How Lever Reads Your Resume, and Where Its Own Pages Disagree

Lever’s technical brief for Talent Fit, its AI resume matcher, says the feature returns “a binary decision” on whether a candidate is a suitable fit. A Lever blog post from June 2025 described the same feature as producing a “ranked list of top candidates” with “transparent scoring”. We read both, and they don’t describe the same thing.

The other thing that struck us is in Lever’s help center. In its own software, the automatic rejection Lever documents is a rule the employer builds, and none of the rule’s documented conditions reads the text of a resume. Tags are among those conditions, though, and the same help center carries a partner article for another company’s tool that scores each resume from 1 to 10 and tags the application with the result.

Lever is hiring software owned by Employ, whose website lists Lever, Inc. among its U.S. subsidiaries. The employer runs it and chooses the settings, and Lever’s article for applicants says the company “plays no part in hiring decisions”. Everything here comes from Lever’s own public pages (help center, developer documentation, website, security page and blog) and from Employ’s, as we read them on October 9, 2026. Nearly all of it is written for employers, so it describes what an employer can set up, which is not the same as what happened to your application. Offermic is not affiliated with Lever or Employ.

Talent Fit, described two ways

Talent Fit is Lever’s own feature for evaluating a resume with AI, and a technical brief in the help center sets out how it works. The resume is anonymized first. A large language model, told to act as a recruiter, then compares it with the job description, and the brief says those two documents are all Talent Fit uses to decide whether a candidate is a fit. The output is a binary decision, with an explanation of the candidate’s strengths, considerations and areas for clarification. The brief adds that “each candidate is evaluated individually, with no consideration of the other candidates in the pipeline”.

The June 2025 blog post tells it differently. There, Talent Fit “instantly ranks candidates against your job requirements” and gives the recruiter a shortlist “complete with transparent scoring and explanations”. Scoring turns up in the help center too, where an article about a job where the feature is paused refers to “the Talent Fit scoring for that job”.

For your resume, the brief’s account has a practical consequence. The tool, it says, “only evaluates education, skills, and experience that are called out as necessary by the job description”, with no formula or set of boxes a candidate needs to tick to be considered a fit. Its examples of areas to clarify are a construction project manager who didn’t list the size of their projects and a school bus driver who didn’t list their type of driver’s license. A requirement you meet but didn’t write down is, by that description, invisible to Talent Fit, so answer the posting’s stated requirements in the resume itself.

Whether any of this touches your application is up to the employer. The brief says customers must turn Talent Fit on, and the help article adds that as of August 4, 2026, it is turned off by default for all newly created jobs. Lever’s Spring 2026 release notes also list admin settings that enable it for all jobs or for newly created jobs, so an employer can make it standard.

Inside Lever, a matched candidate is labeled with Talent Fit next to their name, and hiring team members with access to the candidate can see the summary pinned at the top of the candidate page. We found no Lever page that says the candidate is shown the result, and the Talent Fit pages describe no way for a candidate to decline the evaluation. On what the result is for, the help article’s FAQ says: “Talent Fit is not intended to filter out candidates, but to provide additional context on their background and highlight areas that require clarification.” Lever’s AI features page says: “Recruiters always have the final say in decision-making.”

What a candidate may see is a notice. Lever says every customer that uses any of its AI features automatically has an AI disclaimer on the job posting and career site, with a default message the employer can rewrite. The brief says customers bear the responsibility of informing candidates that AI is being used to evaluate their applications.

Talent Fit automatically reprocesses candidates if the job description or their resumes change. Lever also says the feature is automatically disabled for a whole job when it meets “blank resumes, heavily redacted resumes, or resumes with insufficient content for analysis”, until the employer opts in to showing partial results.

On how well it works, Lever’s Spring 2026 release notes are direct. Since the release in the first quarter of 2025, they say, customers reported “too many matches, not enough matches, or just not relevant to the job description”, and Lever upgraded the model and updated the prompt in response. A separate article describes a daily bias test in which resumes are evaluated twice, with and without words that identify gender, race or military status. For the one day it shows, “Talent Fit made the same recommendation (match or no match) 91% of the time”.

The prompt, the model in use and what separates a match from a non-match are not public, and neither is the content of the “Matching Logic” option that the help article says some postings have. The brief does say that the models are external and closed-source, that Employ does not train them on hiring outcomes, and that “Data is anonymized before leaving Employ’s environment”.

Before Talent Fit existed, a Lever blog post from December 2021 said that “artificial intelligence isn’t an ATS feature that can or should make important decisions regarding who talent teams should or shouldn’t move forward with”, and it named “auto-archiving unqualified individuals” among the ways AI can help. We set these statements next to those of other vendors on our resume builder page.

Knockout rules, and the tags other tools write

Lever’s word for rejecting is archiving, and its help center says so directly: “Archive or Archiving also means to reject or to be rejected.” The automatic version is an automation workflow, which a Super Admin at the employer, or a user given the automation permission, builds from a trigger, conditions and actions. One trigger is “Application form submitted”, and one action is Auto-archive. Lever’s article on the archive says: “This feature is commonly used to auto-archive candidates based off of knockout questions in the application form.” Its published knockout recipe uses a right-to-work question as the example, archives with a reason such as visa status, and sends a rejection email on the next weekday.

The conditions Lever documents for a workflow are these:

• the opportunity (Lever’s name for one application to one job): archived or not, archive reason, tags, stage, origin and sources
• the application form: specific questions and the answers to them
• the job posting: department or team, hiring manager, owner, the posting itself, location and work type

Nothing on that list reads the text of the resume, and nothing on it refers to a Talent Fit result. So read every application question as a possible gate. The automatic archive Lever documents in its own software runs on those answers.

Tags are on the list, which is where the picture gets less tidy. The Talent Fit article doesn’t say whether the label it puts next to a matched candidate’s name is a tag a workflow can use. Lever’s help center also has partner integration articles, which are setup guides for other companies’ tools that an employer can connect, and some of those tools score resumes. The ZYTHR one says the tool “analyzes each resume against the job description and your customized criteria” to assign a match score from 1 to 10, and that it adds a match tag to the opportunity in Lever. The Senseloaf one describes an agent that moves applicants who do not pass its prescreening questionnaire to a stage named Knockout. Workflow conditions include tags and stages. Lever does not document a workflow built on a resume score.

Two paid add-ons that screen applicants sit in Lever’s own AI settings, both marked as costing extra. With AI Screening by VONQ, the application form shows a consent checkbox above the submit button. Tick it and you are taken to a chat stage and a voice stage after you submit. The consent text says: “The tool may generate a score, ranking, summary, recommendation, or similar output for recruiter review and is intended to support, not replace, human review.” It also says your application will still be submitted if you do not complete AI Screening. The other add-on, Candidate Insights powered by ProboTalent, emails assessments automatically to each new applicant when it is enabled on a job, and the scores are posted to the candidate profile.

Then there is Fraud Signals, which Lever describes as flagging “suspicious applications at the top of the funnel” with checks on the email address, the phone number, the web presence of the companies in your work history, and application volume. Lever documents a button for the recruiter to tag a candidate as suspicious. It documents no automatic rejection by Fraud Signals.

A rejection doesn’t have to come with a message. An email on archiving is a checkbox for the recruiter and an optional step in a workflow. An employer that enables Candidate Transparency can have AI draft personalized feedback for a rejection email, one candidate at a time.

Seven file types and a highlight test

Lever’s parsing article lists the file types it can parse:

• Microsoft Word (including .docx)
• Adobe PDF
• Rich Text (RTF)
• Corel WordPerfect
• HTML
• MS Office HTML
• Open Office (ODF)

The article says Lever cannot parse information from image files such as JPG or PNG. Its test is to try highlighting text in the document with your cursor. “If you cannot highlight text, the document is likely not parseable.” The safe course is a text-based Word or PDF file that passes Lever’s own test.

Plain text is a loose end. A second article, on adding resumes to a candidate profile, says the resume uploader “supports most common file types for resumes, such as .doc, .docx, .pdf, and .txt”. Plain text is accepted there and missing from the parseable list, and Lever does not say whether a .txt resume is parsed.

According to the parsing article, the parser adds name, organization and contact details to the candidate profile. The developer documentation lists more: headline, emails, phones, location, company and links, “although no information is guaranteed to be extracted”, plus positions and schools with their start and end dates. Neither the help center nor the developer documentation names the parsing technology or gives an accuracy figure, and neither says how columns, tables, text boxes, headers or graphics affect parsing. The highlight test is the only guidance.

Location is a documented weak spot. If your resume gives a region rather than a city (Lever’s example is “New York Metropolitan Area”), the parser matches it to the closest location on its list, and the parsing article warns: “This can cause locations to be incorrectly attributed.” That’s a reason to write your location as a city.

On a Lever-hosted application form, some of the result is in front of you before you submit. Lever’s article on career site options lists “Auto-parse resumes to pre-populate applications” for the hosted form and says a custom-built form “does not support” it. The hosted form’s built-in fields are Resume, Full Name, Email, Phone, Current Location and Current company, so check those. Current Location is filled from your resume, and you can select a location manually if the resume is not provided or the location cannot be parsed. Lever’s help center does not describe a screen where the applicant reviews the parsed work history. After you apply, the parsing article says, the recruiter can type information into the profile fields.

The recruiter has the original file as well as the parsed fields. Resumes attached to applications are stored automatically in the resume section of the candidate profile, and Lever says “resumes are visible to all candidate followers and interviewers associated with the interview plan”. Its screening view, Fast Resume Review, shows the resume and candidate details for each new applicant, with the answers to application questions one click away.

Since the interviewers on the interview plan are shown the file, be ready to talk through any line of it. Our guide to resume-based interview questions covers what interviewers ask from a resume, and the resume interview questions tool lists the questions your own resume is most likely to raise.

The parsed text feeds search. Lever’s search covers “any parseable content attached to candidate profiles, including resumes, notes, and feedback forms”. Recruiters can also filter by company, job title, school and years of experience, and the advanced search article states: “Years of experience are calculated based on parsed dates from the candidate’s resume.” By that description, a resume with no parseable text adds nothing to those searches and filters. It is also a reason to put plain start and end dates on every role and school. The same article lists a Skills filter without saying where the skills come from.

One email address, one history

Lever joins your applications by email address: it says it “automatically prevents duplicate candidate profiles when a candidate applies to multiple postings using the same email address”. Each application is its own opportunity, and the opportunity holds the resume you provided with it. The profile’s history is “a record of activity for all opportunities linked to the candidate’s profile, including Nurture campaigns, interviews, offers, feedback, and notes”. Someone who opens your new application can therefore see the earlier ones and why each was archived, subject to their access rights. So use the same email address for every application to one employer.

Applying to the same job again is governed by an employer setting, “Block repeat applications”, which Lever says is disabled by default and applies to all postings once it is on. The employer picks the waiting period, and the default selection is 6 months. Lever identifies the repeat by email address and shows a page stating when you last applied and when you can apply again. The block applies whether your earlier application is active or archived. It is tied to the posting’s ID, so a reopened job with a new posting is not covered. Applications submitted via Lever’s API are not blocked, and neither are applications to other postings.

After you apply, you deal with the employer

The confirmation email on a Lever-hosted job site is a toggle the employer turns on. To withdraw an application, Lever’s article for applicants tells you to contact the company you applied to directly. Lever’s security page says Employ “has no direct relationship with the individual employees and jobseekers whose personal data it processes on behalf of our customers”, and that requests to correct, amend or delete inaccurate data in an application go to the employer, who “is responsible for complying with the individual’s request.”

The same page sets no fixed retention period for candidates. Customers “can configure data retention policies specific to their needs”, and otherwise “customer data in the Lever by Employ Platform is stored for the duration of the service contract between the customer and Employ”. Where an employer has set up GDPR or a local data compliance policy, Lever documents consent and unsubscribe links that open a data preferences page, and there a candidate can ask to view, update or remove their data. Lever also notes: “Candidates do not receive a notification when their data is anonymized.”

What you can’t see from outside

Nothing public shows whether a given employer has Talent Fit, knockout workflows, the repeat-application block or Fraud Signals turned on. Knowing that a company uses Lever tells you which of these could be in play, and that is as far as the public pages go.

Lever’s documentation also describes no resume score that a candidate can look up. The score in our resume builder is our own measure, with its rules published on that page. It is not a score from Lever or any other applicant tracking system, and no employer sees it.

We read the public documentation of four other systems the same way, and the same questions have different answers there: Workday, Greenhouse, iCIMS and Ashby. As for the claim that software rejects 75% of resumes, we wrote up the oldest sources we found for that number separately.