Most assessment tools score a submission and stop. NorthAssay reads what the candidate wrote, then interviews them about it — live, with follow-ups. The questions come from their answers, which is why a generated submission stops helping.
See it in actionAn assessment measures an artefact and infers a person from it. That inference held while producing a good artefact required being good, and it does not hold now. A strong candidate and a weak candidate with a good prompt submit work you cannot tell apart.
The industry answer to this is not better detection — every deployed method has a documented bypass. It is a short live conversation in which the candidate explains their own solution. That is the argument in full, and it is the thing this product is shaped around.
Their written answers, verbatim. The interviewer asks what they weighed, what they rejected, and what they would change.
The multiple-choice options they picked. Every wrong option in a NorthAssay assessment is authored with the specific misunderstanding it represents, so a wrong answer names the gap worth probing rather than just being wrong.
Follow-ups written in advance for each area, used when a first answer is shallow or generic.
Signals of a real answer — the concrete things someone who did the work tends to say.
The follow-up is the whole mechanism. A recorded answer to a fixed question can be scripted or read; the second question cannot, because it depends on what the candidate just said.
The criteria the interview is scored against, and its length, are the ones set for the role. Two candidates for the same job are judged the same way — only the questions differ.
The interviewer is never told to suspect anyone and never implies it. There is no cheating verdict anywhere in the product, no risk percentage, and nothing that flags a candidate as dishonest. Those numbers cannot be produced responsibly, so we do not produce them.
What you get is an ordinary interview record: what was asked, what was said, and a score against the role’s criteria with the reasoning shown. A candidate who did the work has a good conversation. That is the point — this is designed to be pleasant for the people you actually want to hire.
“These numbers cannot be produced responsibly.
They create suspicion without proof.”
“We kept both writes running for two weeks because the cutover window was one night.”
Explained the constraint behind the written answer and what they would change. Did not say how the rollback was tested.
Answer defence needs answers. It runs when the candidate has already submitted written or multiple-choice work for the role, which is the normal order: a short written stage decides who is worth a conversation, and the conversation decides the hire.
It raises the cost of faking substantially. It does not make it impossible, and a tool claiming otherwise is selling you something. A well-prepared candidate with real assistance can hold a shallow conversation, and a nervous strong candidate can interview worse than they work.
If you run a live interview on its own, there is nothing to defend, and the interview covers the role instead. That is still a real interview with real follow-ups — it just is not this.
A score here is evidence, not proof, and the decision stays yours — you can override any score, and we ask you to write down why when you do.
Paste a real posting, take the written stage yourself, and let it interview you about what you wrote. It takes about ten minutes and it is the fastest way to judge whether the follow-ups are any good.
Try it on yourself