A reader forwarded me a rejection email last month with one line that stuck with me: "After careful review by our recruiting team, including AI-assisted candidate evaluation, we have decided to move forward with other applicants." She wanted to know if this was some different kind of ATS than the one I've written about before. It is, and it isn't. I spent the last few weeks pulling apart vendor documentation, court filings, and the newest survey data to figure out exactly what has changed since I first wrote about how an ATS works, and the honest answer is that two different systems are now doing two different jobs, and almost nobody explains the difference clearly.
The two systems that get lumped together as "the ATS"
"Applicant Tracking System" traditionally refers to platforms like Taleo, iCIMS, Greenhouse, and Workday's core recruiting module. As I've written before, these systems mostly parse, store, and rank resumes by keyword overlap with the job description rather than issuing an automatic rejection. That description still holds up. What's changed is that many of the same employers now run a second system on top of, or instead of, that keyword ranking: an AI screening layer built on a large language model.
Platforms like Eightfold, HireVue, Beamery, and Phenom People are the ones doing this second job. Eightfold, for instance, [says its screening model is trained on more than 1.5 billion career profiles](https://eightfold.ai/use-case/candidate-screening/) and matches candidates on inferred skills and career trajectory rather than keyword overlap alone. According to [SHRM's 2025 Talent Trends report](https://www.shrm.org/topics-tools/research/2025-talent-trends/ai-in-hr), 43% of organizations now use AI somewhere in their HR function, up from 26% the year before, and of the organizations using AI specifically in recruiting, 44% use it to screen resumes. A [ResumeBuilder.com survey of 948 business leaders](https://www.resumebuilder.com/7-in-10-companies-will-use-ai-in-the-hiring-process-in-2025-despite-most-saying-its-biased/), conducted in October 2024, found 83% planned to use AI to review resumes in 2025. Put together: if you applied to a large company any time in the past year, there's a good chance your resume was read by both systems, not one.
What actually happens when an LLM reads your resume
The mechanical difference matters more than the marketing language around it. A keyword-matching ATS is essentially running a search: does the string "SQL" appear in this document, and how many times. A large language model reading the same resume is doing something closer to comprehension. It can infer that "owned a 40-person support queue and cut average resolution time from 6 hours to 90 minutes" describes both team leadership and process improvement, even if the words "manager" or "leadership" never appear.
| What it checks | Classic ATS | AI/LLM layer |
|---|---|---|
| Primary method | String/keyword matching against the job description | Semantic reading of the full document for context and meaning |
| Rewards | Exact terms, titles, and phrases from the posting | Described scope, impact, and career trajectory, even without exact keywords |
| Common failure mode | Tables, columns, and text boxes it can't parse into fields | Vague, generic phrasing that reads as low-signal regardless of formatting |
| Examples | Taleo, iCIMS, Greenhouse, Workday's core module | Eightfold, HireVue, Beamery, Phenom People |
| What still gates first | Hard knockout questions (license, work authorization, years of experience) | Typically runs after the ATS pass, on whoever survives it |
Same fact, read two different ways
Bullet: "Managed customer support team." A keyword scanner gives you partial credit for "managed" and "customer support." An LLM reading the same line has almost nothing to score: no scope, no outcome. Rewrite: "Led a 12-person support team and cut average ticket resolution time from 6 hours to 90 minutes by rebuilding the triage queue." Now both systems have something to work with. The keyword match is still there, and the AI layer can infer team size, ownership, and a measurable result.
The lawsuits testing whether these tools have to explain themselves
This is the part most resume advice skips. AI hiring tools are currently being tested in court, and the outcomes could change what companies are allowed to keep hidden. In [Mobley v. Workday](https://www.akingump.com/en/insights/ai-law-and-regulation-tracker/court-allows-discrimination-claims-against-ai-hiring-tool-to-proceed-or-mobley-v-workday-inc), a job seeker named Derek Mobley alleged he was rejected from more than 100 roles that used Workday's AI screening tools and sued, arguing the tool discriminated against candidates by race, age, and disability. A federal court allowed the case to proceed on the theory that Workday itself, not just the employers using it, can be directly liable as an "agent" making hiring decisions. In May 2025, the judge certified a nationwide collective action under the Age Discrimination in Employment Act. As of mid-2026 the case is still in discovery, but it's the first major ruling to say clearly that an AI vendor, not only the company that hired it, can be sued directly over how its tool screens candidates.
A second, newer case adds a different angle. In January 2026, two job applicants filed a proposed class action against [Eightfold AI](https://www.akingump.com/en/insights/ai-law-and-regulation-tracker/ai-hiring-platform-faces-fcra-class-action-over-data-use-or-kistler-et-al-v-eightfold-ai-inc), a platform reportedly used by employers including Microsoft and PayPal, alleging it scores candidates on a 0-to-5 scale using data pulled from outside the resume itself, including social profiles and location data, without telling them a score exists, showing it to them, or giving them a way to dispute it. The claim rests on the Fair Credit Reporting Act, the same law that governs background-check companies, arguing that an automated "likelihood of success" score is functionally a consumer report and should carry the same disclosure and dispute rights. The lawsuit doesn't allege the scoring itself is inaccurate. It alleges the process happened without the candidate ever being told it existed. "Qualified workers across the country are being denied job opportunities based on automated assessments they have never seen and cannot challenge," said Jenny R. Yang, a partner at Outten & Golden and former Chair of the U.S. Equal Employment Opportunity Commission, who helped bring the case.
Why this matters for you
Neither case has been decided on the merits yet, so treat both as pending, not settled law. But together they signal that regulators and courts are starting to treat "the AI read your resume and scored it" as something a candidate may have a right to know about, not just something happening quietly in the background.
One city already regulates this, and enforcement just got stricter
New York City has required something like disclosure since 2023. Under [Local Law 144](https://www.nyc.gov/site/dca/about/automated-employment-decision-tools.page), any employer using an "automated employment decision tool" to screen NYC-based candidates must run an independent bias audit annually, publicly post a summary of the results, and give candidates at least 10 business days' notice before the tool is used, along with a way to request an alternative process. A [New York State Comptroller audit published in December 2025](https://www.osc.ny.gov/state-agencies/audits/2025/12/02/enforcement-local-law-144-automated-employment-decision-tools) found the city's enforcement of its own law had been weak, which pushed the Department of Consumer and Worker Protection into a more active, proactive enforcement posture through 2026. If you're applying to roles based in New York City, you may already be legally entitled to a disclosure that AI is being used on your application, whether or not you noticed one.
What this actually changes about how you should write your resume
- ▸Keep doing the keyword work. The classic ATS pass usually still runs first, and matching the job description's exact terms still decides whether you clear that gate.
- ▸Write bullets in full sentences that describe scope and outcome, not fragments. The AI layer scores context; "owned," "led," and "cut X to Y" give it something to reason about that a bare skill list doesn't.
- ▸Don't let an AI writing tool flatten your resume into generic phrasing. What the AI layer rewards is specificity, real numbers, real scope, real outcomes, not polished-sounding filler. See should you use AI to write your resume for where the line is.
- ▸If you're applying in New York City, check the job posting for the required AEDT disclosure. If it's missing and a large employer is clearly using automated screening, that's worth knowing before you assume silence means rejection.
- ▸Formatting discipline still matters at the parsing stage. A clean, ATS-friendly format is what gets your resume in front of the AI layer in a readable state to begin with.
Key takeaway
The "ATS" most resume advice describes hasn't gone away; it's very likely still the first system to touch your application. What's new is a second, AI-driven layer sitting behind it that reads for context and career trajectory rather than raw keyword count, plus a fast-moving legal fight over whether you're entitled to know it's there. Write for both: match the posting's language, and back every claim with a real, specific, quantified sentence. That combination is what actually holds up under either kind of read.