Explainers
AI candidate screening explained: CV scoring, interviews and chatbots
Four different technologies are sold as AI candidate screening. They read different evidence, fail in different ways, and work best in a particular order.
Agentic BluePublished
Four things sold as AI candidate screening
AI candidate screening is any software that helps decide which applicants a recruiter should spend time on. Four quite different technologies are sold under that name, and most of the confusion in buying them comes from comparing one kind with another.
| Kind | What it reads | What it can tell you | What it misses |
|---|---|---|---|
| CV scoring | The CV, against the job description | Whose stated experience matches the role | Pay expectations, notice period, availability, and strong people with weak CVs |
| AI pre-screening interview | A conversation with the candidate | Experience in their own words, plus the facts a first round needs | Whatever the format cannot capture, such as spoken skill in a text interview |
| Recruitment chatbot | Short answers to set questions | Basic eligibility, such as the right to work or weekend availability | Depth, because most cannot follow up |
| Assessments | Performance on a task or test | Whether someone can do a specific piece of work | Logistics and motivation, and many applicants will not finish one |
How CV scoring works, and where it goes wrong
CV scoring software, also sold as AI resume screening or AI shortlisting, extracts the text of each CV and compares it with the job description. Older systems matched keywords. Newer ones use language models to judge how closely the experience described fits the role, and can explain a score in a sentence or two.
It is fast and cheap per applicant, which makes it useful for putting a very large pile in a sensible order. It fails in predictable ways:
- It scores the writing, not the person. A strong candidate with a thin or badly formatted CV ranks below a weaker one who writes well.
- Scanned and image-only CVs. If the text cannot be extracted, there is nothing to score.
- It cannot see what a CV never says. Expected salary, notice period and whether the candidate is still looking decide more first rounds than experience does.
- Protected details. Malaysian CVs often list race, religion, age, marital status and IC numbers. Ask whether the system removes them before a model reads the CV, or scores with them in view.
How an AI pre-screening interview works
An AI pre-screening interview, run by an AI interviewer, holds the first-round conversation with each candidate. It asks about relevant experience, expected salary, notice period and availability, follows up when an answer is vague, and gives the recruiter a summary, a score against the criteria, and the transcript or recording.
Its strength is that it collects exactly what CV scoring cannot. Its weakness is that it only works on candidates who finish it, which makes the format, text, voice or video, the most important decision. The full explanation is in what is an AI interviewer, and what does it actually do?
Recruitment chatbots, including on WhatsApp
A recruitment chatbot talks to applicants on a careers site or in a messaging app. Most do three things: answer common questions, book interview slots, and ask a few knockout questions with fixed answers. Plenty of Malaysian hiring already happens over WhatsApp, and screening bots built to run there are available.
Knockout questions are useful where eligibility really is yes or no: the right to work in Malaysia, a driving licence, weekend shifts. They are poor at anything that needs judgement, because a bot that cannot follow up records “I have some experience” and “six years running the same system” as the same answer.
AI shortlisting: what a ranking should show you
Whatever does the screening, the output is usually a ranked shortlist, and a ranking is only as trustworthy as what sits under it. Before you rely on one, check that:
- you set the criteria it ranks against, and can change them;
- every score comes with its reasoning and its evidence, a quotation from the CV or the transcript;
- nobody is rejected automatically, and a person decides who moves forward;
- you can see the candidates it ranked low, not only the top ten.
The order that works at volume
- Knockout questions for anything that is truly yes or no, at the point of application.
- An AI pre-screening interview for everyone who passes, so each applicant gets a first round and you get the facts a CV leaves out.
- A ranked shortlist with evidence, which a recruiter reads rather than trusts.
- Assessments or a human interview for the shortlist only, where the extra effort is worth asking of the candidate.
CV scoring fits alongside the first two when the pile is so large that even ordering it helps. On its own, it moves the problem rather than solving it. The arithmetic behind that is in how to screen two hundred applicants for one role.
Questions to ask any AI screening vendor
- What exactly does the system read: the CV, a conversation, a test, or all three?
- Does it reject anyone on its own?
- Can I see the evidence behind every score?
- Are protected details removed before a model sees them?
- Are candidates told they are dealing with an AI?
- What does it cost per screened candidate, rather than per seat?
For the products available in Malaysia, see the 8 best AI interviewer tools in Malaysia and our buying guide to AI recruitment software in Malaysia.