Practical guide
Bloom's taxonomy: implications for AI-enabled work
Bloom's taxonomy is a framework for classifying educational objectives, not a standard for AI capability and not a scoring scale for workplace performance. Its most useful role in an AI programme is to help teams distinguish recall from understanding, application, analysis, evaluation, and creation when they design learning activities and assessment opportunities. This guide uses ALIGN: aim,.
Bloom's taxonomy is a framework for classifying educational objectives, not a standard for AI capability and not a scoring scale for workplace performance. Its most useful role in an AI programme is to help teams distinguish recall from understanding, application, analysis, evaluation, and creation when they design learning activities and assessment opportunities. This guide uses ALIGN: aim, level and knowledge, indicators, grounded activity, and non-claims.
Source status and applicability
The original cognitive-domain taxonomy was published in 1956 under Benjamin Bloom's editorship. Anderson and Krathwohl's 2001 revision uses a two-dimensional taxonomy table: a knowledge dimension and the cognitive processes remember, understand, apply, analyze, evaluate, and create. Krathwohl's 2002 overview explains the revision and its use for classifying objectives. Current university teaching centers use the revised taxonomy to align objectives, activities, and assessment. Bloom's taxonomy has no regulatory jurisdiction, mandatory effective date, certification authority, or special legal status for AI-enabled work. Library of Congress record for the 2001 revision Krathwohl revision overview UC Davis teaching guide
Sources: [1] [2] [3]
What changed in the revision
The original taxonomy used knowledge, comprehension, application, analysis, synthesis, and evaluation. The revised version expresses the cognitive processes as actions and places create after evaluate. It also separates the process from the kind of knowledge involved: factual, conceptual, procedural, or metacognitive.
That distinction matters. Asking someone to “evaluate an AI answer” does not specify what knowledge they need, which evidence they should examine, or what observable work would demonstrate evaluation. The taxonomy helps describe cognitive demand; it does not establish that an activity measures the intended capability.
The ALIGN design method
### A - Aim
Write the work or learning outcome as an observable action with an object and conditions.
Given [sources, tools, and constraints], the participant will [observable action] [work object] to support [bounded purpose].
Avoid objectives such as “know AI,” “understand prompting,” or “be confident.” They do not identify observable performance or the conditions under which it should occur.
### L - Level and knowledge
Classify both the cognitive process and the knowledge required.
| Process | AI-enabled work example | Possible evidence | | --- | --- | --- | | Remember | Recall the approved escalation route | Correct route named from policy | | Understand | Explain why a generated claim is unsupported | Explanation linked to source gap | | Apply | Use a verification checklist on a draft | Completed checks and corrections | | Analyze | Separate evidence, assumption, and inference | Annotated claim map | | Evaluate | Judge alternatives against stated criteria | Tradeoff record with counter-evidence | | Create | Produce a defensible work artifact | Deliverable plus evidence trail |
The rows are not a universal ladder of value. A safety-critical recall step can be essential, and a creative output can still be poorly grounded.
### I - Indicators
Define what observable evidence would support the objective and what would not.
| Objective | Strong indicator | Weak proxy | Missingness rule | | --- | --- | --- | --- | | | | | |
For AI-enabled work, indicators may include opening original sources, changing a prompt after critique, reproducing a calculation, rejecting unsupported content, documenting uncertainty, or escalating an exception. Prompt count, tool usage, confidence, and final polish are weak proxies unless the construct specifically requires them.
### G - Grounded activity
Design an activity that makes the cognitive process necessary. If the objective is evaluate, give the participant competing outputs, criteria, source evidence, and a consequential choice. If the objective is create, require a usable artifact under realistic constraints and retain evidence of how it was built and checked.
Align difficulty to the objective. Unfamiliar jargon, missing source files, inaccessible formats, and arbitrary time pressure can introduce irrelevant difficulty. A higher taxonomy label does not excuse a poorly specified task.
### N - Non-claims and review
Record what the activity cannot establish. Bloom classification does not demonstrate validity, reliability, fairness, transfer to work, general AI capability, or future performance. Review the objective, task, evidence, rubric, accessibility, and interpretation together. When any of those change materially, recheck the alignment.
Worked example: revising an AI-literacy module
A programme's original objective says, “Participants will understand AI hallucinations.” The activity is a short multiple-choice quiz asking for a definition. The quiz may show recall or recognition, but it does not create an opportunity to verify a real output.
Using ALIGN, the team rewrites the aim: “Given an AI-drafted supplier recommendation, source documents, and a claims log, the participant will evaluate material claims, correct unsupported statements, and document unresolved uncertainty for an approver.”
The level is evaluate, with conceptual knowledge of evidence quality and procedural knowledge of source verification. Indicators include opening the cited document, matching the claim to the passage, reproducing a material calculation, distinguishing observation from inference, and escalating a missing source. Merely flagging that AI can hallucinate is a weak proxy.
The grounded activity includes one accurate citation, one fabricated citation, a correct calculation, and one unsupported causal statement. The rubric cites observable evidence for each check. A pilot finds that one document title is misleading, so the team repairs it before use rather than treating confusion as weak capability.
The report states that the participant demonstrated specified verification and evaluation behaviours in the simulation. It does not claim that a Bloom level is a score, that the module validates general AI fluency, or that success transfers automatically to every workplace context.
Programme review questions
Does each objective name an observable action, object, and condition?
- Are cognitive process and knowledge type both explicit?
- Does the activity require the intended process rather than merely mention it?
- Are indicators direct evidence rather than usage or confidence proxies?
- Are realistic constraints present without adding irrelevant difficulty?
- Does the rubric align with the objective and available evidence?
- Are accessibility and missingness addressed before interpretation?
- Are transfer, validity, reliability, and benchmark claims separately evidenced?
Two Prune boundary
Two Prune can use learning objectives to help design realistic assessment opportunities and describe observed task evidence. Bloom's taxonomy does not validate a Two Prune assessment, supply a universal score scale, certify a programme, predict workplace performance, or authorize employment decisions.
Sources
Library of Congress: Anderson and Krathwohl's 2001 revision
- ERIC: Krathwohl, A Revision of Bloom's Taxonomy
- UC Davis: Bloom's Taxonomy of Cognitive Processes
- University of Utah: Bloom's Taxonomy
Sources: [1] [2] [3] [4]
Sources
- 1.Library of Congress record for the 2001 revision · www.loc.gov
- 2.Krathwohl revision overview · eric.ed.gov
- 3.UC Davis teaching guide · cee.ucdavis.edu
- 4.University of Utah: Bloom's Taxonomy · cte.utah.edu