AI and Inclusion: Crafting Job Posts for a Diverse Workforce

Job postings carry more history and hidden bias than most recruiting teams realize, and handing them entirely to AI does not fix that. Pulling samples across decades of hiring language shows that postings have always mirrored the biases of their time, and that legislation aimed at equity tends to shift the wording without changing the underlying structure. Feeding that history into a generative tool just launders old bias into new copy.

Attempts to build software that writes a finished job post from a single prompt run into two hard problems. First, most people working in HR cannot actually define what a good job post looks like, so there is no reliable standard for a model to learn from. Second, job titles and role expectations vary wildly between companies, so a generic prompt produces generic, often inaccurate results. Grading tools that simply swap "aggressive" for "driven" do not solve this either, since changing a word without changing the substance of the role does nothing for the candidate reading it.

The bigger argument is about what happens when a bad post goes live. Generic, exaggerated listings drive away qualified applicants, cost real money in wasted sourcing spend, and set up new hires to quit early because the role never matched the description. That damage compounds when candidates immediately notice a lack of authentic detail or care on day one of the process.

Practical points covered include:

• Why job postings resist full automation, including the inconsistency between hiring managers describing "the same" role
• How a hundred years of job listing language still shapes today's templates and default phrasing
• Why word-swap "bias checkers" fail to make postings genuinely more inclusive
• What job postings mean as infrastructure for every other AI-driven recruiting tool built on top of them
• How combining human judgment with AI tools can produce postings that actually widen the candidate pool instead of narrowing it

The overall point is that job postings still need a human writer who understands the role, the audience and the history behind biased language, using AI as a tool for speed rather than as the sole author of the message.