How Applicant Tracking Systems Work: Workday, Greenhouse and 8 More

How does an applicant tracking system work? We studied ten of the most common ones, including Workday, Greenhouse, Lever, Ashby, iCIMS, SAP SuccessFactors, Oracle, SmartRecruiters, Workable and the Eightfold and Phenom layers that sit on top of them. They all do three things in the same order: they check your answers on the application form, they parse your resume into fields, and then they judge each of the employer's requirements against something your resume actually says. The automatic rejections almost always come from the first step, not the third. So getting through is less about keywords and more about answering the form truthfully, proving each must-have in a real line, and handing the parser a file it can read.

How an applicant tracking system works: three steps, in order

Strip away the brand names and every system we looked at runs the same pipeline.

1. The form. Yes/no and multiple choice questions about work authorization, sponsorship, location, years of experience, degrees and licences. This is the only step that rejects people automatically on every platform.
2. The parse. Your file is turned into a profile: job titles, employers, dates, education, skills and contact details. Years of experience are usually calculated from those dates.
3. The judgement. The employer's requirements are compared with your parsed resume. Newer systems do this one requirement at a time with AI and show the recruiter which requirements you meet.

After that, a person decides. On most platforms the AI score sorts and explains; it does not reject on its own. The exceptions are covered below.

If you have been rejected minutes after applying, start with why your resume is getting rejected by an ATS, which walks through the causes in the same order.

What we studied, and what we could not see

We looked only at what any job seeker or developer can see. That meant the public candidate pages where you apply, the public job board APIs that power careers sites, the vendors' own help centers and administrator guides, and published bias audits. We read the code these pages ship to every visitor's browser. We did not sign in to any employer account, submit a real application or test anyone's private system.

That has limits worth saying plainly:

• No vendor publishes its scoring weights or the cut-off between a strong and a weak match.
• The recruiter screens, where an employer sets its own rules and filters, are behind a sign-in.
• We can see which questions an employer asks. We usually cannot see which answers it has set to reject.

Where a claim below comes from a vendor's own documentation, it links to that page.

Most automatic rejections happen on the application form

Every system we studied can reject an application because of an answer, and most employers set up the questions that make this possible. Greenhouse publishes each job's form through a public API, so we read the forms of real Greenhouse employers.

80 of the 89 Greenhouse employers we checked (90%) had at least one required yes/no or multiple choice question of the kind Greenhouse lets employers attach automatic rules to. The topics were consistent. Across the 177 jobs, 134 asked about visa sponsorship and 115 asked about work authorization, far ahead of relocation, years of experience, non-compete agreements and on-site work.

The other platforms work the same way, with some differences in how visible the rule is:

• Workday lets employers put points on multiple choice answers. Its administrator guide describes using those points to eliminate ineligible candidates, and in the public questionnaires we read, five employers gave the answer they did not want minus 1,000 points.
• Oracle Recruiting has disqualification questions, and its documentation tells employers to give a negative score to an answer that disqualifies.
• Ashby can auto-reject on answers, including typed answers that contain, start with, end with or are similar to a phrase the employer chose.
• Workable knock-out questions disqualify a candidate who answers No.
• SuccessFactors, Lever and SmartRecruiters can screen an application out on its answers, and Taleo has its own disqualification questions. iCIMS scores screening questions; we could not confirm from public material whether it rejects on them automatically.

Two things make this step easy to miss. First, most job postings never mention sponsorship or work authorization, so the question appears for the first time on the form. Second, you are usually not told. A rejected application often just shows as received or processed.

What you can control: answer the same question the same way on every application, read the form before you write a cover letter, and check the posting for location, schedule and credential requirements before you start.

The AI score reads your resume one requirement at a time

The second big shift is how the matching works. Older systems mostly counted keywords. The current generation takes the employer's list of requirements and judges each one separately, with a reason.

• Workday's HiredScore grades applicants from A to D. Its grade reference is explicit: an A or B meets all the basic (required) qualifications, a C meets most of them, a D does not meet most.
• Greenhouse Talent Matching sorts applicants into Strong, Good, Partial, Limited or Needs manual review, according to its Talent Matching FAQ, and shows matched skills and experience.
• Ashby's AI review marks each recruiter-written criterion as meets, does not meet or undecided, with reasoning, as its help article describes.
• Lever's Talent Fit gives a verdict per requirement and quotes the line of the resume it used as evidence, or says it found none.
• Workable's agent scores against matching criteria, and its documentation says a must-have that is not met weighs more than any other criterion.
• Oracle Recruiting offers generative AI matching ratings of how well an application fits the requisition.

Put those together and the practical rule is clear. A requirement is judged against evidence. A skill that sits only in your skills list gives the system a name with nothing behind it. A line in your experience that shows the skill in use gives it something to quote.

Use the posting's exact words where they are true. Greenhouse shows recruiters exact keyword matches separately from similar ones, and keyword filters that recruiters set by hand are literal. If the posting says "JavaScript" and you wrote "JS", write the posting's form. Our guide to finding the keywords in a job description shows how to pull them out, and tailoring a resume to a job description shows how to work them into lines you already have.

Mostly, the score advises a recruiter

This is the part most advice gets wrong. On most platforms the AI score does not reject anyone.

Greenhouse is direct about it: its Talent Matching FAQ says the feature does not auto-reject or auto-advance any candidate, including those in the lowest categories, and the hiring team makes every decision. Workday's HiredScore grade, Ashby's criteria, Lever's Talent Fit and Oracle's ratings are all presented to a recruiter as a way to sort and read applications faster.

There are exceptions, and they are switches an employer has to turn on:

• Workable lets an employer mark a criterion as a disqualifier. Its agent documentation says the candidate is disqualified only if the employer enables that action, and the agent can also advance candidates automatically.
• Workday can move top-graded applicants forward automatically.
• Oracle can compute its AI ratings for every new application, which makes the rating the first thing a recruiter sees.

So a low match is rarely an automatic no. It is a lower place in a queue that a busy recruiter may never reach. That is still a reason to prove every must-have, but it is not a reason to panic about a hidden score.

If the parser cannot read your file, you wait for a person

Every system parses your resume before it scores anything. When parsing fails, you are not rejected; you are set aside. Greenhouse labels these applicants Needs manual review and leaves them out of matching, and its FAQ lists formatting or technical issues with the resume as one cause. Lever and Ashby have similar states, and HiredScore gives no grade when the resume is missing or in a format it cannot read.

Some details that matter more than people expect:

• The safe formats are DOC, DOCX, PDF, RTF and TXT. A text-based PDF is fine. A scanned PDF or an image of a resume is not.
• Titles, employers and dates feed the years-of-experience number. Several systems calculate years from your dates and show them next to your name. Write every role as title, employer and dates, the same way each time.
• Some systems trust what you confirm more than the file. SmartRecruiters scores the fields you confirm on the form before it looks at the resume, and Workday's autofill can leave a degree blank when it cannot match it. When the form fills itself in from your resume, check the titles, dates and degree before you submit.
• Two of the ten use the same parser. SAP SuccessFactors and SmartRecruiters both rely on Textkernel to read resumes, so a layout that confuses it confuses both.

Our ATS-friendly resume checklist covers file type, layout and headings in detail.

Gaming the system gets flagged, not rewarded

Hidden white text, pasted job descriptions and keyword blocks were never a good idea, and the newer systems look for them.

Ashby's web app ships a candidate flag labelled "Detected a resume trying to manipulate AI scoring", aimed at hidden instructions and keyword tricks. Ashby also runs fraud checks on new applications that send suspicious candidates to closer manual review. Among the signals we found in its app are a resume email that does not match the application email, and a phone number or connection that does not match your stated country. Lever can tag an application as suspicious for signals such as an internet phone number, an unusual email domain or employers it cannot verify, and Greenhouse's fraud report looks at IP address, email, phone and VPN use.

None of these reject you for an honest mistake on their own, but each one puts a question mark next to your name. The fixes are simple: one email on the resume and every application, a real phone number, and nothing on the page that a person cannot see.

A checklist that covers what actually matters

Here is everything above turned into steps you can run before each application.

• Answer screening questions truthfully, and the same way every time you are asked.
• Read the application form before you tailor; note sponsorship, authorization, location, schedule and credential questions.
• Give every must-have requirement one line in your experience that proves it, in the posting's words where they are true.
• Keep the skills list, but do not let it be the only place a required skill appears.
• Use standard headings: Summary, Experience, Education, Skills, Certifications.
• Write each role as title, employer and dates, in plain text, never inside an image or a text box.
• Save as DOCX or a text-based PDF unless the employer asks for something else.
• Use the same email on your resume and on the application.
• Check what the form filled in from your resume before you submit.
• Never add hidden text, white text, pasted job descriptions or keyword blocks.

Where FitMyRole fits in

We built two checks into FitMyRole from this research. Before you apply, it shows the application questions a job is likely to use to reject people automatically, such as sponsorship, work authorization, years, degree, licences and on-site work, and checks them against the answers you saved. For jobs posted on Greenhouse, Ashby or Workday, it shows the employer's own required questions. It never suggests what to answer.

For required skills that your resume only lists, FitMyRole asks which job you used them in and adds a line in your own words, or drafts one from your own bullet for you to approve. It rewrites your own lines in the posting's words and never adds experience you do not have, and every edit comes with the reason it was made. Plan details are on FitMyRole plans.

The short version: the form decides who gets rejected automatically, the parser decides whether you are read at all, and the AI score mostly decides where you sit in the queue. Answer honestly, prove each requirement in a real line, and give the parser a clean file. That is how you get through all ten.