Evidence extraction and summaries
Long-form responses and voice interview transcripts take time to digest. AI assistance extracts the evidence in a response criterion by criterion — observations mapped to your rubric — and can draft a whole-candidate evidence summary, giving evaluators a fast orientation before they read the underlying material. These are starting points, not substitutes: the full response and the stored transcript remain the evidence, and evaluators score against the shared rubric as usual.
Draft score suggestions
AI can draft criterion-level score suggestions for a response, including suggestions derived from interview transcripts. These drafts map the evidence to your rubric's criteria so the evaluator starts from a structured proposal instead of a blank form. A draft suggestion is exactly that — a draft. It carries no weight in any rollup until a named person confirms it or replaces it with their own judgment.
Transcript-only evidence in interviews
In interview evaluation, only the candidate's own words count as evidence. Accent and fluency are excluded by design: the AI works from the stored transcript, not from how the candidate sounded. This keeps voice interview scoring focused on the substance of what was said and protects candidates from being penalized for how they speak rather than what they demonstrate.
Confirm or override — a human decides
Every AI output passes through a human step: a named person confirms the suggestion or overrides it. The confirmed or overridden score is what enters the record, attributed to that evaluator with a timestamp like any other score. Within a criterion, evaluator scores are averaged exactly as in fully manual scoring — AI assistance changes how fast you get to a judgment, not whose judgment it is.
Model and version provenance
Every AI output used in evaluation is logged with model and version provenance — evidence extraction, score suggestions, transcriptions, transcript evaluations, and candidate summaries. Transcript evaluations and candidate summaries show the model right in the interface; the rest is kept in the platform's AI generation log. If a score is ever questioned, you can trace the assisted step, see the model behind it, and see the named person who confirmed or overrode it.
At a glance
- AI extracts evidence from responses criterion by criterion, evaluates interview transcripts, and drafts a whole-candidate summary.
- AI drafts criterion-level score suggestions, including from interview transcripts.
- In interview evaluation, only the candidate's own words count as evidence — accent and fluency are excluded by design.
- A named person confirms or overrides every AI output; nothing enters the record on AI authority alone.
- Every AI output used in evaluation is logged with model and version provenance.