How AI resume screening reads your CV, and how to get through it

When you apply through a company's careers page or job boards like the Search O Pal job portal, your CV usually reaches software before it reaches a person. That software is strict. In a Harvard Business School and Accenture survey of employers, 88% said qualified high-skilled candidates were rejected outright because they did not exactly match the hiring criteria, and 94% said the same about middle-skilled candidates (Hidden Workers report).

AI is now part of that first pass at more companies. SHRM's 2025 Talent Trends survey of 2,040 HR professionals put AI adoption in HR tasks at 43%, up from 26% a year earlier.

Below is what the software does with your file, where it misreads good candidates, and a checklist to run before you apply.

What happens to your CV after you click apply

Screening usually runs in four stages. Not every employer uses all of them, but knowing each one tells you what each part of your CV has to do.

The parser turns your file into data

MIT's career office describes most applicant tracking systems as being designed like a database. The first thing the system does is parse your file. It pulls out your name, contact details, job titles, employers, dates, education, and skills, and stores each one in its own field.

Everything after this works on those fields, not on the page you designed. If the parser drops your job title into the skills field, or loses a section completely, every later step is judging the wrong information.

Screening questions filter first

Many applications ask a few questions before or after the upload. Do you have the right to work in this country? How many years of experience do you have? Can you work weekends? Many application systems let employers mark these questions as required and filter applicants on the answers. A careless answer here can end an application before anyone, human or model, looks at the CV.

The software scores you against the job post

Some systems then rate or score each applicant by comparing the parsed fields with the job posting (MIT). Older tools count exact keyword matches. Tools built on language models can also recognise that client onboarding and customer onboarding mean the same thing, but you can't know which kind an employer runs. The posting's own words are still the safest choice.

The result is usually a ranked list, with a score or match percentage next to each name.

A recruiter searches and filters

A person usually comes in next, often working inside the same database. Recruiters sort by score, filter by location or by answers to the screening questions, and search for a skill or a job title. If the word they search for is not in your CV, you don't show up in the results, however strong the rest of it is.

What screening software misreads

Parsing is where most avoidable damage happens, and layout causes nearly all of it.

• Tables, text boxes and images. MIT warns that these "may either be distorted, ignored, or erased by an ATS."
• Two-column layouts. MIT names columns and text boxes as common reasons text comes out in the wrong order. A skills sidebar can end up mixed into the middle of your job history.
• Headers and footers. UC Berkeley's career center tells students to avoid headers, footers, text boxes, tables, colors, pictures and graphics. Contact details placed in the header can go missing.
• Creative section names. A parser knows where to file Experience, Education and Skills. It may not know what to do with My Story or Where I've Been. Berkeley recommends standard section titles for the same reason.
• Abbreviations on their own. MIT advises against abbreviating relevant keywords, because the ATS may not count them. Write the term in full once, with the short form after it: Search Engine Optimisation (SEO).
• Scanned or image-only PDFs. If you can't select the text in your PDF with your cursor, the parser can't read it either.

Where AI screening gets it wrong

A perfectly formatted CV can still be filtered out for reasons that have nothing to do with ability. The Hidden Workers report found employers relying on negative filters, such as rejecting anyone without a particular degree or with a gap in their employment history.

Ranking with AI adds problems of its own. In a 2024 University of Washington study, researchers had three large language models rank more than 550 real resumes, changing only the names. The models favoured white-associated names 85% of the time and female-associated names 11% of the time.

You can't fix that from your side. You can make sure the software reads your CV correctly and finds the words it is looking for, so that you are never filtered out for a reason you could have prevented.

How to get through AI resume screening, step by step

Run these steps for every application. Most of them take a few minutes once your base CV is clean.

1. Read the application instructions first. If the posting asks for a Word file, send Word. ORISE's guidance is to follow the posting's file requirements exactly. Where it says nothing, MIT says a .docx or .pdf file is usually fairly safe.
2. Answer the screening questions precisely. Give your real years of experience and read each question twice. Required questions are often the first filter.
3. Use the posting's language. If the advert says Accounts Payable Officer and that describes your work, use that title in your summary. Repeat the posting's terms for the skills you actually have, and write acronyms in full once.
4. Use one column and standard headings. Summary, Experience, Education, Skills. Put contact details in the body of the page, not in the header. If you'd rather not rebuild the layout yourself, Careerkit's free resume builder exports a PDF with selectable text that hiring software can read, and you don't need an account to use it.
5. Write bullets a person will stop on. The score gets you onto the shortlist, and a recruiter decides what happens next. ORISE describes this as writing for two audiences: the scanner that reads first and the person who makes the final decision. Start each bullet with a verb and add a number where you honestly have one. There are two examples after this list.
6. Test the file. MIT suggests saving your CV as plain text (.txt) and reading it from top to bottom. Missing sections or jumbled order mean the parser sees the same mess. You can also upload it to Careerkit's free resume parser, which shows every field a parser pulls out of your CV and every field it misses.
7. Don't stuff keywords. MIT also warns against spamming the ATS with keywords. Listing skills you don't have, or hiding repeated terms in white text, may get past one filter and then fall apart at the interview.

Two bullets rewritten for both readers:

Before: Responsible for social media accounts.

After: Grew the clinic's Instagram from 900 to 2,400 followers in six months by posting three short patient-education videos a week.

Before: Handled customer complaints.

After: Resolved an average of 35 customer complaints a day by phone and email, keeping the team's satisfaction score above 90% for four quarters.

Checklist before you apply

• The file type matches what the posting asks for.
• You can select and copy the text in your PDF.
• One column, with no tables, text boxes or images.
• Contact details sit in the body of the page, not in the header or footer.
• Headings are standard: Summary, Experience, Education, Skills.
• The job title and the main skills from the posting appear in your CV, where they are true.
• Every acronym is written in full at least once.
• Screening questions are answered accurately.
• Every bullet starts with a verb, and most include a number.
• The plain-text version reads in the right order with nothing missing.

Conclusion

AI screening reads your CV as data first and as a document second. Keep the layout simple enough to parse, use the posting's own words for skills you really have, and write bullets for the recruiter who opens the shortlist. Those are the parts of screening you control, and they decide whether a good application gets seen.