AI hiring tools can inherit bias in ways that are hard to see and easy to defend. Here's the checklist we run every draft through.
In job descriptions
- Gendered language ("rockstar", "aggressive")
- Unnecessary years of experience
- Requirements that are proxies for pedigree
In screening
- Inferring protected characteristics from names, schools, or gaps
- Rewarding a specific career shape over demonstrated skill
- Treating career breaks as red flags
In outreach
- Templates that only land with one demographic
- Assuming compensation motivation over mission
None of this is solved by "turning off" AI. It's solved by keeping a human in the loop and giving the AI a rubric that names the bias out loud.