You send an application. Nothing comes back. A week later, still nothing, and the explanation that surfaces in every comment thread is the same: a robot read your CV, scored it, and binned it before a human ever saw your name.
It is a satisfying story, because it makes the silence somebody else's fault. It is also, in the systems most people are actually applying through, mostly untrue. The software does something narrower than the folklore says, and knowing what that is changes where it is worth spending your effort.
What the software is actually for
An applicant tracking system is, at heart, a database with a workflow attached. It receives applications, extracts what it can from the files you upload, stores each one as a structured candidate record, and gives the hiring team a place to move people between stages, leave notes and send email. Its primary job is administrative. Judgement is not part of the specification.
That distinction gets lost because "tracking system" and "screening system" have been used interchangeably for a decade. They are not the same thing, and the vendors are unusually clear about it when you read what they publish for candidates rather than what gets written about them.
Greenhouse — used widely across technology and scale-up hiring — puts it plainly in its own guide for applicants:
AI doesn't score or rank applications, nor does it make any decisions about whether or not you move forward.
Greenhouse, "What really happens after you apply for a job"
The same guide says that applications are typically "reviewed by real people, one by one, in the order received", while noting the obvious exception of very high-volume roles.2
Where applications actually get filtered
Three things happen to an application, and only one of them is automatic.
The application questions. These are the checkboxes and dropdowns underneath the file upload: are you authorised to work here, can you commute to this office, do you have five years of X. In Greenhouse, an employer can set a rule so that a particular answer rejects the application outright. The documentation describes it as being able to "set up application rules so that, based on an applicant's answer to a question, they will automatically be rejected as a potential candidate", and it works only with yes/no, single-select and multi-select questions.1 This is the automated rejection people have heard about. It reads your answers, not your CV.
Recruiter search. Once applications are in the database, a recruiter looking at three hundred of them does not read three hundred CVs from the top. They search and filter — by job title, by a tool or a certification, by location. What that search returns depends entirely on what the system managed to extract from your file. Nothing has rejected you at this point; you have simply not been retrieved.
Human review. Someone opens the record and decides. Most rejections are this, and most of them are never explained, which is exactly what makes the algorithm story so easy to believe.
The failure nobody tells you about: parsing
Here is the part that genuinely costs people interviews, and it has nothing to do with keywords or scores.
When you upload a file, the system tries to pull structured fields out of it: your name, your contact details, each employer, each job title, dates, education, skills. That process is called parsing, and it is more fragile than it looks. Workable, an ATS vendor, describes the failure modes in its own engineering write-up: "Tables and columns will put words, and sadly sometimes letters, on different lines", and "try to avoid using headers and footers as they often get interspersed with the main body of text".3 A name set as an image comes out as nothing, or as scattered single letters.
The consequence is not a rejection. It is worse than that in a quiet way: your record ends up with your job titles missing or mangled, so when the recruiter searches for the title they are hiring for, you are not in the results. You were never assessed at all.
| What you did | What the parser may produce | What it costs you |
|---|---|---|
| Two-column layout with a skills sidebar | Sentences interleaved from both columns | Job titles and dates land in the wrong order or the wrong field |
| Name and contact details in the page header | Header text mixed into the body | Missing or wrong contact details on your record |
| Job title inside a table cell | Letters split across lines | You do not appear in a search for that job title |
| A scanned or photographed PDF | An image with no extractable text | An essentially empty candidate record |
| Icons instead of labels for phone and email | Symbols dropped | Contact fields left blank |
None of this requires a special "ATS-friendly template", and the template industry that grew up around this idea has oversold it. One column, real text, standard section headings, and a PDF exported from your word processor rather than scanned from paper covers almost all of it.
Some systems really do rank
The myth is wrong about the mechanism, but it is not wrong that automation is spreading, and an honest account has to say so.
Workday markets AI-assisted matching in its recruiting product, describing a recruiting agent "powered by HiredScore" that takes on screening work, and AI that reads skills from candidates' CVs to recommend relevant opportunities.4 Products in this category surface and prioritise candidates for the recruiter rather than deleting them, but the effect on a queue of four hundred applications is real: order determines who gets looked at while the reviewer is still fresh.
So the accurate summary is not "no machine ever judges your application". It is that automated judgement varies by vendor, varies by how each employer has configured its system, and is far less absolute than "a robot binned it" implies. Where a hard automatic rejection exists, it is usually attached to a question you answered, not to prose you wrote.
What this leaves worth doing
The myth is expensive because it sends people to the wrong work. An evening spent hiding white keywords in a document that nothing is scoring is an evening not spent on the things that do move an application forward.
- Answer the questions properly. They take eight seconds and they are the one part that can end your application automatically. Read the salary and notice-period fields before typing in them.
- Use the employer's own job title. If they advertise "Customer Success Manager" and your CV says "Client Partner", a title search does not find you. Put their title in your CV where it is honest to do so — as the role you are applying for, or alongside your own.
- Keep the layout boring. One column, text as text, dates in a consistent format. Save the design for a portfolio where a human will look at it.
- Mirror the language of the posting, not the keyword list. Ranking, where it exists, works on skills and phrasing. Recruiter search works on the words in your record. Both reward using the vocabulary the employer used, which is also what makes a CV read as relevant to a person.
- Apply to fewer things, more carefully. Fifty untailored applications produce silence that feels like evidence of a broken system, and mostly is not.
Tailoring a CV to a specific posting is genuinely tedious, which is why most people skip it and then blame the outcome on software. It is the part of this that JobCraftly exists to take off your hands: point it at a job, and it produces a tailored version you can review change by change before accepting anything.
What none of this fixes
Changing your CV does not fix a hiring process that was too narrow to begin with. The 2021 Harvard Business School and Accenture study Hidden Workers: Untapped Talent — surveying more than 8,000 workers and over 2,250 executives across the United States, the United Kingdom and Germany — argued that standard hiring practices systematically exclude capable people, and estimated 27 million such workers in the United States alone.56 The Harvard Gazette's report of the study notes that roughly 99% of Fortune 500 companies use automated tracking systems to handle applicants.5
Two caveats on those figures, because they matter. The study is from September 2021, so it describes the market as it was then, not as it is now.6 And its scope is three countries; hiring practice, and the software behind it, differ elsewhere.
The reasonable conclusion is smaller than either the myth or the counter-myth. The software is not reading your CV and finding you wanting. It is filing you, imperfectly, in a system that a person then searches. Make yourself easy to file and easy to find, answer the questions carefully, and spend the effort you save on the applications you actually care about.
