AI drafts items, suggests scores and flags proctoring events. A trained person decides before any of it affects a candidate, and every step is logged.
Who does what, use by use
Three uses of AI, and nothing beyond them without your written agreement.
| Use | What AI does | What a person does | What is logged |
|---|---|---|---|
| Item drafting | Drafts items from your blueprint | An expert edits, approves or discards each | Draft, model, edits, approver |
| Essay scoring | Suggests a score with matched rubric lines | A trained scorer approves or changes it | Both scores, scorer, reason |
| Proctoring flags | Marks moments in a recording | A reviewer decides within two business days | Flag, clip, decision, note |
Rules we follow
These apply on every plan and every exam.
A person decides
No AI output reaches a candidate, a score report or a live form until a named person approves it.
Everything is logged
Model, version, input, output and the reviewer’s decision stay with the record, and you can export them.
No third-party training
Candidate responses, recordings and personal data are never used to train third-party models.
On only when chosen
Each AI feature is switched on per exam by your program lead, and can be switched off.
Flags are not accusations
A proctoring flag is a moment to review. Decisions about a candidate stay with your program.
Appeals see everything
On appeal you can show the response, the suggested score, the final score and who set it.
Questions about AI and scoring
Does AI ever set a final score on its own?
No. AI suggests a score against your rubric and shows which rubric lines it matched. A trained scorer approves or changes the score before anything is released, and the log records both numbers and the scorer’s name. When the two differ, the scorer adds a short reason that stays with the record.
Is candidate data used to train AI models?
No. Candidate responses, recordings and personal data are not used to train third-party models. We use model providers whose terms bar training on the data we send, and you can ask for the current list of providers and the terms that apply at any time, in writing.
Can AI-drafted items go straight into a live form?
No. A drafted item enters the bank as a draft and cannot join a form until a subject matter expert from your field edits and approves it and a reviewer checks it for bias and accessibility. The item record shows that it began as an AI draft, so its origin is never lost.
Does AI decide whether a candidate cheated?
No. AI marks moments in a recording, such as a second voice or the candidate leaving the camera’s view. A trained reviewer watches each flagged clip within two business days and records a decision. Any action about a candidate stays with your program, under its own published policy.
Can we switch AI features off?
Yes. Each AI feature is set per exam, and your program lead can switch it on or off at any time. With AI scoring off, responses go straight to your scorers in the same queues. With AI drafting off, your authors write in the same editor, without suggestions.
See the review queue yourself
We will show a suggested score, the scorer’s decision and the log behind both.