AI agents · Guide
How AI Can Help Faculty Build Assessments
On this page
What AI can draft#
Building a good assessment takes time because there are many parts to write and align. AI can produce a first draft of several of them quickly.
- Assignment prompts. A task description, the audience and the deliverable.
- Rubrics. Criteria and performance levels for a task.
- Question sets. Questions at different cognitive levels from a reading or a unit, including follow-up questions for oral assessment.
- Scenarios and personas. A situation, a character and hidden information for a role-play. See how to design an AI role-play scenario.
- Variants. Parallel versions of a task, for practice and for graded attempts.
- Alternative formats. A text version, a simplified-language version or an alternative task.
- Course outlines. Modules and assessments mapped to outcomes, for you to restructure.
A draft is a starting point. It saves the time of a blank page and moves your effort to judgment.
A workflow that keeps you in charge#
- Decide the outcome and the evidence. Write what a student should be able to do and what you would accept as evidence. Do this yourself. It is the part AI cannot decide for you.
- Write a brief. Include the course, the level, the outcome, the time students have, the format and any constraints. Attach the readings or notes the assessment should draw on.
- Ask for a draft. Request one specific thing at a time, such as a rubric with four criteria and three levels.
- Review against your outcome. Does the draft assess what you intended?
- Edit. Change the wording to match your field's language and your students.
- Test. Try the assessment yourself or with a colleague. For a rubric, score two or three sample responses.
- Approve and publish. Nothing should reach students without your explicit approval.
A brief that works#
A brief for a rubric might read:
Course: Introduction to Public Health, second-year undergraduates. Outcome: students can evaluate a community health intervention using evidence. Task: a 1,200-word policy memo with an oral follow-up. Criteria wanted: evidence quality, reasoning, recommendation, communication. Four performance levels. Use the attached grading notes. Do not include criteria about formatting.
The more specific the brief, the more usable the draft. Vague requests produce generic rubrics.
Checks before you use a draft#
| Check | Question |
|---|---|
| Alignment | Does each criterion or question map to a stated outcome? |
| Validity | Could a student earn a high score without the knowledge you care about? |
| Accuracy | Are the facts, terms and examples in the draft correct? |
| Level | Do questions ask for the cognitive level you intend, such as application or evaluation? |
| Authenticity | Does the task resemble real work in the field? Gulikers and colleagues' five dimensions are a useful test. |
| Fairness | Is the language clear, and does it avoid assuming background not in your course? |
| Accessibility | Is there a plan for students who need alternatives? |
| Leakage | Do questions or scenarios reveal their own answers? |
| AI-readiness | Does the assessment collect evidence of the student's own reasoning? See authentic assessment in the age of generative AI. |
Villarroel and colleagues describe authenticity in terms of realism, cognitive challenge and evaluative judgement. Those three terms make a quick review lens for any draft.
Common problems with AI-drafted assessments#
- Generic criteria. "Clarity" and "organization" appear in every rubric. Replace them with criteria that reflect your outcome.
- Rubrics that reward polish. Fluent prose is easy to produce. Weight reasoning and evidence.
- Questions at the wrong level. Drafts lean toward recall. Ask for application and evaluation, and check.
- Invented content. A model may cite sources or facts that do not exist. Verify anything it states as fact.
- Hidden bias. Examples may assume a context. Read them as a student from a different background would.
- Over-long output. Ask for fewer criteria. Four to six are easier to apply.
Using AI inside the assessment platform#
Using a general chatbot means copying drafts between tools. AI built into an assessment platform can write the draft directly into the assignment form, the rubric bank or the question bank, using the context already in the platform. The tradeoff is that the assistant's access should be limited to what you are allowed to see, and nothing should be saved or published without you. Our guide to AI agents in higher education lists the control questions to ask.
Privacy#
Do not paste identifiable student information into any AI tool unless your institution has approved it. Course documents such as readings and syllabi are usually safe to use. If in doubt, ask your privacy office.
Where Rocketproof.ai fits#
Rocketproof.ai includes assistants that help instructors build assessments. The Assignment Creator turns a conversation into an assignment or simulation draft, the Rubric Builder drafts criteria and performance levels, the Question Builder drafts question sets for an assignment or a question bank, and the Course Builder plans modules, assignments and rubrics from a brief and up to three supporting documents, created as unpublished drafts. Each draft is saved only when you save it, and nothing reaches students until you publish it. See the built-in AI agents page.
Frequently asked questions
- Can professors use AI to create rubrics?
- Yes. AI can draft criteria and performance levels from a description of the task. The professor should check alignment with the outcome, accuracy and fairness, and test the rubric on sample work before using it.
- Can AI write exam questions?
- It can draft questions from a reading or a unit. Review them for accuracy, cognitive level and leakage of answers, and do not rely on a draft without checking.
- Is it safe to put course materials into an AI tool?
- Course documents such as readings and syllabi are usually fine, subject to copyright and your institution's policy. Do not paste identifiable student information unless your institution has approved the tool.
- Does AI-drafted assessment need to be disclosed?
- That depends on institutional policy. Whatever the policy, the instructor is responsible for the assessment and should review it fully before use.
Sources
- Gulikers, J. T. M., Bastiaens, T. J., & Kirschner, P. A. (2004). A five-dimensional framework for authentic assessment. Educational Technology Research and Development, 52(3). doi.org/10.1007/BF02504676(opens in a new tab)
- Villarroel, V., et al. (2018). Authentic assessment: creating a blueprint for course design. Assessment & Evaluation in Higher Education, 43(5). doi.org/10.1080/02602938.2017.1412396(opens in a new tab)
Rocketproof.ai builds assessment and simulation software for higher education.
Found an error or a missing source? Tell us.
Keep reading
- Oral assessmentOral Assessment Rubrics: A Faculty Guide With a TemplateHow to write a rubric for oral exams and oral defense, with criteria, performance levels, a printable template and tips for scoring AI-assisted sessions.Read the guide
- Simulation and role-playHow to Design an AI Role-Play Scenario and PersonaWrite an AI role-play for students: outcome, situation, persona, hidden information, ending rules, a testing protocol and a persona template you can copy.Read the guide
- AI agentsWhat Are AI Agents in Higher Education?AI agents use tools to work toward a goal. What that means in a course, the jobs agents do for faculty and students, and the control questions to ask.Read the guide
- Authentic assessmentAuthentic Assessment in the Age of Generative AIHow to redesign assessment when AI can draft the artifact: structural changes, evidence of reasoning, oral defense and what the research suggests.Read the guide
Try it with your own course in mind
The interactive demo needs no account. It shows what a student does and what the instructor reviews afterward.