Most hiring teams now use AI somewhere in the process, whether they chose to or not. It arrives inside job boards, ATS add-ons, scheduling assistants and sourcing products. The question is no longer whether to use it, but whether you can show it is being used well.
Regulators have converged on a similar set of expectations. The UK government's Responsible AI in Recruitment guide, the US NIST AI Risk Management Framework, the ICO's 2026 recruitment guidance and the EU AI Act all point the same way. This checklist turns them into actions, organised by when you need them.
Print it, adapt it, and give each item an owner.
Before you buy
The cheapest time to fix a problem is before you sign.
- [ ] Write down the job you want AI to do. Which stage, which decision, which roles. "Help with hiring" is not a scope.
- [ ] Classify the risk. Does the system rank, filter or reject people? In the EU, AI used for recruitment is high-risk, with full obligations applying from 2 December 2027. In New York City, California, Illinois and Colorado, local rules may apply now or soon.
- [ ] Ask for evidence of bias testing, including who did it, when, on what data, and which characteristics were covered.
- [ ] Rule out banned features. In the EU, systems that infer emotions in the workplace have been prohibited since February 2025. Avoid "personality from your face" or "honesty from your voice" anywhere.
- [ ] Check explainability. Can your team see why a candidate was scored or ranked the way they were?
- [ ] Check records. Can you export inputs, outputs and logs? California requires relevant automated-decision data to be kept for four years. The EU AI Act requires deployers to keep logs for at least six months.
- [ ] Agree the contract terms on audit access, data use, deletion and what happens if a problem is found.
Before you switch it on
- [ ] Name an owner. One person accountable for how AI is used in hiring, with the authority to pause it.
- [ ] Set the boundaries in writing. What the AI may do on its own, what needs approval, and what it must never do. For example: it may schedule interviews; it may not reject a candidate without review.
- [ ] Train the people involved. Reviewers need to understand what the system does and where it can be wrong. Under Article 26 of the EU AI Act, oversight must sit with people who have the necessary competence, training and authority.
- [ ] Tell your workforce. In the EU, employers must inform workers' representatives and affected workers before putting a high-risk system into use at work.
- [ ] Prepare candidate notices. Plain language: that AI is involved, what it does, and how to ask for an alternative or a review by a person. Since 2 August 2026, EU rules require that people interacting directly with an AI system are told it is an AI unless that is obvious.
- [ ] Run a baseline. Test the system on past or sample data for your roles before real candidates meet it.
While it runs
- [ ] Monitor outcomes by group, not just averages. Compare selection rates across stages. The ICO suggests considering monthly bias reviews.
- [ ] Make review real. Research from the University of Washington found that when AI recommendations were heavily biased, people followed them about 90% of the time. Give reviewers time, context and an easy way to disagree, and track how often they do.
- [ ] Watch for drift. Roles, markets and candidate pools change. A system that was fair in January can be unfair by June.
- [ ] Keep candidate experience in view. Response times, drop-off rates and complaints are early warning signs.
When something changes
This is where most problems start, and where most checklists stop.
- [ ] Log every change to criteria, weights or thresholds, with who made it and why.
- [ ] Require a reason before the change applies. "We need more senior people" is a reason. "The ranking felt wrong" is a prompt to look closer, not a change.
- [ ] Re-test after material changes, not just at the annual audit.
- [ ] Pause if something looks wrong. The EU AI Act expects deployers to suspend use and inform the provider if a system may present a risk.
When a candidate asks why
- [ ] Have a route to a person, and make sure that person can actually change the outcome.
- [ ] Explain in plain words what the AI did, what the team decided, and on what basis.
- [ ] Respond on time. Set an internal target and meet it.
- [ ] Learn from the question. Every challenge is free feedback on your process.
Every quarter
- [ ] Review outcome data by group and stage.
- [ ] Review the change log and the reasons given.
- [ ] Check notices and policies against new rules. In 2026 alone, Colorado rewrote its law, the EU moved its deadlines, and the UK changed its automated decision-making rules.
- [ ] Check that published audit summaries are current. In New York City, the State Comptroller's December 2025 review of enforcement found problems with published audits, which suggests closer scrutiny ahead.
How MetaDay handles this
At MetaDay, we designed our AI recruiter so that most of this checklist is how the system works by default, not extra work for the hiring team.
- Boundaries first. The AI runs sourcing, outreach, follow-ups, screening calls, interviews, scheduling and negotiation only inside limits the company approves.
- Transparency by default. Candidates are told they are speaking with an AI and can choose to speak with the team instead. Most do not mind: in our published interview, 82% chose to settle salary and terms with the AI first.
- A reason for every change. When a hiring manager changes scoring criteria mid-run, the system asks for the reason first, so the change log in this checklist fills itself.
- People decide. The run hands the hiring team a ranked top three. The final interview and the decision stay with people.
Responsible and fast are not opposites: running inside these controls, World Mobile hired a QA specialist in one working day.
FAQ
What is responsible AI in hiring?
Using AI in recruitment in a way that is transparent to candidates, tested for bias, overseen by trained people with authority to intervene, and recorded so decisions can be explained.
Do small companies need a responsible-AI process?
Yes, in proportion. Illinois's AI rules sit inside its Human Rights Act, which covers employers with one or more employees in the state, and candidates everywhere expect to know when AI is involved. A one-page version of this checklist is a good start.
How often should we audit AI hiring systems?
At least annually where the law requires it, and after any material change. Monitoring outcomes monthly or quarterly catches problems much earlier.
Responsible AI is not what you promise in a policy; it is what you can show on any given day.
