AI recruitment screening calls: who passes

Reading CVs is not selection

When a hundred CVs arrive for one vacancy, the problem is not selection but volume. Most of them get a few seconds of attention, and that glance verifies nothing: whether what is written actually fits only becomes clear in conversation.

That is also where most of the time goes. A recruiter calls dozens of candidates, asks the same five or six questions on every call, notes the answers, and only then does the real selection begin. This work is repeatable, its result is written down, and it requires no judgement — which makes it one of the best fits for automation there is.

This article is about screening calls: what screening establishes, the question set, the criteria, how a rejection is delivered, and the limits of this stage. Putting an interview in the calendar is a separate subject and is not what this article covers.

What screening establishes

The purpose of screening is not to find the best candidate — that is what an interview is for. Its purpose is to keep unsuitable applicants out early and respectfully, so that interview time is spent only on real candidates.

So screening criteria are knockout criteria: they do not score, they check a threshold. The typical ones are these.

• The core requirement of the role: a specific skill, a certificate, a driving licence
• Location and working pattern: office, hybrid, shifts, travel
• Whether salary expectations fall inside the range
• The date they could start, and their current situation
• Right to work or formal status, where the role requires it
• Language level — where the role requires it, checked in practice

Keeping this list short matters. Ten criteria at the screening stage both lengthen the call and artificially cut out suitable people — the real assessment happens at interview.

The question set and the assessment

A screening call is short and its structure has to be the same for every candidate. A stable sequence is what makes comparison possible and keeps the process fair.

1. 1The role is introduced in one sentenceTitle, what the company does, working pattern and location. Days have passed since the application, so the reminder is needed.
2. 2The thresholds are checkedThe criteria above are asked in order. Every answer is written as a field — not as free text, because free text cannot be compared.
3. 3The candidate's questions are answeredSalary range, working pattern, team size and similar are answered within what the company has approved.
4. 4The next step is statedWhen they will hear back and from whom. Without a specific time the candidate waits, and that is the single most complained-about part of hiring.

Which fields the answers are written into has to be decided in advance; the general principles are in the CRM field checklist for calls — that article is about sales calls, but the logic of field design is the same.

Timing: most candidates are working

The most overlooked detail in screening calls is timing. Most candidates are currently employed and cannot speak freely during working hours — particularly when the subject is a job search. The answer rate then looks low, while the problem is not interest at all.

• The first touch can be a short message: the reason for the call and a question about a convenient time
• Evenings and lunch breaks produce a noticeably higher answer rate
• The time the candidate names is recorded, and the next attempt is planned for it
• After two attempts the channel changes — a third call does not raise the rate
• How long the call will take is said in advance: a five-minute call should be five minutes

That last line matters more than people expect. A candidate who knows the call will be short engages with it; an open-ended conversation gets cut off — and a screening call that was cut off has to be made again.

Rejection: the visible part of the process

Most of what screening produces is rejections, and that is the part that forms an impression of the company. A candidate left without an answer leaves more than one vacancy behind — and tells other people about it.

• Every candidate receives an answer, including those who did not pass
• The answer comes within the time promised on the call
• The reason is given in general form: fit against the stated requirements of the role
• Consent is asked for keeping them in the database
• A segment is kept for future vacancies — the cheapest source of candidates there is

Consistency and equal treatment

One practical advantage of automating screening is consistency: every candidate gets the same questions in the same form, and the answers land in the same fields. In manual screening the questions drift with the hour of the day and the fatigue of whoever is calling.

Consistency does not by itself make a decision fair, but it makes it auditable: which criterion closed on which answer is visible in the record. The company's own hiring policy and its regulatory obligations sit above this process — the script does not replace them.

A separate note: the candidate has to be informed about call recording and data retention, and the company's policy applies. That is settled before the script is written.

Limits: who decides

The most important limit at this stage is simple: the final decision is not made automatically.

• Screening checks thresholds; it does not assess a candidate
• Borderline cases are not rejected automatically — they go to a recruiter
• If the candidate has an explanation or extra context, the call goes to a person
• Technical and professional assessment is the interview's work
• Special cases — disability, adjustments, non-standard experience — always go to a person

The second line is the one most often broken in practice: cutting borderline candidates automatically speeds the process up and loses some of the best people. Where there is doubt, the decision belongs to a person.

The last limit is a boundary between stages: screening checks thresholds, while putting the interview in a calendar is separate work, covered in AI interview scheduling calls. Merging the two into one call weakens both — one checks criteria, the other matches calendars.

How Vexvon supports this stage

Vexvon's AI voice agent calls from the company's own scenario, approved information, question flow and escalation rules. The parts screening uses:

• The candidate list becomes a campaign from a CRM filter; preview shows how many will be reached and which rows have no number
• The scenario defines the fields to extract: each threshold as its own field
• The knowledge base keeps the agent speaking only from approved information about the role
• Doubtful or non-standard cases hand the call to a recruiter
• Retry plans and reminders cover candidates who did not answer

How call quality is checked — systematic audit rather than listening to everything — is in AI call quality audit, and it applies to hiring calls as well.

Measurement

Four numbers follow the screening stage.

• Reach rate — how many candidates could be spoken to at all
• Pass rate — the share moving to the next stage after screening
• Share rejected after screening, and the distribution of reasons
• Effect on time to fill — the average length of the screening stage

The second needs reading carefully: a very high pass rate means the criteria are weak, and a very low one means they are too strict. To build this stage, get in touch or see the call center page.