Gender-Coded Language Decoder for Job Ads
Paste a job advert below. The decoder prefix-matches every word against the masculine- and feminine-coded stem lists published by Gaucher, Friesen and Kay (2011), reports the signed difference D, shows coded-word density per 100 words, highlights each match in place and proposes a neutral replacement. Everything runs in your browser.
Decode a Job Advert
Paste an advert above to begin. Nothing leaves your browser.
Advert with matches highlighted
Hover any highlighted word to see which stem it matched.
Neutral rewrite suggestions
One row per distinct coded word found, with the matched stem and a neutral alternative. Suggestions are a starting point, not a mandate: some coded words are genuinely required by the role.
| Word in advert | Code | Matched stem | Count | Neutral alternative |
|---|---|---|---|---|
| No coded words detected yet. | ||||
How the Decoder Works
Gaucher, Friesen and Kay published Evidence That Gendered Wording in Job Advertisements Exists and Sustains Gender Inequality in the Journal of Personality and Social Psychology (2011, volume 101, issue 1, pages 109 to 128). Across five studies they showed that job adverts in male-dominated occupations contained more masculine-coded wording, and that when women read those adverts they reported lower anticipated belongingness and found the job less appealing. Crucially, the wording did not change how qualified women believed they were. The effect ran through belonging, not through perceived skill.
The paper's appendix lists the two sets of word stems used throughout the studies. This tool implements exactly those lists, and the matching procedure is deliberately simple so that the result is reproducible by hand.
The matching procedure
- Tokenize. The advert is split into word tokens. Letters, internal hyphens and internal apostrophes stay together, so
self-reliantandcandidate'ssurvive as single tokens. - Normalize. Each token is lowercased and stripped of leading and trailing punctuation.
- Prefix-match. The token is tested against every stem in both lists. A stem matches when the token starts with that stem, so the stem
analycatches analyse, analytical and analysis in one rule. - Longest stem wins. If more than one stem matches, only the longest one counts. That is what stops
independentfrom being scored twice, and it keeps every token worth exactly one point on exactly one side. - Score. The hits are summed per side and subtracted.
The verdict scale
The published scale reads off the signed difference D. A larger positive number means a stronger masculine tilt.
| Difference D | Verdict | Reading |
|---|---|---|
| D ≥ 3 | Strongly masculine-coded | Likely to depress applications from women. Rewrite several terms. |
| D = 1 to 2 | Masculine-coded | A mild tilt. Swapping one or two words usually clears it. |
| D = 0, hits present | Neutral | Both codes appear and cancel out. This is the target state. |
| D = −1 to −2 | Feminine-coded | A mild tilt the other way. |
| D ≤ −3 | Strongly feminine-coded | Strong tilt the other way. |
| no hits | No coded words found | Either the advert is very short or it is genuinely plain. |
Why density is shown alongside the raw score
The raw difference grows with length. A 900-word advert can accumulate six masculine hits simply by being long, while a tight 120-word advert with the same tilt scores two. Coded-word density expresses hits per 100 words, so adverts of different lengths become comparable. Turning on density mode rescales the difference to a per-100-word rate before applying the same thresholds, which is fairer for long postings. Be explicit about which mode you quote: density is a normalization added by this tool, not part of the original published scale, and the thresholds above were calibrated on raw counts.
The two stem lists
Every stem is matched as a prefix. Nothing else is added to these lists unless you switch on the jargon extension.
Masculine-coded stems (44)
Feminine-coded stems (40)
The jargon extension
Terms like ninja, rockstar, guru, wizard and hustler post-date the 2011 research and appear in none of its lists. They carry a comparable competitive, individualist signal, so the decoder offers them as an opt-in extension that scores on the masculine side. The toggle is off by default and extension matches are marked with a dashed underline so you can always tell what the original list produced and what the extension added.
What this tool does not do
- No context. The matcher is lexical. "Objective" as a noun in "our objective this quarter" scores identically to "an objective decision maker". Read the highlights before acting on the number.
- No claim about a candidate. The score describes wording in a document. It says nothing about the writer's intent or beliefs.
- Not a compliance check. Coded wording is a nudge effect measured in aggregate, not a legal standard. Neutralizing wording does not by itself satisfy any jurisdiction's equal-opportunity obligations.
- English only. The stems are English. Running translated text through the matcher produces noise.
Using the result well
The most useful output is not the verdict, it is the highlighted advert. Read down the masculine matches and ask, for each one, whether the trait is genuinely required. "Aggressive revenue targets" is often shorthand a hiring manager never intended literally; "makes decisions with incomplete information" is a real requirement that happens to be worded as "decisive". Keep the requirements, drop the decoration. The rewrite table gives you a defensible starting phrase for each hit, and re-running the advert after edits shows the score move in real time.
Want the rest of the writing toolkit?
Readability scoring, passive voice detection, tone analysis and more, all free and all client-side.
Browse all Enhio tools