How to screen 200+ applicants without a recruiter

Two hundred applications is not forty applications four times over. Reading each one for two minutes is most of a working day before anybody has been contacted, and a quarter-hour phone screen each would be five working days on the phone. So the pile gets cut on the strength of documents, which is the part of an application least connected to the work.

What happens at that volume is triage by fatigue. The first thirty are read properly, the middle is skimmed, the last stretch gets a glance, and the order applications arrived in becomes a ranking factor nobody chose. The arithmetic behind that is in what screening really costs.

The way out is not reading faster. It is deciding as little as possible from documents and as much as possible from answers, which means the first conversation has to be something that scales.

Before automating anything, separate the requirements that genuinely exclude somebody from the ones you would merely like. The test: if a candidate without it could still do the job well, it is a preference.

Hard requirements are few, and they are facts rather than judgements. The right to work where the role is based. A licence or certification the law requires. Availability for a shift pattern. The language the work is done in. Those belong on the application form as questions answered yes or no, where they cost nobody any time.

Everything else is a preference, and preferences should shape ranking rather than exclude. "Five years of experience" is the usual mistake: a proxy for judgement that one good answer about a hard problem measures directly. Put preferences in the job description so applicants can self-select, and keep them out of the gate.

Write the list down before you open a single application. Deciding what counts after you have met somebody is how a requirement quietly becomes a justification.

Swipe the table sideways to see every column.

The line runs between gathering evidence and judging it. Gathering the same evidence from two hundred people is repetitive work that software does without getting tired at applicant 140. Judging it is what your team is for, and the part that should never be handed to a score.

Keyword filters standing in for judgement

A filter ranks applicants by how closely they mirrored your wording, which rewards whoever guessed your phrasing over whoever did the work. It also inherits whatever else the document carries: in Bertrand and Mullainathan's field experiment, otherwise identical resumes drew 50 percent more callbacks when the name at the top sounded white. Filter on facts, never on quality.

Gut calls, applied two hundred times

Instinct feels like the fast option and is the least reproducible thing in the process. In Hoffman, Kahn and Li's study of job testing across fifteen firms, managers who hired against the test's recommendation ended up with worse hires on average, pointing at error rather than insight.

Unstructured screens

Twenty unstructured conversations cannot be compared with each other. The re-analysis of personnel selection research by Sackett, Zhang, Berry and Lievens revised every validity estimate downwards and still found structured interviews top-ranked among the procedures examined. Structure means the same ground covered with everyone, with room to probe an answer.

Cutting the pile before anybody speaks

The applicant with the worst-written resume and the most relevant experience is precisely who a document-first process loses, and you never find out that it did.

1. Write the role and the must-have list. Requirements as form questions, preferences as prose.
2. Open the interview to everyone who applies. On Career1 applying starts it, so there is nothing to send and nothing to book. The AI recruiter asks about the role and the applicant's own resume, follows up when an answer is thin, and takes about eight minutes in the browser on a laptop or a phone.
3. Let it run for two or three days. Applicants interview whenever suits them, including outside your working hours. You are not in this loop.
4. Read the top of the ranking, not the whole list. Fifteen or twenty reports properly beats two hundred resumes badly.
5. Pick eight to ten for a human round, using the questions each report suggests for that person.
6. Reject the rest promptly, by name. A fast no is the part of hiring candidates actually thank you for.

That shape holds at four hundred applicants as well as two hundred, because the only step that grows is reading further down a longer ranking.

A ranked list is a starting point, not a verdict. You are reading for evidence you could disagree with.

• The quote behind each judgement. If a report calls somebody strong on system design, the answer that showed it should be in front of you. A score with no quote is an opinion.
• The gap between claimed and demonstrated. Career1's report lists the skills from the resume and marks which ones the interview actually demonstrated, with the quote. At volume that gap is the most informative thing on the page.
• Risks and what was missing, not only strengths. A report that only flatters is not telling you where the second round should go.
• Follow-ups rather than first answers. The first answer to any question is the rehearsed one; read what happened after it.
• Candidates strong in one dimension and weak in another. An average hides the people worth arguing about, and those are the interesting ones.

There is a complete example report if you would rather see the shape before running a role through it.

Screening at volume means applying one procedure to a lot of people, which is exactly when consistency stops being a nicety.

• The same structure for everyone. Every applicant to a role gets the same interview shape and the same criteria, or the comparison is not a comparison.
• Keep the records. Transcripts and recordings let a decision be explained months later, to a candidate or to a lawyer.
• Watch your own selection rates. Under the Uniform Guidelines on Employee Selection Procedures, United States enforcement agencies generally treat a selection rate below four-fifths, or 80%, of the highest group's rate as evidence of adverse impact.
• Check the law where you hire. New York City's Local Law 144 requires a bias audit within a year of using an automated employment decision tool, published information about it, and notice to candidates. None of this is legal advice; ask counsel what applies to you.
• Let a person do the rejecting. Career1 ranks and recommends and rejects nobody, which keeps both the decision and the accountability with your team.

• Your first screen is a work sample. Ask for the code, the writing or the design review. A conversation does not substitute for it.
• You interview in another language. The interview is English only at the time of writing.
• You need an applicant tracking system. Two hundred applicants across eight roles with four interviewers is a coordination problem, not a screening one.
• Most of the pile is plainly ineligible. If eligibility alone removes most of it, the form has done the work and no interview is needed.
• You do not have applicants. Career1 does not source or find anybody; it interviews the people who apply to you.

1. Create a workspace: an email address and a password, no card.
2. Post the role, with your must-have questions on the application form.
3. Share the job link wherever those 200 applicants are going to come from.
4. Applicants interview themselves. Nothing gets scheduled by anyone.
5. Read the reports at the top of the ranking and choose a human round.

The free plan vets two candidates a month and needs no card, enough to judge the reports before committing. Paid plans are priced by monthly interview volume, from 25 to 150, on the pricing page. If you are still comparing tools, the criteria are in how to evaluate an AI interview platform.

Each source below was opened and checked on 26 September 2026.

• Sackett, Zhang, Berry and Lievens, "Revisiting meta-analytic estimates of validity in personnel selection", Journal of Applied Psychology: doi.org/10.1037/apl0000994.
• Bertrand and Mullainathan, "Are Emily and Greg More Employable than Lakisha and Jamal?", NBER working paper 9873: nber.org/papers/w9873.
• Hoffman, Kahn and Li, "Discretion in Hiring", NBER working paper 21709: nber.org/papers/w21709.
• Uniform Guidelines on Employee Selection Procedures, 29 CFR 1607.4: law.cornell.edu/cfr/text/29/1607.4.
• New York City guidance on automated employment decision tools: nyc.gov.