Our fake beat both questions, but it only earned one of them. Every keyword was on the page, so a search would find him every time. That much was real. The second question he got for free: because the words were all there, the tool leaned toward crediting the skills, even though the resume proved none of them. The bug was that the tool confused findability with qualification.

The fix

We stopped letting the word check feed the judgment. The two questions now come back as two separate answers.

The word check runs first, because a recruiter has to find you before anyone reads you. If the job asks for “cross-functional collaboration” and your resume says “partnered across teams,” it reports that the exact phrase is missing, so a keyword search might not surface you. It says nothing about whether you can do the work, because it cannot know that.

The verdict is decided second, on evidence alone. Results, outcomes, proof. Keywords do not buy their way in.

What the tool does now

We reran everything.

The keyword-stuffed fake now comes back Not Qualified. The empty resume stopped getting away with it.

The real candidate did not move. He is still a Stretch, exactly where he was. His resume genuinely does bury the good part under a title that does not match, and Stretch is a fair read of that. We want to be precise here, because it matters: we did not put a thumb on the scale to promote the honest guy. Nothing got more generous. The fake just stopped winning. That is the fix being surgical instead of an overcorrection.

This matters more now than it used to. Hiring runs on volume. Recruiters lean on search to cut a pile down before anyone reads it closely, which makes being findable close to survival. But the advice that usually follows is to stuff in more of the job’s words whether or not you did the work, and that is what fills the pile with resumes that say everything and prove nothing. Everyone loses there. Honest candidates learn to write like the fake one, and resumes get harder to trust across the board.

Keeping the two questions apart is how you get out of that trade. The words get you surfaced in the search. The evidence is what puts you in the yes pile once someone reads you. You do not have to pretend to be found.

To be clear about what this does and does not promise: the tool tells you whether a recruiter would consider you qualified, and how to get found. It cannot promise you an interview. That depends on the size of the pile, the competition, and a human being’s choices, none of which live inside the tool.

Try to break it

We are telling you this because a tool that cannot tell a stuffed resume from an honest one is not a fix for hiring. It is one more thing that rewards whoever games it best. The only way to know which kind you are holding is to attack it and say out loud what falls out.

So attack ours. It is free, no account, about ten seconds: synthalyst.com/verdict.

There are two ways to test it, and they are different tests.

Build the emptiest, most keyword-perfect resume you can invent and see if it still gets through. If it does, we have the same bug back, and we want to know.

Then run your own real resume, and read the reason it gives you. If that reason rests on something that is not true about your experience, that is a miss, and we want to hear about it publicly.

We would much rather hear it from you than not know.