Biased Language in Job Descriptions: 2026 Audit of 39,000+ JDs

Biased language in job descriptions is close to universal in 2026: nearly 9 in 10 postings (88.6%) contain at least one masculine-coded word, yet that wording moves candidate behavior far less than most advice assumes. That’s the headline from Pin’s review of 39,000+ real JDs that recruiters wrote and ran between January 2024 and October 2026, paired with candidate reply data from 3,000+ of those jobs. Length, readability, and the requirements a posting lists turned out to be bigger levers than adjectives.

We’re not alone in that conclusion. Emilio Castilla and Hye Jin Rho’s Management Science study tracked about 300,000 postings and 485,000 job seekers and found that gendered wording effects were “statistically significant, but very small” (MIT Sloan, 2023). None of this clears anyone to write “rockstar ninja wanted,” since age-coded wording carries legal risk under EEOC rules and credential creep still screens out qualified people. What follows separates the bias that costs you candidates from the bias that mostly costs you credibility.

The short version:

• Structure beats adjectives. Across 39,000+ job descriptions, length and requirements predicted candidate replies better than gendered wording did, so fix those first.
• Gender-coded words are everywhere. 88.6% of JDs carry a masculine-coded term and 88.4% carry a feminine-coded one, mostly generic words like “lead” and “support.”
• The reply-rate link to gendered wording is fragile. Under one word list, masculine-leaning JDs looked better; under a stricter list, feminine-leaning JDs did. That flip matches the “very small effect” finding from Castilla and Rho (2023).
• Short JDs cost replies. Postings under 300 words drew 10.1% candidate replies versus 14.1% for 300-599 words, a 28% gap that held when comparing JDs from the same company.
• Pay disclosure didn’t change replies, but it’s rising fast. Explicit pay figures climbed from 17.4% to 31.5% of JDs in two years, with no reply-rate penalty.

What Counts as Biased Language in a Job Description?

Biased language in job descriptions is any wording that signals a posting is meant for one group, whether by gender, age, disability, or background. The 2011 study by Danielle Gaucher, Justin Friesen, and Aaron Kay found women rated masculine-worded ads as less appealing (4.16 vs 4.50 on a 6-point scale) and expected to belong less (Harvard Kennedy School Gender Action Portal, 2011).

Five categories show up in real postings, and they carry very different stakes:

Only some of these categories measurably change who applies, and only some create legal exposure, so treating “leadership” and “recent grads only” as the same problem wastes editing time on the word that matters least.

The pattern we keep seeing is that recruiters spend their JD-editing energy on adjectives because adjectives are easy to change. A word checker flags “competitive,” the recruiter swaps in “motivated,” and the posting feels fixed. Yet it still lists a degree requirement the hiring manager would waive in a heartbeat, along with a five-year experience floor nobody can defend. Boilerplate fills three paragraphs before the role is even described.

Having built Interseller before Pin, our team watched this play out across years of outbound campaigns: candidates who never answer usually decided the role wasn’t for them based on the requirements, not the tone. This study tested that hunch against real reply data. It held up for length and requirements, and it held up for gendered wording in a direction most people wouldn’t anticipate.

Vocabulary matters a little. Structure matters more.

How Common Is Gender Bias in Job Descriptions in 2026?

Gender bias in job descriptions, measured by the Gaucher word list, is close to universal: 88.6% of the 39,000+ JDs we reviewed contain a masculine-coded term and 88.4% contain a feminine-coded one. That matters because the application gap is real: after viewing a job, men apply 13% more often than women (LinkedIn Talent Solutions).

Those near-identical shares hide what’s driving them, and the most frequent masculine-coded words were “lead” (38.5% of JDs), “leadership” (36.5%), and “competitive” (22.0%). On the feminine side, “support” appeared in 52.6% and “responsibilities” in 51.1%, because “responsibilities” matches the Gaucher stem respon-, which shows how blunt a stem list can be.

To test that bluntness, we ran a sensitivity check: strip out the four near-universal stems (lead, analy-, respon-, support), and the male-coded share drops to 69.7% while the female-coded share drops to 68.7%. Net lean barely tilts: 35.2% of postings lean masculine, 36.0% lean feminine, and 28.7% come out neutral.

Here are the same shares side by side, plus the neutral-lean figure for each definition:

Anyone comparing this to Pin’s May 2026 study of 29,000+ JDs, which put masculine-coded language at 49.5%, should know that study used a shorter list of high-signal words. Here we used the full Gaucher stem list, which catches far more generic vocabulary. Both numbers are right for what they measure, and the gap is the point: a “biased JD” rate depends almost entirely on the word list behind it.

Blatant examples are considerably rarer than the internet suggests, since bro-culture jargon showed up in just 1.3% of JDs, led by “hustle” at 0.39%, then “unicorn” (0.29%) and “rockstar” (0.19%).

Ninjas, evidently, are mostly retired.

Does Gendered Language in Job Descriptions Change Who Responds?

Gendered language in job descriptions has a measurable but small effect on applicants, and Pin’s reply data can’t separate it from noise. In their follow-up experiment, Castilla and Rho found the effects “so small as to have little practical impact” (INFORMS, 2023).

Across 3,000+ JDs that each had at least 50 candidate outreach messages behind them, the full word list made masculine-leaning postings look like winners: 14.9% of candidates replied, against 12.4% for neutral ones. Switching to the stricter list flipped the picture entirely, as masculine-leaning JDs fell to 12.7%, neutral came in at 13.0%, and feminine-leaning JDs rose to 15.0%.

Identical postings, identical candidates, opposite conclusions.

When a result reverses because four generic stems were dropped, it isn’t telling you about gender; it’s telling you about the word lists researchers and software vendors happen to choose.

Inside individual companies, the stricter-list version showed a modest lift for feminine-leaning phrasing (words like “collaborative” and “supportive”), which survived some controls but not others. We’d call it suggestive, not something to rewrite a JD around.

The strongest recent field evidence lands somewhere in between. Joyce He and Sonia Kang studied 576 postings and 32,834 applicants at a Canadian employer and found more masculine wording went with fewer women applying. Women made up 40.0% of applicants at one standard deviation below the mean, against 34.1% one deviation above (PNAS, 2025). When a second employer debiased one male-dominated posting, its share of female applicants rose about 4% and stayed up for 10 months. The authors describe the effect as “modest,” and it persisted well after the edit. Real, measurable, and smaller than the requirements list: that’s the honest summary.

One limit is worth stating plainly before going further. Because Pin’s matching feeds zero demographic data to its AI, Pin doesn’t store candidate gender, and we didn’t infer it from names, so we couldn’t split replies by gender. That leaves the gendered-wording question where the best public research leaves it: lab studies (Gaucher et al. tested samples of 43 to 118 students) found real effects on perceived belonging, while large field studies find small ones. A replication from the University of St Gallen found the effect held for startups but not established firms (University of St Gallen).

What should a recruiter actually do with all of this? Keep the obvious exclusionary terms out, balance “competitive” with “collaborative,” and stop there. Spend the saved time where the evidence shows a bigger payoff, which is who the posting reaches, since a JD can only shape the pool that sees it. For teams trying to track applicant-pool diversity by stage, outbound sourcing changes the starting pool far more than a word swap can.

Which Job Description Wording Creates Legal Risk?

Age-coded wording is the JD bias with the strongest causal evidence and the clearest legal line. Federal regulation 29 CFR 1625.4 says ads using terms like “young,” “recent college graduate,” or “age 25 to 35” violate the Age Discrimination in Employment Act unless an exception applies (U.S. Government Publishing Office, 2019).

Peer-reviewed research backs the regulation: a field experiment by David Neumark and coauthors found ageist stereotypes in job ad language predicted discrimination against older applicants. The evidence was strongest for men (NBER, 2019; published in the Journal of Labor Economics, 2022). It’s a considerably tighter link between advertisement wording and documented hiring outcomes than any of the gender studies demonstrate.

It happens less than you’d fear, since 1.8% of postings we reviewed used clear age-coded terms like “recent grad,” “digital native,” or an experience cap. We excluded “young” and “fresh” from that count because they usually describe a “young company” or “fresh ideas,” but including them lifts the share to 3.4%.

Physical requirements show up in 4.9% of JDs overall, rising to 15.5% in operations, trades, and industrial roles, which makes sense, and falling to 2.5% in software, data, and IT roles. The EEOC’s ADA guidance warns against listing something like “ability to manually lift and load 50 pound cartons” unless it’s a genuine essential function of the job (EEOC). Whenever a software JD carries a lifting clause, it’s almost always template residue and a two-minute fix.

Credential requirements are the quieter screen: 5.8% of JDs required a degree with no equivalency option. Among postings that mentioned a degree at all, 38.5% already offered an “or equivalent experience” path, which suggests many hiring teams have absorbed the skills-based hiring argument. The wider market moves slowly: Burning Glass Institute and Harvard Business School found only 3.6% of roles at large firms dropped a degree requirement, changing fewer than 1 in 700 hires (Higher Ed Dive, 2024).

Then there’s the stat everyone quotes, that men apply when they meet 60% of the qualifications and women only at 100%. It traces to an unpublished Hewlett-Packard internal report cited by Tara Mohr in HBR (2014), and the report itself has never surfaced. Mohr’s own survey pointed to a more useful finding: people skipped jobs because they believed they wouldn’t be hired without every listed requirement. Shorter requirement lists, not softer adjectives, are what that finding argues for.

What Is the Ideal Job Description Length?

The ideal job description length is 300 to 600 words, because JDs below 300 drew 28% fewer candidate replies in Pin’s data and gains above 600 faded within the same company. LinkedIn’s earlier study of job posts found 1-300 word posts earned 8.4% more applications per view (LinkedIn Talent Solutions), so the answer depends on which funnel you’re filling.

These findings don’t actually conflict, because LinkedIn measured inbound applies from people browsing a job board, whereas we measured candidates who received recruiter outreach and then read the JD before deciding whether to respond. Passive browsers want a fast scan; a sourced candidate, who didn’t go looking, wants enough detail to justify a conversation.

Below are the same reply rates alongside the share of postings that fall in each length band:

Read that chart carefully, because its right-hand bars oversell. Longer postings did better on raw numbers, but comparing JDs from the same company erased the advantage above 600 words. Lengthier descriptions tend to belong to senior roles, and senior candidates respond more regardless. Under-300 postings are different: their penalty held up inside the same company, which is as close as observational data gets to isolating the JD itself.

With a median of 404 words, roughly a third of real postings still sit in the zone that costs responses.

Aim for 300 to 600 words: long enough to describe the work, the team, and the pay, short enough that nobody scrolls past the point. If you need a starting point, these role-by-role job description templates sit in that range.

Should Job Descriptions Be Written at an 8th-Grade Reading Level?

Job description readability matters, but “write at an 8th-grade level” is the wrong rule for most professional roles. 72.2% of JDs we reviewed read above a 12th-grade level (median Flesch-Kincaid grade 13.4), while 28% of US adults score at literacy Level 1 or below (NCES, 2023).

On paper, that gap argues for simpler writing everywhere, but candidate behavior says otherwise. Postings scored at grade 12 or below drew 12.0% replies, while grade 12-14 drew 14.7% and grade 14-16 drew 14.9%. Only past grade 16 did responses decline again, to 12.6%. Even within the same company, the grade 12-16 bands beat the simplest band by about 20% on the odds of a reply.

Why would plainer text underperform? Partly it’s overlap with length, because very simple JDs are often very short ones. Partly it’s substance: a JD that names the tech stack, the reporting line, and the problem the team owns scores as “harder” on readability formulas, because specific nouns have more syllables than vague ones. Candidates evidently reward those specifics with responses.

Write short sentences, use specific nouns, and don’t strip out the detail that makes a role legible to the person who’d actually do it. Since readability formulas punish jargon and bullet lists, treat any score as a smoke alarm rather than a grade. Hourly and entry-level positions, where the literacy gap with the applicant population is widest, deserve a lower target; for a staff engineer, precision beats simplicity.

Does Salary Transparency in Job Postings Still Pay Off?

Salary transparency in job postings is now the norm in the US market: 59% of Indeed postings included pay as of May 2025, and postings with employer-provided salaries received 3.8x more applications (Indeed, 2025). In New York, where a law requires it, 97.7% of postings showed pay (Indeed Hiring Lab, 2024).

Recruiters on Pin are moving the same way: 46.6% of JDs mention pay somewhere in the text, and 28.8% publish an actual number or range, a share that climbed from 17.4% in Q4 2024 to 31.5% in Q3 2026.

Legislation is doing much of that work, because more than a dozen states now require pay ranges in postings, with Illinois, Minnesota, New Jersey, Vermont, and Massachusetts all taking effect during 2025 (SixFifty, 2026). In a location-tagged subsample of our data, 48.5% of postings in pay-transparency states showed a figure, and so did a solid 40.2% elsewhere. Europe’s directive is phasing in too: it requires employers to share pay before the first interview, though not necessarily in the posting itself (Indeed Hiring Lab, 2026). In March 2026, 56% of UK postings showed pay against 12% in Germany.

Here’s the result we didn’t expect: among sourced candidates, JDs with a pay figure drew 12.8% replies against 13.9% for those without. That gap wasn’t statistically meaningful, so pay disclosure neither lifted nor hurt outbound response. Indeed’s 3.8x comes from a different funnel: job-board applicants who filter by salary before they click, whereas a sourced candidate hears about pay in the first conversation either way.

Should you publish pay anyway?

Yes. It’s legally required in a growing share of markets, it lifts inbound applications, and it costs nothing on the outbound side. Just don’t expect it to rescue an outreach message that isn’t resonating.

This Wall Street Journal segment looks at whether New York’s pay transparency law changed behavior once postings had to show ranges.

Examining NYC's New Pay Transparency Law: Is It Effective?

Is AI Making Job Descriptions More Inclusive?

Not measurably, at least not yet. Among employers that use AI in recruiting, 73.0% use it to write job ads (iHire, 2025). Yet masculine-coded density in our data stayed flat at 1.11 coded words per 100 in 2024-25 and 1.15 in 2026.

Signs of AI drafting are visible, though: em-dashes, a common tell of AI-assisted writing, jumped from single digits in 2024 to 18.7% of JDs in Q2 2025 and stayed above 21% after. That’s a proxy rather than proof, and stock phrases like “fast-paced,” “dynamic,” or “passionate” actually fell, from 49.7% of JDs in Q3 2025 to 33.7% in Q3 2026.

AI drafting won’t clean up biased language in job descriptions on its own, because language models trained on old postings reproduce old postings, five-year experience floors and degree requirements included. If your team drafts with AI, run the output through the same checks you’d use on a human draft. For the screening side of the problem, see our look at how AI resume screeners handle bias.

Frequently Asked Questions

What are examples of biased language in job descriptions?

Common examples include masculine-coded words like “dominant” and “competitive,” bro jargon like “rockstar” and “ninja,” age-coded terms like “recent graduate” or “digital native,” and physical requirements unrelated to the job. In Pin’s 2026 audit of 39,000+ JDs, bro jargon appeared in just 1.3%, while age-coded terms appeared in 1.8%.

What are examples of gendered language in job descriptions?

Gendered language includes gender-coded adjectives from the Gaucher, Friesen, and Kay (2011) list, such as “assertive,” “competitive,” and “decisive” (masculine) or “supportive,” “collaborative,” and “understanding” (feminine). It also covers gendered job titles like “salesman” or “chairman,” which “salesperson” and “chair” replace cleanly. In Pin’s audit, 88.6% of JDs contained at least one masculine-coded word.

How long should a job description be?

Aim for 300 to 600 words. Pin’s 2026 data shows JDs under 300 words draw 28% fewer candidate replies than 300-599 word JDs, while gains above 600 words disappear when comparing postings from the same company. In this dataset, the median description ran 404 words.

Does including salary in a job posting increase applications?

On job boards, yes: Indeed found postings with employer-provided salaries received 3.8x more applications in 2025. For sourced candidates, Pin’s data showed no meaningful reply-rate difference (12.8% vs 13.9%). Publishing pay helps inbound, is legally required in more than a dozen states, and costs nothing in outbound replies.

How do you remove biased language from job descriptions?

Cut requirements to the true essentials, offer “or equivalent experience” for degrees, remove age-coded terms and non-essential physical demands, and balance gender-coded adjectives before widening the pool the JD reaches. Pin users report 6x more diverse pipelines in Pin’s 2026 user survey, because multi-source sourcing reaches candidates no wording fix can.

How We Ran the Audit

To measure biased language in job descriptions at scale, we analyzed every eligible job description recruiters created on Pin between January 1, 2024 and October 18, 2026, after removing internal, demo, and test accounts. Each JD needed at least 500 characters of text, which left 39,000+ postings. For reply-rate analysis, we kept only JDs with at least 50 candidate outreach messages, a cohort of 3,000+ JDs.

How each signal was measured:

• Gender-coded language. Whole-word stem matching against the Gaucher, Friesen, and Kay (2011) masculine and feminine lists, run twice: once with the full lists and once excluding four near-universal stems (lead, analy-, respon-, support).
• Age, physical, and bro terms. Phrase matching on curated term lists; “young” and “fresh” reported separately because they’re usually not about people.
• Readability. Flesch-Kincaid grade level via the textstat library, with bullets converted to sentences so lists didn’t inflate scores.
• Pay disclosure. Any pay mention, plus a stricter test for an explicit currency figure or range in the JD text.
• Reply rate. Replies divided by outreach messages per job. We controlled for role family and seniority (inferred from job titles) and re-ran key results within the same company.

These limits matter for interpretation. Reply rates reflect the whole outreach message and the sender, not the JD alone, and observational data can’t prove causation. Role family and seniority are classifications inferred from title text, which leaves roughly a quarter of postings in an unclassified “other” group. Pin doesn’t store candidate gender, so we couldn’t split any result by gender, and we deliberately didn’t try to infer it from names.

How to Fix Biased Language in Job Descriptions in Five Minutes

Biased language in job descriptions is worth fixing, but this study points to a different order of operations than most checklists. Start with structure, then words:

1. Land between 300 and 600 words. Describe the work, the team, and what success looks like in the first 90 days.
2. Cut requirements you’d waive. If the hiring manager would interview someone without it, it’s a “nice to have.”
3. Add “or equivalent experience” to any degree line, and delete experience caps.
4. Remove age-coded terms and non-essential physical demands. These carry the legal risk.
5. Publish the pay range. It’s required in more places every year and has no outbound downside.
6. Balance the adjectives last. Run the draft through a free checker such as Kat Matfield’s Gender Decoder, swap a “competitive” for a “collaborative,” and move on.

Then fix the candidate pool, because a JD influences who applies but can’t reach people who never encounter it.

For teams that want diverse pipelines without waiting on inbound, Pin is the best AI recruiting platform to pair with a cleaner JD. Pin’s AI sourcing searches more than 850 million profiles across professional networks, GitHub, and patent filings, and it feeds zero demographic data into matching. Rated 4.8/5 on G2, it’s one of several options in this roundup of tools built for diversity sourcing. Whichever you choose, the order holds: fix the structure, publish the pay, then go find the people your posting will never reach.