Where AI Helps in Recruitment and Where It Creates Risk

The safe, valuable half

• CV parsing into structured records, with the source attached
• Call and interview summaries into the ATS
• Job description drafting from a briefing call
• Candidate shortlist packs assembled from existing records
• Compliance document chasing and checking

None of those decides anything about a person. They remove typing, which is where recruiters lose most of their day.

The half that needs care

Any system that ranks or filters candidates can produce discriminatory outcomes even without protected characteristics in its input, because proxies exist. That is a legal exposure, not a theoretical one.

If you rank, do it against documented job-related criteria, keep the decision with a person, and monitor outcomes by group for disparate impact.

Practical safeguards

1. Remove name, age, address and education institution from screening inputs
2. Score against criteria drawn from the job specification, written down
3. Show the reason for every score, referencing the CV
4. Never auto-reject — surface, do not filter
5. Review outcome distributions quarterly and record that you did

Candidate experience is a differentiator

Faster acknowledgement, faster feedback and fewer candidates left in silence. Those are the things candidates actually complain about and they are all automatable without any ranking.

A rejection sent promptly with a real reason does more for your brand than any ranking model.

Regulation is tightening

Several jurisdictions now require disclosure, bias auditing or both for automated employment decision tools. The direction of travel is consistent even where the detail differs.

Building with documented criteria and outcome monitoring from the start means the compliance work is already done when the rule arrives.

Frequently asked questions