Screening
Your Resume Score Comes From the Tool That Sold It to You
An AI resume checker's score comes from the same company that offers to fix it, a closed loop with no outside check. The applicant tracking systems the checker claims to predict document no such score, so the tool is measuring its own rules: usually keyword overlap with a posting plus formatting choices the checker itself decided on. A 92 says nothing about how a real system will read the file. Fix what you can verify on your own screen; ignore the number and the percentile.
The takeThe score exists because a number is easier to sell against than an explanation, not because a number is what a resume actually needs. A percentile implies a population and a method, and neither is published anywhere a candidate can check. The parts of these tools worth using are the boring parts: did the file open cleanly, is the contact information where a parser expects it rather than buried in a header, did a second column survive the upload. None of that needs a grade attached, and the grade is the part being sold.
Where Olive fits
Open a role and see what the work shows
No screen can tell which resume a model wrote, and Olive doesn't try: if an employer sends you an Olive assessment, it is a 40-to-60-minute assignment done openly with an AI assistant, and you receive the same six-finding report the employer does, free.
Rank your shortlistWhere the Number in the Score Comes From
A resume checker's score has to come from somewhere, and the somewhere is the checker's own rule set. That rule set is written by the company selling the checker, weighted by that company, and changed when that company decides to change it. Nothing in it was copied from an employer's live system, and nothing in it has been tested against one.
Even the closest thing to a studied market here is thin ground to stand on. A 2020 peer-reviewed review of 18 vendors selling algorithmic pre-employment assessments could analyze only what each firm chose to publish, because the models and the employee data behind them are kept private.1 Those are assessment tools rather than the applicant tracking systems that receive an application, and a consumer resume checker is a third thing again: it carries no employer's blessing that its number predicts anything about that employer's process. The score is a product decision, not a measurement borrowed from the systems it claims to describe. A candidate reading a low number has no reliable way to know whether it flags a real, fixable problem, a quirk of that one checker's weighting, or nothing at all beyond where the free version stops and the paid one keeps going.
Why the Score Has Nothing to Be Checked Against
Because the documentation those systems do publish describes no such number. Greenhouse's talent filtering lets a recruiter search job titles, skills and other keywords in the full text of applications, and states that the keyword from the search must exactly match the keyword in the application to appear in the results.2 That is a search a person runs, not a grade anything assigns.
Workday's administrator guide says resume parsing populates fields from a resume, that results can vary based on resume format and order of words, and that resumes should avoid images and image-based styles for best results.3 It puts no figure on how often parsing goes wrong or in which direction, and it publishes nothing a checker's score could be calibrated against. Each page is one vendor's account of its own product and licenses no claim about anyone else's.
Nothing published anchors the number. A checker selling a precise number is answering a question in a form the systems it targets are not documented to ask. Anyone evaluating a vendor's confident claim about what a tool measures hits the wall an employer hits: ask for the method, and whether one was ever published is the whole question.
Fix What You Can Actually Verify Yourself
The useful part of a resume checker is the part that names a specific, checkable problem rather than a score. Greenhouse publishes the list of what breaks its own parser: a file over 2.5MB, letters with spaces between them, graphics or photos, a resume uploaded as an image rather than a document, a columned layout, name and contact details placed in a header or text box, and a job title abbreviated to Sr. Account Exec instead of Senior Account Executive.4
The stated consequence there is worth knowing too: when a parse fails, the resume stays attached to the candidate and someone types the fields in by hand. That is an inconvenience the vendor documents, not a rejection it documents. Every item on the list is also something you can check on your own screen without a tool telling you a number. Open the file as the software would, look for the photo, the second column, or the header hiding your contact information, and fix the specific thing you find, one at a time, rather than chasing a total that moves for reasons you can't inspect.
That is a materially different exercise from chasing a higher score. A checker that tells you your contact details are sitting in a header its parser can't read gave you something to act on. A checker that tells you "73 out of 100" gave you a reason to buy the next tier, and the two sentences can come from the exact same scan of the exact same uploaded file.
Should You Trust the Percentile Against Other Users?
Not without a real comparison population and a stable published method: who was measured, when, against what. A free consumer tool publishes neither, so there is no way to check whether the group you're being ranked against resembles anyone applying to the same jobs you are.
A number that moves when you pay for the next tier is a retention mechanic, whatever else it also is, and it is not calibrated against anything an employer runs. Whether that describes the tool in front of you is checkable in a single pass: change nothing a parser would read differently, apply only the cosmetic suggestions, and watch which way the number goes. Treat a comparison against other users the way you'd treat a leaderboard sitting next to a purchase button, because that is functionally where it sits. Nothing about your standing on it travels anywhere near the employer you are trying to reach.
Use the Posting's Words for Real Things, Not for the Score
Use the posting's exact term where it names something you genuinely did, because in at least one major system a recruiter's search returns only applications where that keyword matches exactly.2 What a checker asks for is a different move: more of the posting's language, in more places, to raise a number. That instruction is the product doing the selling, not a finding about how you will be read.
Matching every word in a posting stopped separating one resume from another once it became something anyone can do in seconds with any free tool, so a checker that raises your score for more of it is rewarding the one move that has already stopped distinguishing anybody. Which system sits on the other end also explains very little of who gets contacted at large US employers: in an audit study of 108 of them, the differences that mattered ran between the companies themselves rather than between the application systems they had bought.5 A score a consumer tool assigned your file was not among the things that study measured at all.
Spend the time the checker asked for on a specific, true sentence about what you actually did instead, and read what a low match score against a single posting is actually worth before letting either number decide anything for you.
A resume that reads well to a person and a resume that scores well on a checker are not always the same document, and when they diverge, the actual person on the other end is the one deciding whether you get a call back.
Common questions
Is a resume checker's score based on how a real ATS would read my resume?
Not verifiably. The major applicant tracking vendors document keyword search and resume parsing rather than a candidate-facing score, so a checker's number has nothing published to be calibrated against and reflects the checker's own rules instead of a tested prediction of any employer's system.
What parts of a resume checker are actually worth using?
The specific, mechanical findings: a file that's too large, contact information sitting in a header, a columned layout, a resume saved as an image. Greenhouse names all of those in its own documentation as reasons its parser fails, and every one is checkable and fixable on your own screen without trusting any score.
Should I pay for the premium version to fix a low score?
Fix the specific, named problems first, for free, on your own screen. The number itself is not a diagnosis, and a premium tier that raises the score without changing anything mechanical hasn't necessarily changed how a real system reads the file.
Does a high score mean I'll get more interviews?
No. A checker measures its own rules against your file, not what an employer did next, and nothing in the number is tested against interview outcomes. What most of these tools reward is heavy overlap with the posting's own wording, and that stopped setting one resume apart from another once every applicant could do it in seconds.
Why doesn't a real employer's system show me a score like this?
Systems built to route applications internally were not designed to hand a candidate a number, and the vendor documentation that is public describes search and parsing instead. A consumer-facing score is a separate product built to be sold, not a window into the employer's own process.
References
- 1. Mitigating Bias in Algorithmic Hiring: Evaluating Claims and Practices arxiv.org Supports the claim that a review of 18 pre-employment assessment vendors could analyze only what each chose to publish, because models and training data are kept private.
- 2. Talent Filtering (Greenhouse Support - Recruiting) support.greenhouse.io Supports the claim that this Greenhouse feature is a recruiter-run exact-match keyword search, not a published score.
- 3. Concept: Resume Parsing (Workday Administrator Guide) doc.workday.com Supports the claim that Workday documents parsing variation by format and word order without quantifying it and without any score.
- 4. Unsuccessful resume parse (Greenhouse Support) support.greenhouse.io Supports the named list of mechanical parse failures and the documented consequence, which is manual data entry rather than rejection.
- 5. Systemic Discrimination Among Large U.S. Employers (NBER Working Paper No. 29053, revised May 2022) nber.org Supports the claim that at 108 large US employers the differences in who got contacted ran between firms rather than between the application systems they used.
5 sources, numbered by first appearance. How Olive sources claims
General guidance for hiring teams. What works at one company and one volume may not transfer to yours.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.