Rethinking Bloom’s Taxonomy for the AI Era

Jason Blomquist; Liza Long; Joel Gladd

10 Rethinking Bloom’s Taxonomy for the AI Era

How This Chapter Fits in Part 2

The Need for Rethinking Assessment identified the assessment problem: AI can produce fluent work that may hide whether students learned. This chapter moves from the problem to learning outcomes. Before faculty choose a two-lane map, an AIAS level, an authentic task, or a process requirement, they need to decide what kind of thinking the assessment should make visible.

The later chapters build from that decision. Inspire and Assure maps the course, AI Policy Frameworks and the AI Assessment Scale names assignment-level AI expectations, The Role of Authentic Assessments and UDL redesigns the task, Why Process-Oriented Assessment Works Well in the AI Era documents the process, and AI Detection and Accountability handles detection and accountability.

Why Bloom’s Feels Different Now

Many faculty know Bloom’s taxonomy by heart, even if they do not use the name every week. We ask students to remember, understand, apply, analyze, evaluate, and create. We use those verbs to write course outcomes, scaffold assignments, and explain why one task is more complex than another.

That vocabulary still helps. A nursing instructor, a historian, a welding instructor, and a biology instructor all need ways to describe cognitive complexity. AI can now produce work that looks like several Bloom levels at once. A student can ask AI to summarize a reading, apply a theory, analyze a case, evaluate alternatives, and create a polished response in the same conversation.

The finished product may look strong. It may also hide the student’s actual learning.

Tina Austin makes this point in her essay on the evolution of Bloom’s taxonomy. If AI can generate a product that appears to sit at the top of the familiar Bloom’s pyramid, faculty have to ask what the student actually did. Did the student understand the concept? Did they notice weak reasoning? Did they question the AI’s assumptions? Did they make a disciplinary judgment? Did they decide where AI should stay out of the process?

Those questions are assessment-design questions. Bloom’s revised cognitive-process categories can still guide us, but the verb alone does not tell us what evidence to collect. In a study of 940 biology assessment items, Larsen et al. found that action verbs did not reliably predict the cognitive process an item required.

Product Quality and Learning Evidence

Faculty often use final products as evidence of learning, even though a product does not fully reveal the process behind it. A polished lab report may suggest the student worked through the data. A coherent essay may suggest the student read, interpreted, and revised. A strong business recommendation may suggest the student understood the case.

Generative AI makes that limitation harder to ignore. A 2026 TEQSA-commissioned resource (PDF) recommends placing evidence of learning processes alongside assessment products. Students can now submit work that is coherent, well-organized, and superficially aligned with the prompt while skipping the thinking the assignment was designed to produce.

Consider a common outcome:

Students will analyze a business case and create a recommendation.

That outcome uses higher-order Bloom’s language. It also leaves the evidence problem unsolved. An AI tool can analyze the case and create a recommendation. The instructor still needs to know whether the student identified the central problem, weighed tradeoffs, understood the local constraints, and revised a judgment after encountering better evidence.

The outcome becomes stronger when it specifies the evidence of thinking:

Students will compare an AI-generated recommendation with course concepts and local constraints, identify what the AI missed, revise the recommendation, and explain how their judgment changed.

The second version still asks for analysis and creation. It also asks the student to provide more evidence of their reasoning.

The UnBlooms Framework

Austin’s UnBlooms Framework responds to this problem. In the Oxford Zenodo record, she describes UnBlooms as a problem-centered model for learning design in the AI era. It treats learning as recursive. In plain terms, students may enter and re-enter questioning, generating, critiquing, and refining wherever the problem requires, with reflection and evaluation guiding the work. Learning does not always move cleanly from remembering to creating.

Austin’s OpenAI Higher Education Summit record names three ideas that fit faculty assessment design especially well.

First, learning is recursive. Students may begin by generating an idea, then discover they need more background knowledge, then critique a source, then revise the idea, then return to the original problem.

Second, assignments should make student voice and lived experience structurally necessary. AI can generate plausible language about an experience, but it does not possess the student’s experience. In this training, local observation, personal stake, disciplinary judgment, field experience, and live explanation are possible ways to enact that principle.

Third, assessment should include the student’s ability to evaluate AI output. Students need to identify errors, bias, missing context, weak assumptions, and places where AI output sounds fluent while failing the task.

This training connects that principle to ABC-R from Part 1. Students can check accuracy, bias, context/relevance, and responsible use when they critique AI output. A revised outcome may include one or more of those moves, depending on what students need to practice.

For this training, the practical takeaway is a shift from the product to the student’s judgment. Faculty can ask directly: What would provide evidence that the student thought with care?

Bloom’s, Inverted Bloom’s, UnBlooms, and Bloom’s stAIrcase

Several Bloom’s-related models are circulating in AI education. They are easy to blur together. For this training, it helps to keep them distinct.

Inverted Bloom's names an emerging family rather than one standardized model. Pesovski, Vorkel, and Trajkovik describe a reversed sequence for AI-supported programming education, while Megan Workmon Larsen offers a flexible, faculty-facing version that begins with creation and works back toward understanding.

Bloom’s-Related Frameworks: Basic Ideas
Framework	Basic Idea
Bloom’s taxonomy	A familiar vocabulary for describing cognitive moves such as remembering, applying, analyzing, evaluating, and creating.
Inverted Bloom’s	Students may start with AI-generated text or ideas, then work backward into critique, revision, explanation, and understanding.
UnBlooms	Students enter and re-enter questioning, generating, critiquing, and refining as the problem requires, guided by reflection and evaluation.
Bloom’s stAIrcase	A practical AI-literacy activity-design tool that crosses Bloom’s cognitive processes with knowledge dimensions and AI-literacy competencies.

Bloom’s-Related Frameworks: Uses and Risks
Framework	Useful For	Risk If Used Poorly
Bloom’s taxonomy	Writing outcomes, sequencing tasks, and discussing cognitive complexity with colleagues.	Faculty may assume a higher-order verb proves higher-order student thinking.
Inverted Bloom’s	Writing-intensive courses where students can analyze and revise AI output as a learning activity.	The model can remain too linear if students only move from generated output to critique.
UnBlooms	Designing assignments where metacognition, critique, student voice, and disciplinary judgment need to be visible.	Faculty may treat it as a slogan unless the assignment prompt and rubric specify the evidence students must provide.
Bloom’s stAIrcase	Browsing examples and designing classroom activities and AI-literacy tasks across disciplines.	As with any design tool, faculty may add AI activities while leaving the assignment’s evidence problem unchanged.

These frameworks can coexist. A faculty member might use Bloom’s language for outcomes, UnBlooms to rethink the learning evidence, Bloom’s stAIrcase to design an activity, and AIAS in AI Policy Frameworks and the AI Assessment Scale to communicate what AI use is allowed.

From Outcome Verbs to Evidence of Thinking

When an outcome targets a cognitive process, a useful AI-era outcome names both the cognitive move and the evidence students will provide.

Revising Outcomes for Visible Thinking
Original Outcome	AI-Era Revision	Evidence Students Provide
Analyze a historical event.	Compare two AI-generated explanations of a historical event against primary sources, then explain which claims hold up and which need revision.	Source annotations, claim checks, and a short explanation of revised historical judgment.
Create a lab report.	Interpret lab data, identify where AI-generated explanations fit or misread the evidence, and defend the final interpretation using course concepts.	Data notes, AI critique, final report, and a brief defense of the interpretation.
Evaluate a public policy proposal.	Test an AI-generated policy analysis against stakeholder needs, local constraints, and evidence from the course, then revise the recommendation.	Stakeholder map, critique of AI assumptions, revised recommendation, and rationale.
Develop a patient education resource.	Create a patient education resource, check AI-assisted language for accuracy and audience fit, and explain how health literacy shaped the final version.	Draft, accuracy check, audience-fit note, and final resource.

Notice the pattern. The revised outcomes do not prohibit AI across the board. They specify what students must do with content, tools, sources, constraints, and feedback. The assessment can then ask for evidence that matches the outcome.

This is an outcomes audit. It checks whether the stated learning outcome still names the thinking students need to practice. A preprint by Zaphir, Lodge, Lisec, McGrath, and Khosravi offers a related but narrower four-step check. It maps one assessment question to a cognitive skill and disciplinary criteria, tests several versions with an AI system, grades the outputs, and evaluates the question’s vulnerability to AI completion. That process can inform task redesign, but it does not show what a student thought. For faculty, the practical version is direct: decide which part of the thinking belongs to the student, then ask what evidence would show that thinking happened.

The word disciplinary needs to become concrete during that audit. Stolpe, Larsson, and Johansson Falck’s (2026) DiSAIL framework distinguishes a field’s language and genres, cognitive strategies, and standards for producing knowledge. Faculty can therefore ask what students must communicate, how they must reason, and what the field accepts as support. A general verb such as analyze becomes assessable when the outcome identifies those disciplinary practices and the evidence students will provide.

Faculty Spotlight: Marcus Hickens, Business

Marcus Hickens teaches Principles of Management. In The Need for Rethinking Assessment, he identified a case-analysis assignment that had become vulnerable to AI. The original prompt asked students to analyze a business case and create a recommendation. Students submitted polished memos. Some were strong, but Marcus could not tell how much thinking sat behind them.

He begins with the outcome:

Students will analyze a management case and create an evidence-based recommendation.

The verbs sound rigorous. Analyze. Create. Evidence-based. But Marcus realizes the outcome still points mainly toward the final memo. AI can produce that memo quickly.

He revises the outcome:

Students will identify the central management problem in a case, compare an AI-generated recommendation with course concepts and local constraints, revise the recommendation, and explain which tradeoffs shaped their final judgment.

That revision changes the assignment. Marcus can now ask students to submit four pieces of evidence: their own problem statement, the AI-generated recommendation they critiqued, an annotated revision, and a short explanation of the tradeoffs they considered.

The final memo still counts. But Marcus is no longer relying on the memo alone. He can see whether students noticed weak assumptions, used course concepts, and made a judgment the AI did not make for them.

Practical Takeaway: Complete an Outcomes Audit

Choose an assignment from one of your courses. If the assignment you audited in The Need for Rethinking Assessment is still useful, you can reuse it. Start with the learning outcome attached to the assignment, or write the outcome you have been assuming.

Then answer these questions.

What does the current outcome ask students to know, do, or decide?
What can AI make easy in this assignment?
What thinking could a student skip if they used AI carelessly?
Which disciplinary practice must remain visible, such as framing a problem, interpreting evidence, applying a method, explaining uncertainty, or communicating in a field-specific genre?
What evidence would show that the student actually did the thinking?
Where should students question, generate, critique, revise, or resist AI?
What would need to appear in the prompt, rubric, or class activity for that thinking to become visible?

Now revise the outcome in one sentence. Keep the original cognitive goal if it still fits. Add the evidence of thinking that would let you assess it.

If you ask AI for help revising the outcome, use the R-T-C-I pattern from Part 1. Give the AI a role, define the task, provide course context, and specify the kind of output you want. Then use ABC-R to evaluate the suggestion before adopting any language.

You might use this structure:

Students will [cognitive move] by [specific action], using [course concepts, sources, data, tools, constraints], and will show their thinking through [evidence].

For example:

Students will evaluate a workplace communication problem by comparing an AI-generated response with course principles, revising the response for audience and context, and explaining the choices that shaped the final version.

Your audit should leave you with two versions: the original outcome and the revised outcome. Compare them and notice whether the revision makes the intended thinking easier to assess.

Looking Ahead

The next chapter, Inspire and Assure, turns from outcomes to the course map. The Inspire and Assure approach helps you examine how strongly an assessment motivates learning and how readily it can assure learning. The two-lane approach then gives students a simple way to see where work is secured and where AI-supported learning can remain open.

References

Austin, T. R. (2025, September 16). The UnBlooms™ model: A Problem-Centered Framework for Learning Design in the AI Era [Presentation]. Zenodo. DOI record

Austin, T. R. (2025, October 20). Teaching with AI across Disciplines: Reframing How We Learn: The UnBlooms™ Framework Presented by Tina Austin at OpenAI Inaugural Higher Education Summit, October 20th, 2025 [Presentation]. Zenodo. DOI record

Austin, T. (2026, May 11). The evolution of Bloom’s taxonomy, and where it was always heading in the age of AI. Tina’s Substack. Source link

Krathwohl, D. R. (2002). A revision of Bloom’s taxonomy: An overview. Theory Into Practice, 41(4), 212-218. DOI record

Larsen, T. M., Endo, B. H., Yee, A. T., Do, T., & Lo, S. M. (2022). Probing internal assumptions of the revised Bloom’s taxonomy. CBE—Life Sciences Education, 21(4), ar66. DOI record

Lodge, J. M., de Barba, P., Ainscough, L., Brazil, J. R., Broadbent, J., Ebbert, D., Frankland, S., Gabriel, F., Gašević, D., Hennicke, T., Lim, L.-A., Male, S. A., Mirriahi, N., Oliveira, E. A., Pacitti, H., Raković, M., Russell, J., Taylor-Griffiths, D., & Yang, S. (2026). Assuring quality learning in a gen AI-integrated future: The role of adaptive capabilities. Tertiary Education Quality and Standards Agency. PDF source

Meskell, B., & Baker, N. (n.d.). Bloom’s stAIrcase. Retrieved July 16, 2026, from Source link

Pesovski, I., Vorkel, D., & Trajkovik, V. (2024). AI-powered education: Rethinking the way programming is taught using AI tools and reversed Bloom’s taxonomy. In 2024 21st International Conference on Information Technology Based Higher Education and Training (ITHET) (pp. 1-6). IEEE. DOI record

Stolpe, K., Larsson, A., & Johansson Falck, M. (2026). Discipline-Specific AI literacy (DiSAIL): A theoretical framework for situated engagement with generative AI in education. International Journal of Technology and Design Education, 36, 1917–1932. DOI record

Workmon Larsen, M. (2024, December 5). Process over product, mindset over toolset: Inverting Bloom’s taxonomy for teaching AI. Medium. Source link

Zaphir, L., Lodge, J. M., Lisec, J., McGrath, D., & Khosravi, H. (2024). How critically can an AI think? A framework for evaluating the quality of thinking of generative artificial intelligence [Preprint]. arXiv:2406.14769v1. DOI record

License

Icon for the Creative Commons Attribution-NonCommercial 4.0 International License

A Guide to Teaching and Learning with Artificial Intelligence Copyright © by Jason Blomquist; Liza Long; and Joel Gladd is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License, except where otherwise noted.