"ATS" gets used as one word for a wide range of software, which is part of why the advice about it is so muddled. An applicant tracking system is, at its core, a database — it stores applications, lets recruiters search and filter them, and routes resumes to the right hiring manager. The screening and ranking functions that job seekers worry about are a layer on top of that database, and that layer has changed more in the last few years than most advice online has caught up with.
The keyword-stuffing era is over
The old model was crude: the system counted how many times your resume contained words pulled from the job description, and ranked accordingly. That is why a decade of advice says to copy phrases straight from the posting. It worked, sort of, and it also produced a lot of resumes that read like they were written for a machine instead of a person — because they were.
Modern systems do something closer to semantic matching. They are built on natural language models that can tell the difference between someone who "managed a team" and someone who "managed a project timeline," and they weigh context — job titles, seniority signals, how skills are described in relation to actual outcomes — rather than raw term frequency. Repeating "cross-functional collaboration" four times in slightly different phrasing does not help the way it used to. It can actually hurt, because it reads as padding to both the algorithm and the human who eventually opens the file.
What the scoring actually weighs now
Based on how these systems are built and what vendors themselves describe in their own documentation, the scoring generally comes down to a few things:
- Title and seniority alignment. Does your most recent title, and the trajectory before it, plausibly match the level of the role?
- Skills in context, not in a list. A skill mentioned inside a bullet about what you built with it scores differently than the same word dropped into a skills section with nothing behind it.
- Evidence of outcomes. Numbers, scope, and scale are increasingly used as signals of seniority and impact, not just nice-to-haves.
- Clean parsing. This part has not changed — tables, text boxes, headers/footers, and unusual fonts can still cause a parser to misread or drop content entirely, regardless of how smart the ranking model is.
That last point is worth its own read if you have not checked it recently — see what actually makes a template ATS-friendly.
The catch nobody talks about: what screens well and what reads well have started to diverge
Here is the part that matters most right now. Because these ranking systems are themselves built on AI, and a lot of job seekers now write their resumes with AI, there is a real feedback loop: AI-generated resumes are frequently structured in ways that other AI systems parse and score easily — clean bullet structure, predictable phrasing, explicit keyword-to-context matching. That is exactly what makes them easy to spot for an experienced recruiter. Generic, evenly-paced, slightly-too-polished language is now a tell, not a strength. Hiring managers who read hundreds of resumes a year notice the pattern fast, and a resume that clears the automated screen only to read as interchangeable in the human review is not actually a win.
The practical takeaway is not "don't use AI tools." It is that the AI-written first draft is a starting point, not a finished resume — the specificity has to come from you. We wrote a longer version of this argument in AI resume builders: what they actually do.
What to actually do with this
- Pull the job description and identify the 8-10 terms that describe the actual responsibilities and required skills — not filler phrases like "fast-paced environment."
- Use those terms inside real sentences that describe what you did, not as a standalone list bolted onto the bottom of the page.
- Match your most recent job title's seniority level honestly. Inflating a title to match the posting tends to create a mismatch the system (and the recruiter) can detect elsewhere in the resume.
- Keep formatting boring: standard headers, no tables, no graphics, exportable as a clean PDF or .docx.
- Read it out loud before sending it. If it doesn't sound like something you would say about yourself in an interview, a recruiter will feel that gap too.
None of this requires guessing which specific ATS a company uses — Workday, Greenhouse, and the dozen others behave differently in the details, but the underlying principles above hold across all of them, because they are increasingly built on the same category of language model.