How AI Is Reshaping the Hiring Process for Job Seekers

In October 2024, researchers at the University of Washington ran over three million comparisons between AI resume-screening tools and real job descriptions. The result was hard to sit with: the systems preferred white-associated names 85% of the time, and never ranked a Black male-associated name above a white male one. That study was presented at the AAAI/ACM Conference on AI, Ethics, and Society. And yet the companies deploying those same tools keep scaling them up.

That’s the actual situation right now. AI is all over hiring. Whether that’s good for candidates is genuinely unclear, and anyone who tells you otherwise is selling something.

How widespread AI screening actually is

According to SHRM’s 2025 Talent Trends report, 51% of organizations now use AI to support recruiting, up from a much smaller share just two years prior. The most common applications are writing job descriptions (66%) and screening resumes (44%). Candidate searches, job posting customization, and applicant communication trail behind those.

Among the HR professionals who do use AI in recruiting, 89% report it saves time or increases efficiency. That number is real. It’s also the whole problem: when something makes an employer’s life easier, they adopt it fast, and the downstream consequences for candidates take longer to surface.

The practical upshot is that your resume probably hits an automated filter before any human sees it. At large employers, that’s been true for years. What changed is that the filters got smarter, and in some ways, harder to game.

What AI screening actually does to your resume

Older applicant tracking systems ran keyword matches. Submit a resume without the exact phrase “machine learning engineer,” get filtered out. Simple, visible, gameable.

Newer systems are doing something closer to semantic matching. They compare the overall shape of your experience against the job description and assign a relevance score. This is harder to game with keyword stuffing, which is actually a mild improvement for candidates with genuine experience. The problem is that the semantic models absorb biases from training data at scale, as the UW study showed.

The short version of what these systems measure: how closely your history maps to a “successful” past hire. If those past hires skew demographically in any direction, the model learns that skew.

What helps in practice is being specific. Vague descriptions like “worked on backend systems” score lower than “built a Kafka-based event pipeline handling 40k events/sec.” Not because the AI is cleverer than a recruiter, but because specificity overlaps with what the model was trained to surface. This is an imperfect proxy for actual quality, but it’s how the system works right now.

The video interview problem

Some employers add an AI-analyzed video component after the resume filter. HireVue is the most common vendor. The pitch is that these tools reduce interviewer scheduling overhead and detect patterns in speech and facial expression that correlate with job performance.

The legal record is worth knowing. In March 2025, the ACLU of Colorado filed a complaint alleging HireVue’s platform discriminated against a deaf and Indigenous applicant. CVS settled a similar lawsuit in 2024 after applicants alleged the company used AI video tools to assign “employability scores” based on facial expressions. Workday faces a conditionally certified class action (as of May 2025) covering millions of applicants over 40 who were allegedly filtered out by its AI screening tools.

I don’t know whether these lawsuits will change employer behavior in the short term, or whether regulatory enforcement will catch up to adoption rates. The honest answer is that nobody does. What I’d say to candidates: if a company is using AI video analysis, practice your delivery on camera anyway, because the communication skills it rewards (clear pacing, direct answers, minimal filler words) are skills that will help you in human interviews too. Don’t let the ethics debate stop you from preparing.

How to actually use AI on your side

Candidates who struggle most with AI screening tend to do one of two things: ignore it entirely, or over-optimize to the point that their resume reads like a keyword list. Neither works.

What works better, based on what the research and legal cases actually show:

• Write to specificity, not keywords. Quantify scope and impact wherever you can. “Led a team” is weak. “Led a team of 6 engineers shipping a feature used by 200k daily active users” is not.
• Mirror the job description’s language where it fits naturally, not everywhere.
• Don’t lean on AI to generate your resume from scratch. Recruiters who screen post-AI-filter are increasingly good at recognizing AI-generated prose and it reads as a signal that you can’t communicate your own experience.
• Run an AI screening interview before the real thing. Mock interviews that simulate structured assessments let you see where your phrasing gets vague before it costs you.
• Check your resume formatting. Some ATS parsers still fail on tables, headers in text boxes, and two-column layouts. Plain, single-column resumes with standard section names parse cleaner.

What AI still doesn’t do well

There’s a version of this conversation that ends with “humans still make the final call, so build relationships.” That’s true but it understates how much of the funnel AI now controls before a human is involved.

Still, current AI hiring tools are genuinely bad at a few things. They struggle with non-linear career paths. Career changers, people with gaps, people whose experience spans disciplines, these profiles often score badly not because they’re weak candidates but because the model’s baseline is a clean linear trajectory through one function. They’re also bad at evaluating portfolios, open-source contributions, and anything that doesn’t fit a text field.

If your background is non-linear, you’re probably better off finding companies where your resume reaches a human before a screen. That means referrals, smaller companies, and applications through employees. Not revolutionary advice. Just more true than it used to be.

Practicing for AI-adjacent interviews

The final stage of AI involvement, for most tech hiring, is the actual interview assessment: automated coding tests, take-home projects, and live structured interviews that sometimes use AI scoring. For these, the preparation advice is more concrete.

Structured practice against the specific question formats helps. Resources like behavioral interview question frameworks and AI-powered mock interviews let you rehearse under realistic conditions, which is different from reading about what to say. Reading and doing are not the same thing. I think most candidates underestimate how much of interview performance is physical, not just knowledge-based: the pace, the pauses, the way you organize a response under pressure. Those only improve with repetition.

Separately: if you’re going through an AI coding screen, the landscape of AI coding interviews in 2026 has shifted enough that it’s worth understanding what these platforms actually measure. It’s not just whether your code runs.

The broader hiring market is using more AI, not less. Some of that is good for candidates (faster feedback loops, less scheduling friction) and some of it is a serious problem (documented bias, opaque scoring). The sensible response is to understand how the tools work, prepare for the parts you can prepare for, and not assume the system is neutral just because it’s automated.