Rejected in an Hour: How AI Résumé Screeners Work, and How to Get Seen by a Human

If you've ever applied for a job and gotten a rejection email before you finished making dinner, you're not imagining things.

A lot of applications are read first by software, not people. Sometimes that software just sorts. Sometimes it scores. And sometimes it decides you're out before anyone ever sees your name.

Let's open the hood, talk about the lawsuit that's testing the rules, and then get practical.

What happens after you click "Submit"

Most mid-size and large employers use an applicant tracking system (ATS). Think of it as the database and workflow tool for hiring. Here's the typical pipeline:

1. Parsing. The system extracts text from your résumé and tries to fill structured fields: name, job titles, employers, dates, skills, education. This is where fancy formatting hurts you. Two-column layouts, text inside images or graphics, and tables can scramble the parse, so your "Senior Analyst" title ends up in the wrong field, or vanishes.

2. Knockout questions. "Are you authorized to work in the U.S.?" "Do you have a valid driver's license?" "Are you willing to work weekends?" Answer one wrong, and many systems reject you automatically. These are rules the employer sets.

3. Matching and ranking. This is where AI comes in. Some systems compare your parsed résumé to the job description and produce a match score or ranking. Older tools did simple keyword matching. Newer ones use machine learning models that look for patterns associated with "successful" candidates.

4. Human review. A recruiter looks at the top of the list. If there are 400 applicants and they have time for 40, your rank matters a lot.

That third step is where the big questions live.

Why AI ranking can go wrong

Here's the technical problem in plain language. If a model learns what a "good candidate" looks like from a company's past hires, it can learn the company's past patterns, including unfair ones. If past hires skewed young, the model may learn that recent graduation dates or short work histories are "good signals." No one has to program bias in for it to show up.

Some signals act as proxies. A gap in employment might correlate with medical leave, caregiving or disability. A zip code might correlate with race. A model doesn't need to know your age to learn patterns that track age.

The lawsuit testing the rules

The most closely watched case on this is Mobley v. Workday, in federal court in California. Derek Mobley, a Black man over 40, says he applied for more than 100 jobs at companies using Workday's hiring tools and was rejected from all of them, often within hours, according to the American Bar Association. The plaintiffs allege Workday's AI screening tools disproportionately exclude Black, female, older and disabled applicants.

Workday denies wrongdoing. A spokesperson told Reuters its AI tools "only look at job qualifications—not protected traits like race, age, or disability," and the company says its software is highly customizable by each employer (Reuters).

Where things stand, per legal summaries from Epstein Becker Green and Warden AI's case tracker:

• In May 2025, the court preliminarily certified a nationwide collective action for applicants over 40 on the age claim.
• In March 2026, the judge rejected Workday's argument that federal age discrimination law's disparate impact protections don't cover job applicants.
• In June 2026, the judge allowed California civil rights claims and a disability claim to proceed. That disability claim is based on allegations that the tools treat employment gaps tied to medical leave as a proxy for disability.
• A hearing on whether to certify broader class subclasses is set for March 9, 2027.

Important: the court has not found that Workday's tools discriminated against anyone. These are allegations being tested. But the legal theory that's survived so far matters: when a vendor's software does a hiring function, like screening or ranking, the vendor can potentially be treated as the employer's agent under anti-discrimination law.

What you can actually do as a job seeker

You can't see the algorithm. But you can stack the odds. Here's what I tell people in my classes.

Make your résumé machine-readable

• Use a simple, single-column layout. Standard section headings: Experience, Education, Skills.
• No text in images, text boxes or headers/footers. Parsers often skip them. Put your contact info in the body.
• Save as .docx or a text-based PDF, unless the posting says otherwise. A PDF made by "printing" a scan is just a picture.
• Spell out and abbreviate. Write "Structured Query Language (SQL)" once. Some systems match one form but not the other.
• Test it. Copy all the text from your PDF and paste it into a plain text editor. If it comes out jumbled, a parser will see it jumbled too.

Match the language honestly

Read the job description and mirror its words where they're true for you. If they say "stakeholder management" and you wrote "worked with partners," use their phrase. Don't stuff keywords you can't back up. A human will eventually read it, and so will an interviewer.

Handle gaps directly

If you have an employment gap, a single honest line can help a human reader: "Family caregiving, 2023–2024" or "Career break: completed data analytics certificate." You don't owe anyone your medical history. But a gap with no context leaves the story to be filled in by someone else, or by software.

Don't rely on the portal alone

This is the big one. The best way past an algorithm is a person.

• Find the hiring manager or a team member on LinkedIn. Send a short, specific note: who you are, the role, one sentence on why you fit.
• Ask for referrals. Many companies flag referred candidates for human review.
• Go to meetups, career fairs and community events where recruiters actually show up.

Know your rights

If you believe you were screened out unfairly because of a protected characteristic, you can contact the EEOC or, in Washington, the Washington State Human Rights Commission. Keep records: the posting, the date you applied, and any rejection emails with timestamps.

For employers: ask your vendor hard questions

If you use screening software, you can't outsource your responsibility. Ask your vendor how it tests for adverse impact by race, sex, age and disability, how often, and whether you can see the results. Make sure a human reviews before anyone is rejected on anything beyond objective, job-related knockouts. And watch the rules coming: California's No Robo Bosses Act and Connecticut's notice requirements, which I covered in Your Boss's Algorithm Now Needs a Human, and my 12 questions to ask before you buy any AI tool.

You're more than a match score

If the rejections are piling up, please hear this: an automated "no" is often a statement about a filter, not about you. Keep learning, keep connecting with real people, and keep applying. If you're working toward a data career, my roadmap is in How to Break Into Data Without a Computer Science Degree.

Pour me another: What's the fastest rejection you've ever gotten? Tell me through the contact form.

Sip slow, stay curious.D