How companies use AI to screen applicants

The common fear is a machine rejecting you outright. The reality is usually less dramatic and more consequential: software parses your resume into structured fields, then ranks or filters candidates so a recruiter reviews a shortlist rather than nine hundred applications.

You are rarely rejected by an algorithm. You are more often simply not surfaced — which has the same effect while being far easier to fix, because it is a visibility problem rather than a judgement about you.

The distinction changes what you should do. If you believe a score is gating you, you try to game a number nobody computes. If you understand that a recruiter is running a search over parsed fields, you make sure you appear in it — a concrete, checkable goal.

Different employers automate different amounts, but the sequence is fairly consistent across mid-size and large companies.

The amount of automation correlates with applicant volume rather than with company sophistication. A small firm receiving four hundred applications for one role automates more aggressively than a large one hiring for a specialism nobody applies to.

• Parsing — extracting your details, employers, dates and skills into a database
• Matching and ranking — scoring against the job description so recruiters review a shortlist
• Knockout questions — hard filters on work authorisation, location or experience thresholds
• Assessments and asynchronous video, sometimes with automated scoring
• Scheduling and messaging, which is now largely automated

Two things matter far more than they intuitively should. First, whether your resume parses correctly — a multi-column layout, a table, or contact details in the page header can make your work history invisible to the system regardless of how strong it is.

Second, whether your vocabulary matches the posting. If the job says "PostgreSQL" and you wrote "SQL databases", a keyword search will not return you. That is not dishonesty on either side; it is two words for one thing, and the search does not know it.

Both are mechanical problems with mechanical fixes, which is the encouraging part. They are also invisible from the outside: nothing in a rejection tells you that your experience section never made it into the database.

Knockout questions are where automatic elimination actually happens, and they sit on the application form rather than in the resume parser. Right to work, minimum years with a named technology, required licences, location and start date are all configured by the employer and applied before any human sees anything.

These are answered carelessly more often than you would expect. A question asking for years of experience with one specific tool is narrower than total career length, and an answer given quickly is not recoverable once submitted.

Read the form as carefully as the posting. It is the one place in the process where a single wrong click ends the application regardless of everything else you did.

It is a legitimate concern. Systems trained on historical hiring decisions can reproduce historical patterns, and there have been well-documented cases of exactly that. Regulation is tightening in several jurisdictions, and some employers now audit these tools.

As a candidate you cannot fix this, and it is worth being clear-eyed rather than reassured. What you can control is the mechanical part — that your application parses cleanly and uses the vocabulary being searched for — which is where most avoidable losses occur.

A growing number of jurisdictions require employers to disclose automated screening and, in some cases, to offer human review on request. Where that applies, asking is reasonable and is not held against you, though the rules differ enough that it is worth checking what applies locally.

It is worth acknowledging the other side. Before automation, a recruiter facing four hundred applications read perhaps the first eighty. Everyone after that was rejected by arrival order, which is not a fairer process than a search over parsed fields.

Response rates have improved too. Automated status updates and scheduling remove much of the silence candidates used to experience, and applications now get acknowledged at volumes where that was previously impossible.

The realistic view is that automation replaced one arbitrary filter with a different one that is at least consistent and, unlike arrival order, something you can prepare for.

Use a single-column, parse-safe layout with standard section headings, mirror the posting’s terminology where it genuinely applies to you, and export a text-based PDF. Then check the result against the actual job description before submitting.

Two minutes of that per application does more for your odds than another hour of formatting, because it targets the specific mechanism that is filtering you out.

And remember that a referral bypasses most of this. A referred application is read by a person; a cold one has to survive a search first. Where you have any route to one, it is worth more than any amount of resume optimisation.