Bloom's Taxonomy is a six-level model of cognitive demand that you can use to make sure an assessment tests the depth of thinking you actually care about. The levels run from lowest to highest: Remember, Understand, Apply, Analyse, Evaluate, Create. A question that asks a learner to recall a definition sits at the bottom; one that asks them to design a solution sits at the top. The practical value is simple - if you only ever write Remember-level questions, you only ever find out what people can memorise, never what they can do. The taxonomy gives you a deliberate way to aim higher.
The six levels, with what each asks of a learner
- Remember - recall facts and basic concepts. Verbs: define, list, name, recall. "List the four rules of evidence."
- Understand - explain ideas in your own words. Verbs: explain, summarise, classify, describe. "Explain why evidence must be current."
- Apply - use knowledge in a new situation. Verbs: apply, demonstrate, solve, use. "Apply the rules of evidence to this portfolio and decide which items qualify."
- Analyse - break something down and see how the parts relate. Verbs: compare, contrast, differentiate, examine. "Analyse why this learner's evidence is sufficient for one unit but not another."
- Evaluate - judge against criteria and justify the call. Verbs: justify, critique, assess, defend. "Evaluate whether this assessment decision was defensible and explain your reasoning."
- Create - produce something new from the parts. Verbs: design, construct, develop, propose. "Design an assessment task that gathers valid evidence for this competency."
The note that matters: the verb is not the whole story. "Analyse this" is only an analysis question if the learner could not answer it by recalling something they were told. Cognitive level depends on what is new to the learner, not just the word in the prompt.
Why it matters for assessment
Two reasons. First, alignment. If your learning outcomes promise that learners can evaluate options but your assessment only asks them to remember definitions, the assessment does not measure the outcome - a gap auditors, accreditors, and honest reflection all catch. The taxonomy is a checklist for whether your questions reach the level your outcomes claim.
Second, it stops accidental floor-hugging. Lower-level questions are far easier to write and to mark, so under time pressure assessments drift downward into recall. Naming the level you are targeting before you write the question is the simplest defence against an assessment that quietly tests nothing but memory.
Higher levels need open-ended responses
Here is the practical constraint. Remember and Understand can be tested with closed formats - multiple choice, matching, true or false. But Apply, Analyse, Evaluate, and Create mostly cannot. You cannot truly assess whether someone can design a solution or justify a judgement with a tick box. The higher you go up the taxonomy, the more you need an open-ended response - a written answer, a worked case, a project, a recorded explanation - because the thinking you are testing only shows up when the learner produces something.
This is the catch that pushes people toward recall-heavy tests: closed questions are cheap to mark and open-ended ones are expensive. So assessments get written to the marking budget rather than the learning outcome. That is the wrong trade, and it is the specific problem worth solving.
How AI marking changes the trade-off
If open-ended questions are what let you assess the higher levels, the blocker has always been the cost of marking them at scale. AI rubric marking changes that arithmetic. You can ask the Analyse and Evaluate questions you actually want - the ones that reveal real capability - and have the model read each response against your rubric, map it to the performance level it reaches, and cite the evidence, with a qualified person reviewing and signing off. The higher-order questions stop being a luxury you ration.
The rubric is what carries the cognitive level into the marking: a well-written criterion for an Evaluate question describes what a strong justification looks like, so the model marks for reasoning, not keywords. If you are writing questions, pair this with how to write better assessment questions, and for the marking side see how AI rubric marking works. Scorafy is built to mark exactly these open-ended, higher-order responses against your own rubric. See it on a real submission.