The recruiter's guide to assessing candidate responses.

A recruiter screen usually ends with one judgment of the whole candidate, written as something like “strong, good communicator, comp is fine.” The hiring manager reading that note can’t tell which answer earned “strong,” or whether “fine” was a number or a shrug. Even with a good interview rubric open beside the call, plenty of a screen’s answers have no row on it.

Some answers in a screen are facts, like a start date. Others are reasons, like why the candidate wants to move. One or two are evidence, a story about past work that the rubric was written to rate. Rated on one scale, they blur into a single impression, and the hiring manager inherits it with no way to check it.

I’d sort the answers before judging any of them, and ask another question before a thin answer counts against a candidate. Each judgment then points back to words the candidate said, so the hiring manager gets a note they can argue with line by line, and the next interview starts from what the screen found out.

Sort each answer by what it was meant to settle.

Before you judge an answer, work out what kind it is, because the kind decides what you hold it against. A screen mixes the kinds within one short call, and an answer measured against the wrong thing can make a candidate sound better or worse than they were.

A candidate who is articulate about why they want the job and vague about what they did in the last one can leave the call rated “strong” because the reasons carried the score. Nothing in that note shows the hiring manager that the evidence never arrived.

Check each fact against the requirement, word for word.

Each fact meets the requirement or misses it, so the judgment holds up only as well as the wording you keep.

Write each fact in the candidate’s words. “Four weeks, maybe less if they let me go early” and “four weeks” are different answers, and a hiring manager planning a start date needs the first. “I can travel when the work needs it” leaves open whether the candidate can manage the two client trips a month the role carries, so ask about those trips before you hang up.

On pay, share the role’s range first and ask whether it works. A candidate who answers that they’re “flexible” has told you how they plan to negotiate, and nothing yet about the range. Ask once more, plainly, whether an offer inside it would work for them. If they still won’t say, that is the answer, and it goes in the note as said.

A fact that misses is a mismatch for the hiring manager to weigh. It carries no weight on any other answer in the screen, so an answer outside the range shouldn’t pull down how you rate the story that came after it.

Hold every reason up against the job as it is.

Reasons are easy to grade on how convincing they sound. I’d grade them on fit instead, by asking one thing: does this role give the candidate what they say they’re moving for?

Wanting to manage people is a good reason to leave a job, and a poor fit for an individual contributor role with no team on the horizon. A wish to build something new fits badly with a job that is mostly keeping a mature system running. Each of those reasons is sound, and the role still can’t give the candidate what they asked for.

When the reason and the role pull apart, say so in the call: describe plainly the part of the job that conflicts, and ask how it sits with them. What they say once they know the job is what you judge. Some candidates had the wrong picture of the role and are still keen once it’s corrected, and some realize it isn’t for them, which saves both sides the rest of the loop.

Write down the reason as they first gave it, and then what they said once they knew the job.

Judge the evidence against the interview rubric.

The evidence answers are the ones the rubric was written for, so this is where its written levels do the work. Read the level, then ask whether the answer contains what the level describes.

Your rating covers only what you can check. A recruiter screening an engineer can hear whether the candidate names a decision as theirs and says what it changed, and those are the things a level usually names. Whether the caching approach they describe was a sound design is beyond what a screen can weigh, so it moves your rating in neither direction. Write it under the rating as the candidate said it, so the number’s blind spot is on the page for the hiring manager and the engineer on the panel can choose to pick it up.

Keep how the answer sounded apart from what it held. A fluent answer on a clear video call and a halting one on a line with a lag can carry the same evidence, and the level asks about the evidence. The exception is a role where how someone explains things is itself on the rubric, as it often is for customer-facing hires, and there it gets its own line.

Rating against a level assumes the level names something an answer can contain. If yours only say how much of a quality the candidate showed, rewrite them before the next screen. Building them from answers candidates have given is the subject of a guide to interview rubric levels.

Ask one more question before an answer counts against them.

Every kind of answer has a thin version, and a thin answer is often a candidate guessing how much detail you wanted. Before one lowers anything, ask a question aimed at what the original question was asked to find out, and judge what comes back.

Each kind has its own probe. A fact hedged with “probably” gets a question about what it depends on, such as whether the start date hangs on a bonus or on a notice period. A reason given only in terms of pay gets asked what they’d want from the role if the money were equal. A story that stops short of what the rubric level names gets asked about that part in the level’s own terms, so if the level names who signed off, you ask who signed off.

Keep going until the answer settles the point its question was asked for, or until it’s clear the candidate can’t settle it. That second outcome is evidence too, and you can rate it.

Which probe to pick depends on what the first answer left out, a subject with its own page that goes through the usual gaps one at a time.

Keep answers that fit no level off the scale.

Some answers won’t sit on any level, and the pull is to give them the nearest number anyway. I’d write down what happened instead, because a number tells the hiring manager how the candidate did, and these answers mostly tell you what the screen didn’t find out.

• The candidate answered a question you didn’t ask. Steer them back once. If the answer still goes around the question, record the competency as not assessed and note that it was asked twice, which the hiring manager will want to know.
• The candidate raised something no one planned for, like a counteroffer due this week or a start date tied to a bonus payout. Write it down word for word and get it to the hiring manager while the next interview is still unbooked, because it belongs to the plan for the hire and none of your ratings should absorb it.
• The candidate gave strong evidence on a competency this screen doesn’t own. Pass it, in their words, to the interviewer who does own it, and leave it off your own ratings.

One rule covers all three: no number for no evidence. A 2 for an answer that never reached the competency tells the hiring manager the candidate is weak, when the screen hasn’t found out yet.

Where Metaview fits in assessing candidate responses.

A thin answer is only fair to judge if you can see whether it got its follow-up, and a judgment the hiring manager can check needs the candidate’s words behind it.

The Notetaker joins an interview as a visible participant and records and transcribes the conversation, and structured notes are written from it. Whether a call is recorded rests on the consent both people give under your team’s own process.

Metaview captures every spoken word of an interview it joins, so when you sort the answers afterward, a thin answer and the question you followed it with are both there to judge.

Whether an answer gets its follow-up is decided live, on the call, by you, and no recording can go back and ask the question you skipped.

Before you rate an answer as thin, find it in the transcript and look at what you said next. If you moved straight to the next question, the gap may be yours as much as the candidate’s, and “not assessed” is the fairer entry.

From the same recording, the Notetaker drafts a scorecard. Nothing on that draft counts as a rating until the person who held the screen picks one against the answer and writes the recommendation. They then submit it: directly into Ashby or Lever, and by pasting it into Greenhouse. In Ashby and Greenhouse, a browser extension can fill in the form’s objective sections, while the recommendation and every subjective field stay blank.

Later, when the hiring manager asks exactly what the candidate told you on start dates, the Metaview Assistant takes the question as the hiring manager would phrase it and answers from your team’s recorded interviews, start date included. A disagreement about one answer can then be settled by reading it.

Ask the hiring manager to find the words behind each judgment.

The method has worked when someone who wasn’t on the call can follow each judgment back to an answer, and there’s one test of that worth running on a real screen. Hand your note from one recent screen to the hiring manager for the role, and ask them to point, for each judgment on it, at the candidate’s words behind it.

Where they can point every time, the note is doing its job. Where they find a verdict like “comp fine” with no words under it, the call asked the question and the note dropped the answer, so the fix is in how you write. Where a rating points at nothing, it was an impression, and it should have gone back to the candidate as another question or into the note as not assessed. And where they find the words but read them differently from you, the test has turned up the argument this method is for: two people disagreeing about one answer, with the answer in front of them. The same test works on the interview feedback the panel writes after later rounds.