Configure AI grading beyond provider setup
Connect question-level model answers and rubrics to ordered providers, personal credentials, pending status, correctness review, and human fallback.
Before you start
Own the quiz, configure at least one enabled AI provider, and keep its credential private. AI authoring and correctness-review controls require Premium or above; answer grading also needs a model answer on each enabled question.
Work through the product
Configure and test enabled provider profiles in quiz AI settings, ordered from first choice to last fallback. On each supported question, write the model answer, add an optional rubric, and enable AI grading. A respondent's compatible personal provider can grade future attempts on that person's quota; otherwise the quiz profiles are tried in order. Monitor pending status, use Analyze and Grade for unresolved answers, and treat Evaluate by AI or Evaluate correctness as author review tools, not automatic persistence of a final human decision.
Expected result
Automatically graded answers move through pending to a suggested score; failures fall through the configured provider order and then remain available for human grading.
Limits, privacy, and control
Never place API keys in question text, screenshots, or shared documentation. Review AI suggestions before relying on them; a human grade takes precedence and unsupported or failed requests fall back safely.
Troubleshooting
If a control is unavailable, confirm that you own the quiz, have the required organization role, and are within the current plan limits. Save the current tab before moving on, then preview the quiz with a test response before inviting respondents.