Methodology
What we measure, how we measure it, and what we cannot know.
Most tools in this category show you a percentage and leave you to guess what it means. This page is the opposite: the actual method behind every number inteller.ai shows you, the weights it is computed from, and an explicit list of the questions we cannot answer. If you only read one section, read what we cannot know.
The number is not the product
A match score cannot tell you whether you will get an interview, and you should be suspicious of any tool that implies otherwise — including this one, if we ever start talking that way. What a scan can honestly do is name the specific things the posting asked for that your resume never demonstrates: the stakeholder management you have done but never wrote down, the SQL that is on your resume only as a tool name, the forecasting experience that is missing entirely. That list is actionable. The percentage on top of it is just a way of sorting the list.
Layer 1 — is this posting real?
This layer is deterministic. It runs in well under a second, uses no AI, and produces the same answer for the same text every time — which matters, because a fraud verdict that changes between runs is not a verdict.
The posting is checked against more than 30 documented fraud and ghost-job patterns: requests for money, bank details, or identity documents before an interview; contact moved to a personal messaging app; a salary that is absent or implausible for the role; evergreen and “always accepting applications” language; responsibilities vague enough to describe any job; and requirement sets no real candidate could satisfy. Each pattern that fires is shown to you individually, with the phrase that triggered it, so you can disagree with any one of them.
The same pass runs the culture decoder, which translates specific stock phrases into what they commonly signal — “fast-paced environment” as a marker for chronic understaffing, “we’re like a family” as one for blurred boundaries. These are pattern-matched phrases with a fixed reading, not an AI interpretation of the employer.
Layer 2 — which requirements have you not demonstrated?
This is the part we consider the actual product. Rather than scoring your resume against a bag of keywords, we extract the posting’s individual requirements and check each one against your resume separately. Every requirement is classified by category (experience, skill, domain, education, certification, logistics), by importance (must-have or preferred), and by verdict — covered, partial, or missing.
Every claim has to cite its evidence
When the model says your resume covers a requirement, it must quote the line that proves it, word for word. We then search your actual resume text for that quote. If it is not there, the claim does not stand: a covered verdict is demoted to partial and the citation is stripped.
This is the one place in the product where we deliberately overrule the model’s own confidence. An invented quote — telling you that you demonstrated something you did not — is the single most damaging output this feature could produce, so it is the one thing we verify mechanically rather than trust.
The coverage figure is arithmetic, not opinion
The model classifies requirements. It is never asked for a score. The coverage percentage is computed in our code from the verdict list, with two fixed weights:
- Importance: a must-have requirement counts three times as much as a preferred one.
- Verdict: covered scores 1, partial scores 0.5, missing scores 0.
Those are the only policy numbers involved, which means any figure we show you can be recalculated by hand from the requirement list on your screen. If our arithmetic and yours disagree, ours is wrong and we want to hear about it.
About the ATS score specifically
We do publish an ATS score, because it is genuinely useful for ranking your own resume versions against one job and because it is what people search for. It is built from a database of 200+ skills with weighted categories and alias handling, so that “Postgres” and “PostgreSQL” are not counted as different things, and it is always computed against a specific posting — a resume score with no job attached measures nothing.
Do not treat 80%, 90%, or 95% as a prediction — from us or from anyone else. A high score means your resume covers what this posting asked for. It does not mean you are a strong candidate, that the job is real, or that a human will read it. Use the score to choose between your own resumes; use the gap list to decide what to change.
What we cannot know
Every one of these is a real limit of the method, not a feature we have not shipped yet.
That a job is definitely real, or definitely fake.
We read the posting's text. A careful scammer can write a clean posting, and a legitimate company can write one full of the same signals we flag. Our verdict is a probability shaped by the wording, not a verification of the employer. We say "this posting shows N of the patterns associated with ghost listings" — never "this company is not hiring."
What a specific employer's ATS actually scores you.
We have no access to Workday, Greenhouse, Taleo, or any other applicant tracking system. Nobody outside those companies does. Any tool showing you "your Workday score" is modelling, not measuring — and so are we. Our score reflects how well your resume covers this posting's stated requirements, which is a different question.
Whether you will get an interview.
Interview decisions turn on referrals, timing, internal candidates, budget, and a recruiter's judgement — none of which appear in the job description. A resume can cover every requirement and still not get a call. We can tell you what the posting asked for and whether you demonstrated it. That is the whole claim.
How long a posting has really been live.
We analyse the text you paste, not a longitudinal record of the listing. Repost age is one of the strongest ghost-job signals available, and we currently infer it only from language in the posting itself. We would rather state that limit than imply a data source we do not have.
Whether a salary figure is what they will actually pay.
We extract the range, currency, and pay period the posting states and normalise it for comparison. If the posting omits pay, we flag the omission — we do not estimate a number and present it as theirs.
Where our published numbers come from
Statistics we cite in marketing and on the blog are attributed to named third parties. We do not publish first-party statistics from our own scan data yet — our scan volume is not large enough for that to be honest, and we will say so on the day it changes.
- Ghost-job prevalence (~43% of postings): reporting from NPR, Forbes, and ZipRecruiter research.
- Job-scam losses: US Federal Trade Commission consumer reports and the FBI's IC3 annual report.
- ATS usage among large employers and resume filtering rates: Harvard Business School's "Hidden Workers" (2021) and Jobscan's ATS usage research.
- Employer detection of AI-generated applications: published survey work on recruiter screening practices.
Questions about the method
Is the ATS score a prediction of whether I will pass a real ATS?
No, and you should not treat any tool's score that way — ours included. Our score measures how much of this posting's stated requirements your resume demonstrably covers. It is a measure of coverage, not a forecast of an outcome. The useful output is not the number; it is the list of requirements you have not evidenced.
Does an AI write the score?
No. The coverage figure is computed in our own code from the verdict list, using fixed weights — a must-have requirement is worth three times a preferred one, a covered requirement scores 1, a partial 0.5, a missing 0. The model classifies each requirement; it is never asked for the score. That means any coverage figure we show can be recalculated by hand from the requirement list on screen.
How do you stop the AI from inventing evidence?
Every time the model claims your resume covers a requirement, it must quote the line that proves it, verbatim. We then search your actual resume text for that quote. If the quote is not there, the claim is demoted — a 'covered' verdict drops to 'partial' and the citation is removed. An invented quote is the most damaging thing this feature could produce, so the model's confidence is overruled in exactly that one place.
What happens if the analysis fails?
The requirement panel does not render, and the scan still returns its fraud check and keyword coverage. We would rather show you less than show you a number we cannot stand behind.
Do you sell or share my resume?
No. Your resumes and scans are stored against your account under row-level security and are not shared with employers, recruiters, or other users. You can export or delete your data at any time from Settings.
Run it on a posting you are unsure about
Free plan, no card. Judge the method on a job you already have an opinion about — that is the only fair test of a tool like this.
Try inteller.ai freeSee what each plan includes on our pricing page.