The job market has always been competitive, but the mechanics of a first-round review have changed. The most-cited figure here is a 2018 eye-tracking study by Ladders, which put the average initial resume scan at 7.4 seconds. It is an old study and it gets quoted carelessly, so treat it as what it is: evidence that the first pass is a scan, not a read.
Add an automated parsing layer in front of that scan and you get the core tension of resume design.
The ATS paradox
You need a resume that works for a machine and a human, and they want different things. Parsers reward clean structure, consistent headings and recognisable section names. Humans reward visual hierarchy, whitespace, and a story that holds together at a glance. Optimising hard for one has historically meant compromising the other.
This tension is narrower than most advice implies, though. Having written the parser behind our own resume checker, the actual structural constraints are a short list: single column, no tables, no image-only pages, standard heading words, consistent date formats. Everything else in visual design is free. Typography, spacing, a restrained accent colour and a well-set header all survive extraction untouched.
So "ATS-friendly" never required an ugly document. It required a document whose reading order is unambiguous.
Beyond templates
Traditional builders hand you a grid of templates and leave you to fill in blanks. Templates are generic by construction: a marketing manager and a software engineer tell fundamentally different stories and end up in the same three-column layout.
The useful shift is starting from content and working outward. A dense project history wants tighter typography and more vertical space; a handful of high-impact roles wants larger type and more air. The layout should follow the material, not the other way around.
Keyword intelligence
Tailoring a resume to each posting is the most mechanical part of applying: read the description, identify the vocabulary, weave it in. It is a matching problem, which makes it a reasonable thing to automate.
What that should mean in practice is narrow. Surfacing skills you already have in the language the employer used. Reordering emphasis so the most relevant role reads first. Flagging genuine gaps between the posting and your experience so you can decide what to do about them.
What it should never mean is generating experience. Our ATS scorer will happily score a fabricated resume at 95, because nothing in an automated pipeline verifies anything. The verification happens in the interview, and that is a much worse place to be found out.
What has not changed
- A recruiter still forms an impression in seconds, and that impression is driven by hierarchy and clarity, not by cleverness.
- Numbers still outperform adjectives.
- Two pages is still the practical ceiling. Our own parser stops reading long documents, and it is not the only one.
- Someone still has to decide which story to tell. That part has not been automated and probably should not be.
The honest summary
AI removed the busywork between your experience and a well-presented document: the formatting, the re-tailoring, the parser-safety checks that used to require knowing how PDF text extraction works. It did not change what makes a resume good.
If you want to see where your current resume sits on the structural side, run it through the resume checker. It reports the specific parsing hazards it finds, by name.