How ATS actually works

How an ATS Actually Screens Your Application

The real order is knockout questions on the form, then parsing, then a recruiter typing two or three words into a search box. No ranking score decides it.

There is a story about applicant tracking systems that almost everybody believes. Your resume goes in, software gives it a score out of a hundred, the top few are printed out for a human, and the rest are deleted. It is a tidy story. It explains rejection without requiring you to think about anything uncomfortable, and it sells a great deal of resume optimisation.

It is also not how the machinery is built. The order of operations is different, and once you know the real order, the list of things worth fixing changes completely.

The real order of operations

Four things happen, in this sequence, and each one is a different kind of gate.

1. The form screens you. Answers to the application questions, before anybody opens the document.
2. The parser converts you. Your file becomes a set of database fields. This is where most damage happens, silently.
3. A recruiter searches. Two or three words in a box, over the pile that survived.
4. A human reads what came back. Quickly, from the top of the page down.

No step in that sequence is a ranking score applied to everybody. Where a score exists at all it appears late, it is optional, and it is a different thing from the score you are shown by a resume checker.

Stage one: the questions on the form

The most consequential screen is the one that costs you thirty seconds and no thought. Greenhouse documents three application rules that fire on answers to custom application questions: Auto-Tag, Auto-Advance and Auto-Reject. The Auto-Reject rule, in Greenhouse's own words, exists to help "recruiters with high-volume roles by filtering, rejecting, and automatically emailing candidates who do not meet essential criteria, like certain license or location requirements". Their documented examples of non-negotiable answers are blunt: not enough experience, underage, requires a visa the company cannot provide. When an Auto-Advance rule and an Auto-Reject rule both fire, Greenhouse applies the rejection.

JazzHR calls the same feature knockout questions, and flags disqualifying answers in red on the candidate profile. Every major system has some version of it.

This is a real filter, it runs before parsing, and no wording on your CV reaches it. The practical consequence is unglamorous: answer the form carefully, in the format it asks for, and treat every dropdown as a decision rather than admin.

The Harvard Business School and Accenture study Hidden Workers: Untapped Talent (2021) surveyed employers on exactly this behaviour and found that more than 90% of them use their recruiting system to initially filter or rank candidates, 94% for middle-skills roles and 92% for high-skills roles. The report is worth reading because it is honest about what the criteria are: employers use "proxies (such as a college degree or possession of precisely described skills)" and exclude on "a failure to meet certain criteria (such as a gap in full-time employment) ... irrespective of their other qualifications". 88% of employers agreed that qualified high-skills applicants get vetted out for not matching the exact criteria in the job description.

That is the filtering everybody blames on the algorithm. It was configured by a person.

Stage two: parsing, where the damage is silent

Your file is not stored as a page. It is converted into fields.

Oracle publishes the list for Taleo, and it is short: name, address, city, postcode, phone, email, place of residence; education level, institution, programme, start and graduation dates; and for each job, start date, end date, employer, job function and responsibility. Oracle also documents that a resume submitted for parsing "cannot exceed 100 kilobytes or the size defined by the system administrator". Workable documents that when a name field is left blank, the name is extracted from the resume, and "if no resume is found, or the name cannot be parsed, the name will be listed as Uploaded Candidate".

That is the whole risk, stated by the vendors. What extracts becomes you. What does not extract does not exist.

Workday describes the step in general terms on its own site: "the system's parsing engine ... extracts details from resumes, such as education, skills, and work history. Then it organizes them into structured profiles." Greenhouse lists what its Talent Matching feature pulls out: skills, job titles, years of experience, start and end dates of employment, company names, and a derived industry classification.

Notice what is on none of those lists. Your design. Your icons. The clever two-column layout. Those either survive as text or they do not, and the parser never reports back to you either way.

Stage three: the search box

This is the part the myth gets most wrong. A recruiter with four hundred applications does not scroll. They search.

Ashby documents a full-text search over the candidate's resume with four match modes and boolean operators, and the modes are instructive:

Lever's search "will look for candidate names and any parseable content attached to candidate profiles, including resumes, notes, and feedback forms", and sorts what it finds by last interaction date, most recent first. Greenhouse's full text search returns matching candidates with "a small snippet under their name" showing how the search terms are used in the profile.

Three vendors, three documented behaviours, and not one of them is a ranking of everybody by fit. They are retrieval tools with a human attached. Which produces the single most useful sentence in this whole subject: if your resume does not parse correctly you will not appear in that search, regardless of qualifications.

Stage four: a person, quickly

Ladders ran an eye-tracking study in 2018 and reported an average of 7.4 seconds on a first pass, up from six seconds in 2012. Treat the number as indicative rather than precise: it was a vendor study of thirty recruiters, and it has been argued about ever since. The direction is not in dispute, and it matches what recruiters say themselves. The first screen is fast, and it looks at titles, employers and dates.

What each stage can actually do to you

About that 75% figure

You have seen the claim that 75% of resumes are rejected by software before a human sees them. It is worth knowing where it came from, because the answer is nowhere. HR practitioner Christine Assaf followed the citation chain in 2020 and found every article pointing back to a job-services firm called Preptel, which published no methodology and went out of business in August 2013. Her conclusion: "there's no concrete source data, or research to even back up the statement."

We repeat this not to be contrarian but because the figure changes behaviour. If you believe a bot deletes three quarters of applications, you optimise for the bot. If you believe what the vendor documentation actually says, you spend that hour on the form answers, the parse and the first third of page one.

What to do with this

1. Treat the application form as the first interview. It is the only stage that can reject you without a person.
2. Check that your document parses before you send it, rather than hoping. We wrote a ten-minute method for doing this yourself.
3. Pick the two or three terms a recruiter would type for this role and make sure your page contains them as words, attached to real work.
4. Put your current title, employer and dates where a seven-second read lands. Top left, not page two.
5. Stop optimising for a percentage. It is either not there, or it is not yours.

Run a posting against your CV and start with what did not survive the file, not with the score.

Sources: Greenhouse, Application rules overview, Greenhouse, Search resumes for keywords, Oracle Taleo, Candidate Management, Ashby, Candidate Search, Lever, Searching the database for candidates, Workable, Importing candidate data, Harvard Business School and Accenture, Hidden Workers: Untapped Talent, HRTact, Your job application was rejected by a human, HR Dive on the Ladders eye-tracking study