Note · Job evaluation · Updated on

AI job descriptions and equal pay

A drafted job description reads well and describes what similar jobs do elsewhere. That becomes a problem once it is your pay evidence.

Short answer: AI drafts a job description from what similar jobs look like in general, not from what your job demands. Drafts tend to add duties and make different jobs sound alike. Descriptions become evidence when two jobs are compared for equal pay. Use AI for structure and wording. Take the content from the job holder and the manager. Check it before it feeds a grade or a band.

What the data and the rules say

Drafting job descriptions was one of the first jobs HR handed to AI. In a January 2024 survey of 2,366 US HR professionals, nearly two in three organizations that used AI in recruiting used it to help generate job descriptions. One employer in that survey described using AI to add skills and responsibilities to more than 200 existing descriptions and counted it a success [1].

The descriptions were often off before any tool touched them. In a report from October 2025, 79% of employees said their role differs substantially from its job description [2].

The rules compare jobs on the work. The EU pay transparency directive requires pay structures that compare jobs on skills, effort, responsibility and working conditions. It also requires job vacancy notices and job titles to be gender-neutral [3]. Under the US Equal Pay Act regulations, whether two jobs are equal depends on the actual job requirements and performance, not on titles or classifications [4]. A description that overstates the work does not change the work. It changes what the company has put in writing.

What it means for a company of 50 to 500 people

Drafts inflate. Ask a tool for a senior analyst description and it writes what senior analysts do across many companies: Stakeholder management, strategy, leading projects. Your analyst may do none of it. Added duties score higher in a job evaluation. In a dispute, the inflated description is the one a colleague in a plainer job will point to.

Drafts blur. Run every job through the same prompt and the jobs come out sounding alike, with the same verbs and the same collaboration lines. The real differences in skill, effort and responsibility that justify different pay drop out of the record. Our guide on proving equal value covers why the record matters when two jobs are compared.

A hiring document becomes a pay document. A description written to attract applicants is saved as the job's record. It is then used to set the level, to match the job to a salary survey and to score it. Words chosen to sell the job end up setting its pay.

A bulk rewrite changes every comparison at once. Rewriting all descriptions in a week replaces the written record for every job. If nobody checked the new text against the work, the company now holds a full set of claims it may not be able to support.

Our view

We would use AI for structure and plain wording, not for content. What the job requires comes from the person in it and their manager. Someone checks it against the work before it feeds a grade, a band or a survey match. Keep the job ad and the job's record as separate documents. If your descriptions have already been rewritten in bulk, check first the ones that set pay: Jobs near a grade boundary and jobs that are often compared with each other.

Sources

  1. SHRM, 2024 Talent Trends: Artificial intelligence in HR (survey of 2,366 HR professionals, 10 to 19 January 2024), pages 3 and 20, checked 2026-11-03
  2. Korn Ferry, This Week in Leadership: Not what I signed up for (20 October 2025; survey sample not stated), checked 2026-10-27
  3. Directive (EU) 2023/970 on pay transparency, Articles 4(4) and 5(3) (EUR-Lex), checked 2026-11-03
  4. 29 CFR 1620.13(e), Equal Pay Act regulations: Equal work, what it means (via LII), checked 2026-11-03

Are two different jobs of equal value?

Titles and job families do not settle equal value. The demands of the work do. Five traps in proving the comparison and what sound evidence looks like.

Job matching: Where benchmarks go wrong

Most bad benchmarks are bad matches, not bad data. Five traps in matching jobs to a salary survey and what sound matching looks like.

AI in compensation: What to automate

Automate the data work and keep pay decisions with people who can explain them. Five traps in using AI for matching, pricing and raises.

See Resources for every guide and the self-check.