What is a structured interview?

Structure does not make a decision right, it makes decisions comparable. The distinction matters, because the method is usually sold as a certificate of impartiality. GoTeam makes no such claim.

What is written at the code level is narrow and concrete: both the question generation prompt and the answer grading prompt forbid drawing inferences from a candidate's age, gender, marital status, health, faith or political views. Even when that information appears inside an answer, it never enters the call. This is a written rule, not a guarantee about the outcome.

The grading prompt carries two more prohibitions: treating a short answer as automatically weak, and penalising the clipped sentences a recruiter writes while taking notes on their grammar. Both are common in interview notes, and both cost the candidate points that have nothing to do with what they said.

Some tools that get mentioned in the same breath do not exist in GoTeam: video interviews, skills tests, sourcing and social media screening. The module is built on interview scheduling, the question pool and answer grading. Writing that down upfront beats hunting for a tab that is not in the panel.

The items below are not configured in the panel, they come from the code. What makes structure structure is exactly this: they do not change from candidate to candidate.

• Three categories: Technical, Behavioural, Culture Fit.
• The round range is 1 to 5, default 1. One generation request returns at most 8 questions per category, default 5.
• The order of the pool is set by four keys: round, category, the manual order, then creation time. Two questions generated in the same batch keep a stable place.
• Writing a new order sends the full id list of that section; the same drag landing twice changes nothing.
• The decision scale and its weights. There is no mark outside the four steps.

The question text itself changes, and that is deliberate. The generation prompt requires every question to rest on a specific criterion of the position or on a specific claim in the CV. Tell me about yourself style questions, questions that close on a yes or a no, and questions that bolt two topics together with an and are forbidden. So is generating questions about the candidate's private life, age, marital status, health, faith or political views.

Next to every question sits a hint written for the recruiter: one sentence saying what a good answer has to contain. It is never read out to the candidate, it is the interviewer's own yardstick. This is usually the most useful part of the structure, because it does not leave the definition of a good answer until after the conversation.

The generation request arrives with the list of existing questions, and producing the same ones, or ones that mean the same thing, is forbidden. So a second generation does not inflate the pool, it fills the gaps that are still open.

The replace existing option only clears the rows the AI generated. Questions a recruiter wrote by hand are never touched and they still count in the duplicate check. A question you do not want to delete can be parked: a passive row stays out of the default list and can be called back when you need it.

The answer box under a question opens closed, and saving takes a deliberate click. There is no autosave, on purpose: a half sentence typed while the candidate is still talking would later be graded as the candidate's answer. Interviewer notes are a separate field, and autosave lives there.

The pool and the grading of a cancelled or completed interview are never deleted, they stay read only. For structure to be worth anything, the record has to outlive the conversation.

The answer score has bands of its own: 75 and above strong, 50 and above a second round. These are not the decision bands of the position, and the two should not be mixed up.

The answer score and the CV fit score sit side by side on the evaluation screen and are never added together. They measure different evidence: one is what the candidate told you, the other what they documented. The analysis result is written to the interview record, not on top of the fit record. The record also keeps the version of the analysis, so when a decision is read back later, the rubric it came from is known.

The evaluation section opens without an analysis being run: the CV evidence is already filled in, and the answer column sits empty. The reason is written down, an interviewer should see what the analysis will be placed next to before spending a credit.

Position criteria are weighted, and under full weighting the total has to be exactly 100. If a criterion is marked required and is not met, the system suggests neither an offer nor a rejection; the candidate goes to human review. A closed gate is not an elimination, it is the point where the automatic call stops.

When the answers are strong but the CV evidence is missing or weak, an offer is not suggested directly, a second round is. A suggestion never climbs until both kinds of evidence point the same way.

Every suggestion closes with the same sentence: this is a suggestion, and the application status changes only with your approval. The sentence is not decoration, it is the rule itself. The AI eliminates nobody at the end of an interview.

The button on the suggestion card does not change the status either, it takes you to the application centre. The status transition happens there, with your own approval and your own audit record. Who generated which questions and who analysed which round can be read back at any time.