The candidate screening process, stage by stage

Candidate screening is the work of reducing everyone who applied to the few people worth a serious interview, using the same criteria for each of them. It starts when an application lands and ends when a shortlist exists. Nothing in it is a hiring decision.

• Applicant screening process means the same thing.
• Candidate vetting process is the checking half: establishing that claims are true, then references and background checks.
• Candidate selection process is wider. It contains screening, plus the full interviews, the work sample and the offer, where the decision is made.

One thing has to exist first: written agreement on what the role requires, which a recruiter intake meeting produces. Screening against requirements nobody wrote down is sorting by taste.

There are seven, and you can run all of them on Monday. The time column is arithmetic on one role that draws 100 applicants, on the assumptions in the cells. Change them and the total moves, but the shape does not.

Swipe the table sideways to see every column.

About 22 hours for one role, three working days, and 80 of the 100 never spoke to anybody. At 200 or 400 the reading and the calls are what break, and how to screen 200+ applicants without a recruiter is the playbook for that volume.

Ask only what you will use to make a decision, plus the facts a resume cannot be trusted to state. Every extra field costs you applicants and buys nothing.

• The eligibility facts, as yes or no: right to work, a required licence, willingness to work the pattern the job really has.
• One thing that is hard to fake, as a short answer. Not "why do you want this job", which everyone answers the same way now.

What this stage fails at: free text. Anything typed at leisure can be written by somebody else or by a model, so treat prose answers as claims to check.

Is yours working? Count how many people you disqualify on form answers alone. Near zero means the questions are decorative; most of the pile means your advert describes a different job from the one you are filling.

A resume reliably tells you three things: the facts somebody chose to state, the claims they want judged, and how well they write about their own work. That is enough to remove the plainly ineligible, which is all this stage is for.

It cannot tell you whether a claim is true: "led the migration" reads identically whether they led it or sat near it. It cannot tell you scope, because "built the platform" might be two years alone or one quarter as one of thirty.

What this stage fails at: ranking. A resume read sorts documents by how closely they mirror your wording, which rewards whoever rewrote theirs most recently.

Is yours working? After each round of first conversations, count how many contradicted what the resume made you believe. None means you only call people whose documents already agreed with you.

The phone screen exists to hold one short, identical conversation with each plausible applicant and find out whether the claims survive a follow-up question. It is the most valuable stage, and the first to collapse, because its cost rises with every applicant.

1. It does not scale. Twenty calls plus the arranging is most of a week for somebody whose real job is something else.
2. So the call list gets cut from documents, which hands the ranking back to the stage that cannot rank.
3. The conversations drift. Ten unstructured calls are ten different interviews, and cannot be compared.
4. Scheduling loses people, at every step between interest and conversation.

Is yours working? Measure the days from application to first conversation, and the share of plausible applicants who never get one at all.

Compare candidates on criteria written down before you met any of them, scored one criterion at a time, with the evidence visible beside each score. The point is to stop the process re-deciding what the job needs once it has seen who applied.

• Score the criterion, not the person. Fill one column for everybody before moving to the next.
• Attach the evidence. A score with no quote behind it is an opinion with a number on it, and cannot be defended when a candidate asks.
• Keep the disagreements, and never average away a spike.

The template is in the structured interview scorecard.

What the decision stage fails at: memory. Decide in one sitting, from the written record, and write the reason down. It is what you reread in six months, and the only thing that tells you whether the earlier stages worked.

Late, on very few people, and never as a filter for the pile. Both are slow, cost money per candidate, and answer one narrow question: is there anything disqualifying nobody would volunteer.

What this stage fails at: references almost never say anything negative, because the candidate chose them. The useful question is specific and factual ("what did they own, and what would you have them do differently"), not "would you recommend them".

If a third party runs the checks, the law is specific. In the United States the Fair Credit Reporting Act requires "a clear and conspicuous disclosure" in writing, "in a document that consists solely of the disclosure", plus written authorisation, before a consumer report is obtained for employment purposes, and, before any adverse action on it, that you give the person "a copy of the report" and a written description of their rights: 15 U.S.C. 1681b(b), read on 30 September 2026. This is not legal advice, and local rules differ.

Is yours working? If a check has never changed an outcome, you are running it too early and too widely.

An AI interview replaces exactly one stage: stage 4, the first conversation. That is the stage that is identical for every applicant, that nobody has time to hold with all of them, and that therefore gets skipped for most of the pile.

On Career1, applying starts it. The AI recruiter speaks with each applicant in the browser for about eight minutes, in English, with nothing to install and nothing to book.

Its questions come from your role and that applicant's own resume, and it follows up when an answer is thin. Everyone is invited automatically while your monthly allowance lasts.

Each applicant arrives as a match score against the role, an overall score, a recommendation from strong hire to no hire, scored sections, a ledger of which claimed skills the interview demonstrated and the quote that shows it, risks, and questions for your own round, with the transcript and recording beside it. The sample interview report is a complete one to read first.

What it does not replace:

• The decision. Career1 ranks and recommends. It has no reject action of any kind. A person on your team decides and writes the reason.
• The work sample. For anyone who makes things, a short paid piece of real work beats a conversation.
• References and background checks, which nobody can self-report.
• The offer, and the negotiation after it.
• Tracking your candidates. Career1 is not an applicant tracking system and does not sync with one: a careers page, application forms and simple stages, nothing more.

Stage 2 shrinks rather than vanishes: the resume is read inside the interview, as the source of the questions. Stage 3 is untouched, and Career1 filters nobody out on your form answers.

Four checks, none of which needs a benchmark or an industry average, because each compares your process with itself.

1. The conversation rate. What share of plausible applicants ever spoke to anybody, human or machine. If it is small, the order applications arrived in is doing your screening.
2. The contradiction rate. How often a first conversation changes what a resume made you believe. Zero means it is decorative.
3. Your own selection rates, group by group. In the United States, under the Uniform Guidelines, a selection rate below four-fifths (80%) of the highest group's rate is generally regarded by the federal enforcement agencies as evidence of adverse impact (29 CFR 1607.4(D), read on 30 September 2026).
4. The written reasons, read backwards. Six months on, reread what you wrote at the decision. If what mattered on the job was never mentioned, fix the criteria, not the tooling.