How AI screening is changing the way you should apply for jobs

Almost every application you send to a mid-sized or large employer now passes through software before a human reads it. That has been true for years, but the software has changed. Applicant tracking systems that used to be glorified filing cabinets are now bolted to ranking engines, chatbots, skills-inference tools and, increasingly, large language models that summarize your resume for a recruiter who will spend well under a minute on it.

A whole industry of advice has grown up around “beating the bots”, and much of it is wrong or out of date. This piece explains what the systems actually do, what gets people rejected in practice, how to format so nothing is lost in parsing, and why the old trick of stuffing your resume with keywords is now more likely to hurt than help.

What an applicant tracking system actually does

An applicant tracking system (ATS) is a database with a workflow attached. Workday, Greenhouse, Lever, iCIMS, SAP SuccessFactors and Oracle Taleo are the names you will see most often in the URL when you apply. When you upload a resume, the system parses it into fields: name, contact details, employers, job titles, dates, education, skills. Those fields feed a candidate record that recruiters search, filter, tag and move between stages.

The important word there is “recruiters”. In the large majority of ATS setups, the software does not silently reject you on a keyword score. What it does is make you findable or not findable. A recruiter with 400 applicants for a role will run a search (“SQL” AND “Tableau”, location within 30 miles, applied in the last 14 days) and look at whoever surfaces. If your resume was parsed badly, or you described your skills in words the recruiter did not search for, you are still in the database. Nobody is looking at you.

The other thing an ATS does reliably is enforce knockout questions. “Are you legally authorized to work in this country?” “Do you hold a current commercial driver’s license?” “Are you willing to work on site in Denver?” Answer “no” to a required question and the system will move you to a rejected pile automatically. This is the single most common way software rejects people, and it has nothing to do with AI.

Where the AI comes in

Layered on top of the ATS are tools that do more than store and search. Some rank applicants against the job description using semantic matching, so “managed a team of eight” and “people leadership” are treated as related rather than as different strings. Some infer skills you did not list from the roles you held. Some run the first-round screen as a chatbot conversation. Some score recorded video interviews, though that practice has retreated after regulatory pressure and bad press. LinkedIn Recruiter, which many recruiters use alongside their ATS, has its own matching and recommendation layer.

Regulators have noticed. New York City’s Local Law 144 requires employers using automated employment decision tools to have them independently audited for bias and to tell candidates the tools are in use. Illinois has rules on AI analysis of video interviews. The EU’s AI Act classifies AI used in recruitment and worker management as high risk, with obligations that phase in over several years. In the US, litigation over whether screening vendors can be held liable for discriminatory outcomes is ongoing, and the outcome will shape how aggressively employers automate the top of the funnel. None of this means the tools are going away. It means employers are being pushed to keep a human accountable for the decision, which is good news for you if you can get in front of that human.

What actually gets people filtered out

Set aside the myths and the list of real filters is short and mostly mundane:

• Knockout answers. Work authorization, location, licenses, minimum years of experience, willingness to travel, shift availability. Read these questions slowly. A careless “no” on a question that was actually optional or that you misread will end your application without anyone seeing your resume.
• Recruiter-set filters. Recruiters filter on the fields the parser extracted: current title, years in role, education level, distance from the office, date applied. If the parser missed your most recent title because it sat inside a text box, you drop out of the “current title contains ‘accountant’” search.
• Hard requirements in the posting. A CPA, a security clearance, a nursing license, a specific certification. If the posting says required and you do not have it, the ranking tool will place you low and the recruiter will not scroll that far.
• Volume and duplicates. Several systems flag people who apply to a dozen unrelated roles at the same company in a week. It reads as spraying, and recruiters treat it that way.
• Bad parsing. Two-column layouts, headers and footers, tables, icons, unusual fonts and image-based PDFs can all cause the parser to scramble dates, drop sections or attach your skills to the wrong job. You never find out. You just do not get a call.

Notice what is not on that list: a mystical keyword density score. That is not how most of these systems are built.

Formatting for the parser

Parsers have improved, especially those built on language models, but the safe format has not changed much. It is worth following even if a given employer’s system could cope with more.

Use a single column with standard section headings: Summary, Experience, Education, Skills, Certifications. Put your name and contact details in the body of the document, not in a header or footer, because some parsers ignore those areas. Write dates in a consistent format (Mar 2022 to Jan 2025, or 03/2022 to 01/2025). Avoid tables, text boxes, columns, graphics and skill-rating bars. Use a common font. Save as a .docx unless the employer asks for PDF; if you send a PDF, make sure it was exported from a text document rather than scanned, so the text is selectable.

For job titles, use the standard industry title and put your internal title in parentheses if it differs: “Customer Success Manager (internal title: Client Partner)”. Recruiters search on the standard term.

For skills, mirror the posting’s phrasing once. If the job asks for “Salesforce administration” and you have it, write “Salesforce administration”, not “CRM management”. Then stop. One clear mention in a skills section and one in context under a role is all a search or a semantic matcher needs.

If you are starting from scratch, our guide on writing a resume with no experience includes a full plain-text layout that parses cleanly.

The case against keyword stuffing

The old trick was to paste the entire job description into your resume in white text, or to add a “keywords” block listing every tool ever mentioned in your industry. Do not do this, for four reasons.

First, modern parsers and ranking tools increasingly work on meaning rather than string matching, so repetition buys you nothing and unrelated terms can actively confuse the skills the tool infers for you. Second, several systems now convert the document to plain text before indexing, which turns hidden white text into a visible wall of nonsense at the bottom of your resume. Third, the recruiter reads what the tool surfaces, and a resume that lists 60 skills with no evidence of using any of them reads as exactly what it is. Fourth, some employers now run checks for exactly this behavior and treat it as an integrity flag.

The better version of the same instinct is targeted mirroring: pick the three to five requirements the posting cares most about (usually the ones repeated in the title, the first bullet and the “must have” list), make sure each appears in your resume in the posting’s own words, and back each with a sentence showing you did it.

Using AI tools on your side

Employers can tell when a cover letter was generated and not edited. The tells are familiar: generic enthusiasm, a paragraph that restates the job posting, no specific claim that could be checked. Some recruiters now say openly that they discard letters that read this way. A growing number of application forms ask you to confirm whether AI was used to prepare your materials, and a few employers have begun designing screening questions that a language model answers badly without personal input.

Use the tools for what they are good at. Ask one to pull the key requirements out of a long posting. Ask it to tighten a bullet you drafted. Ask it to check your resume against the posting and list what is missing. Then write the actual sentences yourself, with the specific numbers and named projects only you know.

Avoid auto-apply services that submit hundreds of applications on your behalf. They tend to answer knockout questions generically, trip duplicate detection, and leave you unable to remember what you applied for when a recruiter calls.

The human step that still decides

Every one of these systems ends in a person. A recruiter screens a shortlist, a hiring manager picks who to interview, a panel decides. The tools shape who gets into the shortlist. They do not choose the hire.

That is why the highest-value move in a screened process is still to reach a human independently of the queue. A referral from a current employee typically lands your application in a separate, prioritized pile in the ATS. A short message to the hiring manager or recruiter on LinkedIn, sent after you apply, means someone searches for your name rather than waiting for your resume to surface. Our piece on using LinkedIn to get a job without posting covers the exact wording.

A routine for each application

Use this in order. It takes about 25 minutes per role, which is why you should be applying to fewer, better-matched roles rather than more.

1. Read every application question before you upload anything. Note which are required and which are knockouts. Answer them precisely, and do not guess on licensing or authorization questions.
2. Identify the three to five must-have requirements in the posting. Confirm each appears verbatim in your resume, with one line of evidence.
3. Check your current title matches the standard industry term, with your internal title in parentheses.
4. Confirm the file is single column, no header or footer content, no tables, .docx or a text-based PDF.
5. If there is a free-text field or cover letter, write two short paragraphs that make one specific, checkable claim about your relevant work and one about why this employer.
6. Within 24 hours, find one human connected to the role and send a two-line message that names the job and one reason you fit.

If you do only the first and last steps, you will already be ahead of most of the applicant pool. The systems are built to find people. Your job is to make sure there is something clear and specific for them to find, and then to make sure someone goes looking.