Writing a good assessment is real work. You need questions that map to the role, that are fair, and that a rubric can actually grade. AI is genuinely useful for the first draft — and genuinely risky if it’s the last word. So we put a person in the middle.
AI drafts, a human approves
The platform can draft questions for an assessment from the role and the criteria you care about. But drafting is where the AI’s authority ends. Every generated question goes through a human review step before it can be used — you read it, edit it, keep it, or throw it out.
- Generate candidate questions from the role context to get past the blank page.
- Review each one, with AI-generated content clearly labelled as such.
- Edit wording, adjust difficulty, or rewrite entirely — you have the pen.
- Approve only what you want. Nothing an author hasn’t signed off on reaches a candidate.
The approval gate is the point
The value isn’t “AI writes your test.” It’s a faster first draft with a human firmly in control of what’s fair to ask. The approval step isn’t a formality — it’s the design.
Why we won’t skip the human
The same principle runs through the whole product: AI accelerates the work, people own the decisions that matter. A question that’s biased, ambiguous, or off-target is a problem the model can’t be trusted to catch on its own — so it doesn’t have to. An author reviews everything before it counts, and that review is recorded.
It’s the grading philosophy — AI grades, you decide — applied one step earlier, to the questions themselves.