How AI Helps Screen Candidates, and What HR Must Watch Out For

How AI screens applications, the traps of bias, PDPA and over-trusting scores, plus a checklist and a step-by-step rollout plan for HR teams.

A Marketing Executive role has been open for three days and several hundred applications have come in. Two HR people need to read them all before first-round interviews next week. The company's ATS has a ready button for AI to rank candidates. So the question is not whether to use it, but how to use it to find the right people without creating fairness and legal problems later.

What AI candidate screening is

AI candidate screening uses a system that reads applications or resumes and compares them with the role's requirements to rank or score fit. Newer systems do not just match exact words; they use language processing that understands context and compares the meaning of experience with the work required. Put simply, they try to answer how close this person's past work is to this job, not just which words they typed.

The consequence is that the criteria HR feeds the system matter more than the technology. If the JD is vague, for example “highly responsible, good communicator” without saying what the work actually is, the AI's ranking will be inaccurate too.

What AI genuinely helps with

• First-round ranking: lets the team start with the closest matches.
• Summarising applications: condenses long experience into short points for faster reading.
• Checking minimum requirements: such as legally required professional licences or languages the job truly needs.
• Drafting interview questions: creates questions that probe each person's experience.

Three traps HR often falls into

1. Bias built into the data

If the system learns from past hiring that was biased, such as hiring only from a few universities, it repeats that bias at a larger scale. The fix is to set criteria based on measurable skills and results and to sample-check whether any group is being screened out unusually.

2. PDPA risk

Candidate data is personal data protected under Thailand's Personal Data Protection Act. Collection, use and retention need a legal basis and notice to the data subject. If you send candidate data into an external provider's AI system, check that the use matches what you disclosed and that your contract with the provider covers data protection. Read more in PDPA and candidate data. This article is general information; consult a legal adviser or your organisation's DPO before setting up a real system.

3. Over-trusting scores

AI scores help with ranking; they are not verdicts. Letting the system reject people automatically with no human review hands a life-affecting decision to a formula you may not fully understand. Career changers and people with non-standard paths are often the ones wrongly screened out.

AI should help HR decide better, not decide instead with nobody accountable.

Keep only the data you need

A core principle of data protection is to collect only what you use and keep it only as long as necessary. Many organisations keep unsuccessful candidates' records for a long time without reason, a risk that is easy to overlook.

• Tell candidates how their data will be used, including automated screening, and how long it will be kept.
• Limit access to people directly involved.
• Have a process to delete data when the retention period ends.

For data after hiring, read PDPA and employee data.

Checklist before switching on AI screening

• Write a clear JD, separating must-have skills from nice-to-haves.
• Set scoring criteria based on skills and results, not gender, age or institution.
• Keep a log of why the system ranked each person, for later review.
• Always have a person review borderline scores; do not auto-reject everyone.
• Review results every quarter for unfair screening patterns.

Where to start

1. Pick one pilot role with many applicants and clear criteria.
2. Write scoring criteria point by point with the hiring manager before turning the system on.
3. Let AI rank, but have HR read in parallel in the first round, then compare how closely they match.
4. Adjust criteria where they differ, then expand to other roles.
5. Write a policy on AI use in recruitment so the team shares the same understanding.

Candidates should also understand how systems read applications, which we explain in how ATS reads resumes in 2026. If your organisation is hiring, post a job on Plants.work, and if your HR team wants a principled grounding in AI, browse courses from our learning partners.