ATS-Friendly Resume: How Screening Software Actually Reads Your CV

Open your CV, press Ctrl+A, and look at what highlights. If your name and phone number sit in a page header and don't light up with the rest of the text, a large share of applicant tracking systems will never see them. An ATS-friendly resume is one whose text can be pulled out cleanly by parsing software and mapped to the correct fields — employer, job title, start and end dates, skills, education — so that a recruiter searching the database actually finds you. It's a formatting problem, not a scoring problem. Almost everything circulating online about ATS gets that backwards.

Does an ATS reject resumes on its own?

Very rarely, and not in the way the internet suggests.

The claim you've seen — that 75 per cent of resumes are rejected by an ATS before a human sees them — has no study behind it. Writers who have tried to trace it end up at Preptel, a resume-optimisation vendor that was pitching this figure around 2012 and shut down the following year. It was repeated in a 2014 Forbes column written by the founder of a resume service, and again in a 2018 CIO.com piece, neither citing any underlying data. No ATS vendor has ever published a number like it. It survived because it is frightening and easy to quote, and because the people repeating it were selling something.

What does happen is more mundane and, in some ways, harder to beat.

Knockout questions do reject people automatically. These are the application-form questions about work authorisation, minimum years of experience, notice period, location, or a required licence. Answer one wrong and you can be filtered out before anyone opens your CV. That is a genuine auto-reject, and it comes from the form, not the resume.

The larger effect is exclusion by search. Harvard Business School and Accenture surveyed 2,275 executives in the US, UK and Germany in early 2020 and published the results as "Hidden Workers: Untapped Talent" in September 2021. Of those executives, 88 per cent agreed that qualified, high-skilled candidates were being filtered out of their processes because they didn't match the exact criteria configured in the system. For middle-skilled roles the figure was 94 per cent. Around 48 per cent said they filtered out middle-skilled applicants with an employment gap of six months or more. Notice what that describes: not software judging your writing, but humans configuring filters that are too narrow, and candidates disappearing behind them.

One regulatory point is worth knowing as a candidate. The EU AI Act classifies recruitment and candidate-selection systems as high-risk under Annex III, though the Digital Omnibus on AI — now law — deferred most of those stand-alone obligations to 2 December 2027. What did apply from 2 August 2026 is the Article 50 transparency duty: if you're interacting with an AI system, you're supposed to be told. New York City's Local Law 144 goes further, requiring an annual independent bias audit of automated employment decision tools and a public summary of the results. Tools that genuinely decide are increasingly required to say so.

What actually happens when a parser reads your CV?

The file arrives and the system extracts a text layer. For a Word document that's straightforward. For a PDF it depends entirely on how the PDF was made — a PDF exported from Word carries real, selectable text, while a PDF exported as an image from Canva or a scan of a printout carries none, and the parser gets a blank page unless the vendor runs OCR.

Then the parser tries to impose structure on a flat stream of text. It looks for section headings it recognises — Experience, Education, Skills, Certifications — and treats everything between two headings as belonging to the first. Inside your experience block it hunts for date ranges, then attaches the nearest preceding or following lines as the employer and the job title. Titles and companies are usually identified by a mix of pattern rules and a trained model checking against known company names.

That chain of guesses is where things break. A single mis-read heading throws every subsequent block into the wrong bucket. An unreadable date range means your tenure at a job can't be computed, so a filter for "5+ years experience" silently skips you.

Parsing has improved. An October 2025 arXiv paper on layout-aware resume extraction, from a team whose pipeline runs inside Alibaba's HR platform at 240 to 300 resumes a minute, found that adding layout understanding ahead of a large language model raised extraction F1 from 0.919 to 0.959. It also noted that roughly 20 per cent of resumes use non-linear, multi-column layouts that break normal reading order. Modern systems handle those better than the engines of 2015. Older ones, still widely deployed, do not.

What breaks a resume parser?

• Tables. Text inside table cells is often read across rows rather than down columns, or dropped. Indian CVs are especially exposed here, because the common convention of a photo, date of birth and contact block laid out as a table means the parser can lose your phone number and email in one go.
• Multi-column layouts. Jobscan published a test in April 2026 in which a design-heavy two-column CV was run through Lever's parser: it captured the work experience and nothing else — no skills, no summary, no contact details, no portfolio links.
• Headers and footers. Many parsers ignore those layers entirely. This is the single most common way people lose their contact details.
• Text boxes and graphics. Skill bars, rating dots and infographic timelines contain no extractable text. A five-dot rating for Python tells the machine nothing at all.
• Text inside images, and PDFs exported as images. Nothing to extract.
• Non-standard section names. "My Journey", "Where I've Been", "The Toolkit" are invisible as structural markers. The parser doesn't know they mean experience and skills.
• Unusual date formats. Jan '21 with a typographic apostrophe, or a bare 2021–2023 with no months, frequently fails. Jan 2021 – Mar 2024 works.
• Non-standard or unembedded fonts. When the parser can't map a glyph it substitutes something else, which is why parsed CVs sometimes come back reading "Pro?le" instead of "Profile". Arial, Calibri, Georgia, Cambria, Helvetica and Verdana are safe.
• Emoji and icon fonts in the contact line. A phone icon becomes a stray character or nothing.

Which ATS myths should you stop believing?

The white-text one deserves a blunt answer. It does not work, and it is the fastest way to end a candidacy. Built In reported in October 2025 that ManpowerGroup detects hidden text in roughly 10 per cent of the resumes it scans, while Greenhouse puts the figure closer to 1 per cent. Because parsers discard formatting, the white text and the smuggled instructions to the AI arrive in the recruiter's view as ordinary black text sitting oddly in the middle of your CV. Recruiters quoted in that piece said they almost always eliminate candidates who try it. Former Google recruiter Farah Sharghi's framing is the one worth remembering: it signals that you don't trust your own experience.

What do recruiters do with the parsed data?

This is the part that changes how you should think about keywords.

Once your CV is parsed, it becomes rows in a searchable database. A recruiter working a live role rarely reads the pipeline in order. They apply knockout filters, then run a search — often boolean, in the shape of "data engineer" AND (Spark OR Databricks) AND SQL NOT intern — against the indexed text, and review whatever surfaces. Jobscan's own guidance describes exactly this sequence of parsing, indexing and boolean retrieval, and its research puts ATS use at 98.4 per cent of Fortune 500 companies.

So keywords matter, but not because a robot is scoring you out of a hundred. They matter because they determine whether you appear in the result set at all. A CV that says "worked extensively on distributed data pipelines" and never once writes the word Spark will not be returned by a search for Spark. You weren't rejected. You were never retrieved.

This is also why most applicant tracking system resume tips built around "beating the algorithm" miss the point. There is no algorithm to beat. An ATS-friendly resume is one that lands in the index intact, with the right words in it.

What does an ATS-friendly resume look like, section by section?

Copy this shape. The ATS resume format that survives parsing is dull on purpose.

1. Contact block in the body of the document, first thing on page one, never in a header. Name, city, phone with country code, email, LinkedIn URL written out in full. Skip photo, date of birth, marital status and father's name — they add nothing and, in a table layout, actively cause damage.
2. A four-line summary naming your function, years of experience and two or three core technologies or domains. This is prime keyword real estate and it's the one section recruiters do read.
3. Skills, as a plain comma-separated list under a heading that says Skills. Group them by category if you like. No bars, no dots, no star ratings.
4. Experience, under the heading Experience or Work Experience. For each role, on separate lines: job title, then company name and city, then date range as Mon YYYY – Mon YYYY. Then three to six bullets. Reverse chronological.
5. Education, under the heading Education. Degree, institution, year of completion.
6. Certifications, Projects, Publications as needed, each under a heading that says what it is.

Single column. One font. Standard bullet characters. Left-aligned text. No tables anywhere in the document, including for the skills grid.

If you're in India, put notice period and current CTC in the application form fields where they're asked for, not as a decorative box in the CV. If the form doesn't ask, one plain line at the end reading "Notice period: 60 days" is enough.

PDF or Word: which file should you upload?

Send a .docx unless the posting specifies otherwise. Every parser reads Word, and a .docx contains an explicit structural model of the document — paragraphs, headings, lists — rather than a set of positioned glyphs the parser must reassemble.

A PDF exported from Word or Google Docs is close to as good and preserves your layout on the recruiter's screen. A PDF exported from a design tool as flattened artwork is the worst possible file you can send. If you're unsure which you have, open the PDF and try to select a line of text with your cursor. If you can't, neither can the parser.

How do you tailor without keyword stuffing?

Pull the job description into a document and mark every noun that names a tool, a method, a certification or a domain. Ignore the adjectives. Now check which of those you have genuinely done, and make sure each one appears somewhere in your CV in a sentence describing real work — in a bullet, in the summary, in the skills list. That's it.

Use the employer's vocabulary rather than your own where they differ and mean the same thing. If they say Power BI and your CV says "BI dashboarding", you'll lose the search.

The line between tailoring and stuffing is whether a human reading the sentence would find it plausible. A skills list with 60 entries reads as noise to a recruiter and gains you nothing, since retrieval is binary rather than weighted by repetition.

How do you test your own resume?

1. Select all the text in your document. Anything that doesn't highlight — headers, footers, text boxes, images — is invisible to a parser.
2. Save a copy as plain text. Open it. If sections are interleaved or scrambled, your layout has a column or table problem.
3. Copy the whole CV and paste it into a blank email as unformatted text. Read what a recruiter would see after parsing.
4. Apply to one of your own roles through a real ATS if you can, and look at the parsed profile the system builds. Several career sites let you upload a CV and show you the extracted fields immediately.
5. Check that every job has a title, a company and a Mon YYYY – Mon YYYY date range that survived the round trip.

The honest version of how to make your resume ATS friendly is that it takes an afternoon, most of it deleting things. Plain structure, real text, the employer's words for the work you've actually done.

None of this fixes the underlying problem the Hidden Workers research described, where narrow filters exclude people who could do the job. That sits with employers. Some platforms are trying to shift the weight of screening away from resume text altogether — Xakal, for instance, pairs a built-in ATS with structured Xara AI Interviews at thexakal.com, so a candidate's answers count for something before a keyword search decides whether they're seen. Until that's the norm, the formatting is worth getting right. It costs you one afternoon and it is the only part of the process you fully control.