Lever and Resume Parsing: What's Documented, and What's Guesswork

A platform a lot of startups run, and not much public documentation to go with it

If you've applied to a seed-stage startup or a mid-size tech company any time in the last several years, there's a decent chance the application sat on top of Lever. It's one of the more common names that comes up alongside Workday, Greenhouse, and iCIMS in the usual list of platforms applicants try to "optimize" for, and a search for "Lever ATS resume" turns up a surprising number of dedicated guides from resume services and ATS-checker vendors, each one confident about exactly how to format a resume for it.

Lever's own help center does publish some specifics, which puts it closer to Greenhouse's documentation than to iCIMS's thinner public footprint. But most of what circulates about Lever still comes from third parties inferring how the parser behaves, not from Lever itself. Here's what's actually confirmed, and where the rest is a reasonable guess dressed up as a rule.

What Lever's own documentation actually says

It parses text formats, not images. Lever's help documentation states plainly that it can pull structured information out of text-based files like Word documents, PDFs, and RTF files, and that it cannot extract anything from an image file such as a JPG or PNG. That's a narrower, more useful fact than most of the vague "use a simple format" advice floating around, because it gives you something to actually check: does your file have real text in it, or is it a picture of text wearing a document's extension.

There's a built-in way to check for a text layer. The same documentation points applicants toward a simple test: try highlighting a line of text in the document with your cursor. If it highlights normally and you can copy and paste it somewhere else, the file has a text layer a parser can read. If your cursor only selects the whole block as a single image, or nothing highlights at all, there's no text underneath for Lever, or any parser, to extract. It's the same logic behind the copy-paste test already covered elsewhere on this site, just confirmed here as something Lever itself recommends rather than something inferred from general parsing behavior.

Email extraction isn't treated as foolproof, even by Lever. One detail that's easy to miss: Lever's admin documentation includes a setting called "Skip email address parsing," which an employer can turn on for a specific job posting. Turning it on means every applicant has to type their email address into a form field by hand, because the system isn't relying on the parser to pull it correctly. That a platform ships a manual override for one of the simplest possible fields to extract is a small but genuine signal that automatic parsing, even for something as structured as an email address, isn't treated as a solved problem internally.

Outages happen, and Lever has acknowledged them publicly. Lever runs a public status page, and resume parsing has shown up on it more than once: incidents where resumes failed to attach to a candidate's profile correctly, or where an application page returned an error saying it couldn't read a resume, while the file upload itself still went through. These get resolved, generally within a reasonable window, but the point for an applicant isn't the specific timeline, it's that an "unable to read resume" error you hit once is sometimes about Lever's systems that day, not about your file. If the same error shows up across several applications in a short span, that's a reason to check status.lever.co before you spend an hour reformatting a resume that was never the problem.

What's a reasonable inference, not a confirmed fact

Multi-column layouts probably survive without erroring, but lose structure. Several independent guides describe Lever as able to parse resumes with tables or side-by-side columns without the upload failing outright, unlike some older or more rigid parsers that choke on anything outside a single text flow. The catch, according to those same guides, is that the data comes out in the wrong order: a date detaches from the job it belonged to, or a skill from a sidebar ends up spliced into the middle of an unrelated bullet point. Lever hasn't published its own account of how it walks through a multi-column page, so this is an informed pattern across outside testing, not something confirmed in Lever's documentation. The same caution about multi-column resumes generally applies here: a single column removes the uncertainty instead of resolving it.

Word stemming without abbreviation matching. A handful of third-party guides claim Lever's search can match a word to its variants, "manage" to "managed," for instance, but won't connect an acronym to its spelled-out form on its own, so a resume that only says "SEO" might not surface in a recruiter's search for "search engine optimization." That's a specific, checkable-sounding claim, and it's plausible given how a lot of search-and-filter tools are built. It isn't something Lever has confirmed in public writing, though, so the safer move is the one that costs nothing either way: spell out an important acronym at least once if there's room, the same hedge recommended for resume keywords more generally.

AI screening is real, but its role is still narrow and new. Lever has been rolling out AI features, including its own screening tools and at least one third-party integration that scores candidates at a chosen pipeline stage. Lever's own marketing describes these as decision support for a recruiter, flagging or ranking candidates rather than rejecting them outright, and that distinction matters: a flag is something a person still looks at, a rejection is an outcome. Whether that stays true as these tools mature isn't something an applicant can verify from the outside, so it's worth treating "AI decides" claims about any ATS, Lever included, with real skepticism until a platform documents otherwise.

The actual rejection mechanism, most of the time

As with every other platform covered in this series, the single most reliable cause of an instant "no" on a Lever-powered application isn't the parser at all. It's a knockout question: a yes/no or multiple-choice screening question the employer set up, tied to something like work authorization, a required license, or a minimum years-of-experience threshold. Answer one of those in a disqualifying way and the application can close out immediately, well before anything about your resume's formatting becomes relevant. That's the same pattern documented loosely across Taleo, iCIMS, and Workday, and nothing about Lever specifically suggests it works differently.

Where a parse-check actually helps

None of this tells you what your specific resume file contains once it's exported, which is the part you can check yourself before any of it reaches Lever's servers. ResumeReadout runs your actual file through a parser in your browser and shows you the extraction directly, whether there's a real text layer, how your layout holds up outside a single column, and whether your contact details come through as recognizable fields. It won't tell you anything about a specific employer's Lever configuration, since that's not public information, but it will show you the one thing you can actually control: what's really in your file, which is a different and more concrete check than pasting your resume text into a chatbot and asking if it looks fine, since a chatbot never sees the document itself.

The bottom line

Lever publishes more about its parsing behavior than some competitors do, enough to confirm a few genuinely useful, checkable facts: it reads text, not images, it offers a manual override even for something as simple as an email field, and its outages are public record rather than a mystery. Past that, a lot of the "optimize your resume for Lever" content circulating online is informed guesswork, reasonable in places, overstated in others, built on testing nobody has published a methodology for. The advice that holds up regardless of platform still holds up here: export a real, text-based file, keep your core content in one column, and read every screening question carefully, since that's doing more of the filtering than your resume's formatting probably is.

FAQ

Keep reading

• Does ATS Software Really Reject 75% of Resumes?
• ATS-Friendly Resume Format: What Actually Matters (and What's a Myth)
• What Happens After You Submit a Resume Online: A Plain-Language ATS Walkthrough

Or skip the reading and run the free parse-check →