Back to guides
3
5 min

AI-Powered Assessment

Question Banks, Rubrics & Bloom's Taxonomy

Beyond "Write 10 Questions"

Assessment is where most teachers first turn to AI — and where they are most likely to be disappointed. "Write 10 questions on photosynthesis" produces generic, textbook-regurgitation questions that test memory and little else. This chapter teaches you to generate assessment items that actually measure understanding, using Bloom's taxonomy as your guide.

Bloom's Taxonomy: Your Assessment Framework

Benjamin Bloom's taxonomy classifies thinking into six levels, from simple recall to complex creation. Each level requires a different type of question. Here is how it maps to assessment:

LevelThinking SkillQuestion TypeExample (Grade 8, Photosynthesis)
RememberRecall factsMCQ, fill-in-the-blank"Name the pigment that absorbs sunlight in leaves."
UnderstandExplain conceptsShort answer, matching"Explain why leaves appear green."
ApplyUse knowledge in new situationsWord problems, scenarios"A farmer notices her plants are yellowing despite adequate water. What could cause this?"
AnalyzeBreak down and examineCompare/contrast, data interpretation"Compare photosynthesis and cellular respiration using a Venn diagram."
EvaluateJudge and justifyArgument-based, opinion with evidence"A classmate says plants do not need sunlight to survive. Do you agree? Justify with evidence from our lab investigation."
CreateProduce something newDesign, project, open-ended"Design a controlled experiment to test whether light color affects the rate of photosynthesis."

Most teacher-made tests cluster at Remember and Understand. Standardized tests — whether SAT, AP, NAPLAN (Australia), or GCSEs (UK) — increasingly include Apply and Analyze questions. The new SAT, for example, emphasizes evidence-based reasoning over rote recall. AI can help you balance across all six levels — but only if you ask for it.

Prompting for Bloom's-Aligned Questions

The key is to tell AI exactly which level you want. Compare these two prompts:

Weak prompt: "Generate 10 questions on the American Revolution for Grade 8."

Strong prompt: "Generate 10 questions on the causes of the American Revolution (Grade 8, aligned to CCSS.ELA-LITERACY.RH.6-8.2 and C3 Framework D2.His.1.6-8). Distribution: 2 Remember (MCQ), 2 Understand (short answer), 2 Apply (scenario-based), 2 Analyze (primary source analysis), 1 Evaluate (argumentative), 1 Create (design task). Include an answer key with point values."

The strong prompt produces a balanced, standards-aligned assessment in under a minute. The weak prompt produces ten Remember-level MCQs.

Look at data/question-bank.json for 200+ pre-generated questions across five subjects (Science, Math, Social Studies, ELA, World Languages), three grade bands (6-8, 9-10, 11-12), and all six Bloom's levels. Use these as reference examples when crafting your own prompts.

Building a Question Bank

Instead of generating questions one test at a time, build a reusable question bank for each unit you teach. Here is the process:

Step 1: Map the Unit

For each unit, list 4-6 key concepts. For Grade 7 Math, Geometry (CCSS 7.G):

  • Concept 1: Scale drawings and scale factors
  • Concept 2: Drawing geometric shapes with given conditions
  • Concept 3: Area and circumference of circles
  • Concept 4: Angle relationships (supplementary, complementary, vertical)
  • Concept 5: Area, volume, and surface area of 2D and 3D shapes
  • Step 2: Generate Questions per Concept

    For each concept, ask AI to generate questions at 3 difficulty levels:

    DifficultyBloom's LevelsNumber of QuestionsPurpose
    EasyRemember, Understand5-8 per conceptHomework, warm-ups, support group
    MediumApply, Analyze5-8 per conceptClass assessments, core group
    HardEvaluate, Create3-5 per conceptChallenge problems, extension group, AP/IB prep, gifted learners

    For 5 concepts with ~15 questions each, you get a bank of 75 questions. That is enough for an entire unit — weekly quizzes, the unit test, and remedial worksheets — without writing a single question from scratch.

    Step 3: Review and Curate

    AI-generated math questions sometimes have calculation errors. AI-generated science questions sometimes state facts incorrectly. AI-generated history questions sometimes get dates wrong. You must solve every question yourself before giving it to students. Budget 30 minutes to review a 20-question set.

    Formative vs Summative Assessment

    AspectFormativeSummative
    WhenDuring learning (daily, weekly)After learning (end of unit, semester)
    PurposeIdentify gaps, adjust teachingMeasure achievement, assign grades
    StakesLow — not graded or lightly gradedHigh — counts toward final grade, GPA
    AI Use CaseExit tickets, quick quizzes, think-pair-share promptsFull assessments, rubric-graded projects, practice tests
    FeedbackImmediate, specific, actionableDelayed, summative, comparative

    AI excels at generating formative assessment items because they are short, frequent, and low-stakes. A daily 3-question exit ticket takes AI seconds to generate and gives you immediate insight into who understood the lesson.

    Prompt for a formative exit ticket:

    "Generate a 3-question exit ticket for Grade 6 Science (NGSS MS-PS1-2), today's topic: Chemical Reactions. One factual recall, one application, one 'what would happen if' question. Keep language at a Lexile level of 800-900."

    For UK teachers: exit tickets work well at KS3 and KS4. Adapt the prompt to reference the National Curriculum programme of study (e.g., "KS3 Chemistry — Chemical Reactions"). For Australian teachers, reference ACARA content descriptors.

    Rubric Design with AI

    Rubrics turn subjective grading into consistent, transparent assessment. But writing a good rubric takes time. AI can generate a first draft that you refine.

    Prompt for a rubric:

    "Create a 4-level rubric (Exemplary / Proficient / Developing / Beginning) for a Grade 8 ELA argumentative essay (500-750 words). Criteria: claim and thesis, evidence and reasoning, organization and transitions, conventions and style. Include specific descriptors for each level that a 13-year-old can understand. Align to CCSS.ELA-LITERACY.W.8.1."

    Check data/rubric-templates.json for 12 pre-built rubrics covering:

  • Argumentative, narrative, and informational writing
  • Science lab reports (NGSS practices-aligned)
  • Math problem-solving (showing reasoning, not just answers)
  • Social studies research projects and presentations
  • Group collaboration and discussion participation
  • Art, design, and multimedia portfolios
  • Each rubric includes four performance levels with student-friendly language. Adapt them to your specific assignments.

    Tips for Better Rubrics

  • Use student-friendly language. "Demonstrates sophisticated synthesis of multiple perspectives" means nothing to a Grade 7 student. "Uses ideas from more than one source to build your argument" is better.
  • Include anchor examples. For each level, include a one-sentence example of what that level looks like in practice. Many US districts call these "exemplars."
  • Share the rubric before the assignment. Students who know the criteria produce better work. This is not "teaching to the test" — it is transparent expectation-setting. In the UK, this aligns with Assessment for Learning (AfL) principles.
  • Handling AI-Generated Homework

    Your students have access to the same AI tools you do. When Sophia submits a perfect essay that sounds nothing like her usual writing, you have a choice: play detective, or redesign the assignment.

    Redesigning is almost always the better option:

  • Process-based assignments: Require students to submit drafts, outlines, and reflections alongside the final product
  • In-class components: The final essay is written in class; homework is the research and planning
  • Oral defense: "Tell me about your essay. What was the hardest part? What would you change?" A student who used AI without understanding cannot answer these questions
  • AI-transparent assignments: "Use Claude to generate a first draft, then improve it. Submit both versions with annotations explaining what you changed and why." This teaches critical evaluation — a more valuable skill than writing from scratch
  • Assessment Assembly for High-Stakes Testing

    Once you have a question bank, assembling a balanced practice test is straightforward. This blueprint works for SAT prep, AP practice, state testing, or GCSE revision:

    SectionBloom's LevelPointsNumber of Questions
    Section A (Multiple Choice)Remember, Understand2020 x 1 point
    Section B (Short Response)Understand, Apply186 x 3 points
    Section C (Extended Response)Analyze, Evaluate204 x 5 points
    Section D (Performance Task)Evaluate, Create122 x 6 points
    Total7032 questions

    Ask AI to select questions from your bank that match this distribution. Review the paper for topic coverage — make sure you are not accidentally testing Unit 3 six times and Unit 7 zero times.

    Key Takeaways

  • Use Bloom's taxonomy in every assessment prompt. Specify the exact distribution of Remember/Understand/Apply/Analyze/Evaluate/Create questions — this is the single biggest improvement you can make to AI-generated assessments.
  • Build question banks per unit, not per test. A bank of 75 questions per chapter gives you material for quizzes, tests, remedial work, and exam prep without repeated effort.
  • Always solve AI-generated questions yourself. Math errors, factual mistakes, and ambiguous wording are common — budget review time for every question set.
  • Redesign assignments for the AI era. Process-based, oral-defense, and AI-transparent assignments teach deeper skills than trying to catch AI use after the fact.
  • This is chapter 3 of AI for Educators (Global).

    Get the full hands-on course — free during early access. Build the complete system. Your projects become your portfolio.

    View course details