Write a resume that passes AI screening
You have read that 75% of resumes are rejected by an applicant tracking system before a human ever sees them. It is one of the most repeated statistics in job hunting, and there is no evidence for it.
The number traces to a 2012 sales pitch by a company called Preptel, which sold resume-optimisation services. Preptel ceased operating in 2013. No methodology was ever published. Career consultant Christine Assaf, tracing it, concluded the statistic was created without any study, survey or context.
This matters because the myth produces bad behaviour: people stuff keywords, strip formatting into unreadable plainness, and optimise for a filter that does not work the way they think.
What actually happens to your resume
Applicant tracking systems do not autonomously reject candidates. They sort, rank and filter according to criteria a human configured. The software is a filing cabinet with a search box, not a judge.
Recruiter Jan Tegze's estimate, cited in the same analysis, is that 90–95% of applications are reviewed by a human at some point. What gets you cut is usually a human-set rule - a required credential, a location, an employment-gap threshold - not a machine deciding you are unqualified.
That said, 79% of organisations using an ATS now integrate AI into it, so the direction of travel is real. Write for both audiences.
The AI-training platforms are a different case
Here the screening genuinely is AI-led, and it is a different problem. On Mercor and micro1 your CV is not filtered against a job description - it is used to decide what you can credibly be asked about in an AI interview.
That changes what a good CV looks like. You are not trying to match keywords. You are trying to make specific, checkable claims that you can defend out loud for three minutes.
“On these platforms your resume is not a filter to sneak past. It is the script your interview will be built from.”
What to actually do
| Instead of | Write |
|---|---|
| Managed a team | Led a team of six across a nine-month migration |
| Healthcare experience | Clinical trial documentation editing, oncology, four years |
| Expert in Python | Python: built and maintained the ETL pipeline processing 40M rows/day |
| Improved efficiency | Cut review turnaround from nine days to four |
That specificity advice is not ours - it is Mercor's own guidance, which tells applicants to "be specific about domains" and warns that inflated credentials hurt acceptance. Calling yourself an expert at something you cannot demonstrate is worse than not claiming it.
Formatting that survives a parser
- One column. Multi-column layouts scramble reading order in many parsers.
- Real text, never images. A logo or a graphic skills chart is invisible to a parser.
- Standard section headings. "Experience", not "Where I've Made An Impact".
- Dates in a consistent format. Ambiguous dates create the gaps that trip human-set rules.
- PDF unless told otherwise. Preserves layout and parses reliably in modern systems.
- No text in headers or footers. Some parsers skip those regions entirely.
Keywords, honestly
Keywords matter, but not as magic words. Use the actual vocabulary of your field, in sentences that describe real work. A parser looking for "clinical trial documentation" will find it in a sentence about clinical trial documentation. It does not need a keyword-stuffed skills block, and a human reading afterwards will penalise one.
Before you submit
- Every claim has a number, a date or a name attached.
- You can talk for three minutes on any line without notes.
- One column, standard headings, consistent dates.
- Nothing on it you would not want asked about.
- Tailored to one field, not to five.
Then read our Mercor interview guide - the CV you just wrote is what that conversation will be about.
Some outbound links to platforms are referral links.