Assessing communication skills when you hire

Communication assessment in hiring is the structured measurement of whether a candidate can explain their work, adjust it for the person listening, and handle being questioned about it. It is a job-relevant competency, judged against the role.

It is not a personality test, a clinical or speech-language assessment, or a general "communication skills" quiz of the kind used in training and development. Those are all real instruments and none of them predicts whether an engineer can explain a trade-off to a product manager, which is the thing you are actually hiring for.

The distinction is worth holding onto because the wrong instrument is easy to reach for. A quiz produces a tidy score and measures self-report. What you want is evidence of the behaviour, which means watching someone do it.

A written exercise measures written communication, which is a genuine skill and a small part of what the job needs. Everything else about communication is interactive, and interaction is precisely what a written test removes.

The competency that predicts performance is not "writes clearly". It is the ability to notice that the person opposite has not understood, and to change approach — a different level of detail, a different analogy, a question back. None of that is observable in a document, because a document cannot be confused.

Take-home writing has a second problem now: it is the format least able to tell you who produced it. A candidate with an hour and a model can submit prose well above their own baseline, and you will have measured neither their writing nor their thinking.

Break the competency into parts you can observe. Vague scoring is where bias enters, and "good communicator" is as vague as it gets.

The third row is the highest-signal and the most overlooked. A candidate who marks the edge of their knowledge is telling you their self-assessment is calibrated, which is worth more over a year than any amount of polish.

The format has to allow a second question. That rules out written exercises and recorded one-way video, both of which capture a rehearsed answer to a fixed prompt and stop there — a candidate can re-record until the answer is smooth, which measures editing rather than communication.

• Anchor on something they built. Ask them to explain a piece of their own work. It removes the guessing-game quality of an abstract prompt and gives you ground truth to probe against.
• Change the audience mid-answer. "Now explain that to the person who has to sell it." The switch is the test, and it takes ten seconds.
• Push once on something they are right about. You are watching whether they hold a correct position under mild pressure, which is a different signal from whether they can be persuaded.
• Ask what they would want to know if they were on the other side. Strong candidates answer this immediately; it is the perspective-taking question with the least room to prepare.

Structure it the same way for everyone. The same opening, the same audience switch, the same push — otherwise you are comparing four different interviews and calling it a comparison. This is the argument for a live interview rather than a recording: the follow-up is the measurement, and a recording cannot produce one.

This is the competency where careless scoring does the most damage, because the thing being judged is so close to the things you must not judge. Accent, first language, speed of speech, pauses, hesitation markers and cultural directness norms all read as "communication" to an untrained ear and none of them is the competency.

The defence is a rubric that names behaviours rather than impressions. "Adjusted the explanation when the listener was confused" is observable and reviewable. "Communicated well" is a feeling, and feelings absorb accent bias without anyone noticing. Score against the four signals above, write one line of evidence per score, and review the evidence rather than the number.

NorthAssay scores interviews from the transcript as text, and the scorer never receives a name, photo or country. That removes some of this and not all of it — word choice and phrasing survive transcription, and they carry signal about background. We have not commissioned a bias audit; our full position is on the responsible AI page, including what we have not done.

An interview is a conversation with a stranger who knows they are being judged, which is close to the highest-pressure communication setting a job contains and unlike almost all of the real work. Someone who is quiet under that condition may be entirely fluent in a team they know.

It also tells you nothing about written communication over time, which for many remote roles is most of the job — how someone writes a status update nobody asked for, or whether their handover documents get read. For where this competency sits among the others, see what candidate assessment is and how to choose a method, and for a role where this half decides the hire, assessing a data engineer.