HR can use AI in recruitment safely by limiting it to support tasks, such as drafting job ads, structuring interview questions and summarising information the recruiter has already collected, while keeping every hiring decision with a named person. The bias risk is highest when AI ranks, scores or rejects candidates. Use it to prepare and to document, not to decide, and review its output for unfair assumptions before anything reaches a candidate.
Where AI helps in the recruitment process
Recruitment involves a lot of writing and organising: job descriptions, advertisements, screening notes, interview plans and candidate updates. These are the places where a language model saves time without taking over a judgement call.
- Job ads. Turn a rough list of duties into a clear advertisement, and ask the AI to flag wording that may discourage some groups from applying.
- Interview question banks. Generate role-related questions tied to the competencies you define, so every candidate is asked a consistent set.
- Screening support. Help recruiters organise their own notes against a fixed checklist of requirements.
- Candidate communication. Draft polite, consistent updates and rejection messages for a human to review and send.
For the wider role of AI in people processes, see how HR can use AI for job descriptions and onboarding.
How bias enters AI-assisted hiring
AI models learn from large amounts of human-written text, and that text reflects human assumptions. The model can then repeat them without anyone noticing. This is explained in general terms in what AI bias is. In recruitment, bias can appear in several ways:
- Loaded language. A generated job ad may favour certain ages, genders or backgrounds through word choice.
- Proxy signals. If you ask AI to judge "culture fit" or "potential", it may lean on names, schools, locations or career gaps that say little about ability.
- Inconsistent treatment. The same CV summarised twice can produce different emphasis, which makes comparisons between candidates unreliable.
- Overconfidence. A fluent ranking looks authoritative, and recruiters may stop questioning it.
The safest design is to keep AI away from the final judgement of people. Where AI touches candidate information at all, define the criteria first, in writing, based on the job.
Keep a person accountable for every decision
The principle of keeping a human responsible is covered in what human-in-the-loop means. For hiring, put it into practice like this:
- Write the selection criteria before reviewing any candidates, tied to duties and required skills.
- Let AI assist with preparation, but have a recruiter or hiring manager read the actual applications.
- Never reject a candidate solely because an AI tool said so.
- Record who made each decision and on what basis, so the process can be explained and reviewed.
- Allow candidates to be considered through a normal human route if they raise a concern.
Take care with candidate personal data
CVs contain names, contact details, work history and sometimes sensitive information. Before pasting any of it into an AI tool, check what your company allows. The questions to ask are covered in AI and PDPA: 10 questions to ask and whether you can put customer data into Claude, which applies equally to candidate data.
- Use only tools your company has approved for personal data.
- Share the minimum needed, and remove identifiers where the task does not require them.
- Tell candidates how their information is used, in line with your privacy notice.
This is general information and not legal advice. Confirm your obligations with your legal or compliance team before using AI on candidate data.
A practical review checklist for AI-drafted hiring material
Before any AI-assisted job ad, question set or message goes out, ask:
- Does it describe the work and requirements, not a preferred type of person?
- Would every requirement still make sense if stated to every applicant in the same way?
- Has a human checked facts such as salary range, location and reporting line against the approved role?
- Are the interview questions the same for every candidate in the same role?
The habits in how to check if AI output is accurate apply directly here.
Building HR team skills
HR teams do better when they share approved prompts, a bias checklist and clear rules on candidate data. Agmo Academy offers hands-on sessions for this; see AI training for HR teams in Malaysia, the AI training overview and the Claude Certified Associate workshop, delivered by Agmo Studio, a Select Partner of the Anthropic Claude Partner Network. HRD Corp claimability is subject to eligibility and HRD Corp approval.