Job description grader:
score your post out of 100.
Paste a job description and get an objective read on completeness, coded language, requirements bloat, readability, transparency and jargon — plus the five fixes worth making first.
Your job post never leaves your browser — the scoring runs entirely on your device. We only use the email you enter to follow up.
What we check
Six dimensions, weighted by how much each one matters. Every penalty is listed on screen with the points it costs.
Weighting: bias and coded language 25%, completeness 20%, requirements bloat 15%, readability and focus 15%, transparency 15%, length and jargon 10%. Each weight is repeated on its row below.
25%
Bias and coded language
Gendered wording measured as an overall lean, plus phrasing that carries legal risk.
Masculine and feminine-coded word stems from Gaucher, Friesen & Kay (2011), scored as an overall lean rather than word-by-word pass or fail — that is what the research measures.
The published list is not applied verbatim. Structural words that happen to share a stem are excluded, and a lean is only reported once at least three matches are present, so "depending on experience" is not counted as gendered language.
Separately, we flag phrasing that carries age, nationality, gender or accessibility risk.
20%
Completeness
Seven sections a candidate needs before they can decide whether to apply.
What the role is, what they'd do, what you need, pay, location and work model, the process, and your inclusion statement. Each missing section costs a fixed number of points, listed on screen.
15%
Requirements bloat
How long and how absolute the requirement list is.
Bullet count, a missing essential / nice-to-have split, years-of-experience thresholds, degree gates with no equivalent-experience route, and stacked "must have" language.
15%
Readability and focus
Reading grade, sentence length, passive voice and how much of the post is about the candidate.
Reading grade is calculated on prose blocks only and suppressed on short posts — see the methodology below. The candidate-to-company ratio counts second-person against first-person references; we report it, we do not claim a threshold for it.
15%
Transparency
Whether pay, work model, contract type and process are stated.
A missing pay range triggers a dedicated note on the EU Pay Transparency Directive (2023/970), whose transposition deadline passed on 7 June 2026.
10%
Length and jargon
Word count against LinkedIn's published apply-rate bands, and how much filler is in the prose.
The word-count band is the one rule here with published apply-rate evidence behind it. Alongside it we count buzzword density, undefined acronyms and stock opening lines — those are editorial judgement, not measured effects.
Methodology
Where the rules come from, and what they deliberately leave out.
How the score is built
Each dimension starts at 100 and loses points against a published rule. The six are combined on the weights above to give the composite. Nothing is inferred, weighted by a model, or adjusted after the fact.
Where we're conservative
Reading grade is calculated on prose blocks only — two or more sentences, no leading bullet — because Flesch-Kincaid is calibrated for prose and reports nonsense on a bulleted list. Below 100 prose words it is suppressed rather than guessed. The gendered-wording list is applied with an exclusion list for structural words and a three-match minimum before any lean is reported. Where a rule uses bands, the penalty is capped at the band, not scaled indefinitely.
What this isn't
The tool flags risk indicators in language. It is not a legal compliance check and does not constitute legal advice.
Sources
- Gaucher, D., Friesen, J., & Kay, A. C. (2011). Evidence That Gendered Wording in Job Advertisements Exists and Sustains Gender Inequality. Journal of Personality and Social Psychology, 101(1), 109–128.
- Seong, J. et al. (2024). Replication and extension of Gaucher, Friesen & Kay (2011). Strategic Entrepreneurship Journal.
- LinkedIn Talent Blog — job post length and apply rates.
- Directive (EU) 2023/970 on pay transparency; transposition deadline 7 June 2026.
Questions, answered.
No. Every dictionary, rule and calculation runs in this browser tab. There is no server call to produce a score, so nothing to intercept, log or train on. Text is transmitted only if you explicitly ask us to email you the report — and that says so at the point you click.
No. It is a rules engine — word lists, regular expressions and arithmetic. That means it costs nothing to run, can't hallucinate, can't be rate-limited, and gives the same score for the same text every time. It also means it will miss nuance a human would catch. Treat the score as a structured second opinion, not a verdict.
The gendered wording lists are the masculine and feminine-coded stems published in Gaucher, Friesen & Kay (2011) in the Journal of Personality and Social Psychology — the standard academic source, replicated since. We report the overall lean rather than calling individual words good or bad, because that is what the research actually measures.
Because the report is designed to be shared with the person who wrote the post, who is usually the hiring manager. A damning score doesn't get forwarded, and a report nobody forwards changes nothing. The bands go down to "significant issues" and stop there.
No, deliberately. It gives you single-word swaps from a fixed dictionary where a clean one exists, and tells you what to change. Rewriting is a different job — that's what Cara's job post generator does.
Not yet. We won't publish medians until we have graded a large enough corpus to make them meaningful. Anything else would be a made-up number, and this tool only works if the numbers on it are defensible.
This tool tells you what’s wrong. Cara rewrites it. The job post generator in Cara takes your role, your brand voice and this diagnosis, and produces the post you meant to write.
See Cara’s job post generator