How to Make Student Thinking Visible When AI Is Part of the Process

A student can now produce work that is clearer, cleaner or more complete than their thinking.

It might be a polished paragraph, a slide deck, a research summary, a design concept, a data interpretation, a coded solution, a visual prototype, a video script or a set of discussion notes. The work may look finished. It may even meet the criteria. Yet the teacher may still be left with the most important question: where is the learner inside this?

What did the student actually understand? What did they ask? What did they choose? What did they judge? What changed in their thinking? What are they prepared to defend?

These questions have always mattered, but AI makes them harder to ignore. Generative AI can improve the visible quality of student work while making the learning process less visible. It can help students express ideas more clearly, organise information, generate possibilities, test explanations, create prototypes, revise solutions and prepare for discussion. At the same time, research on AI and education raises concerns about learner autonomy, cognitive offloading, transparency and the need to protect human agency when AI becomes part of learning (Roe & Perkins, 2024; Viberg et al., 2026).

AI-supported learning needs a different kind of evidence. The final product still matters, but it cannot carry the whole evidentiary burden anymore. If AI is part of the learning process, then the learner’s decisions must also become part of what teachers can see.

In AI-supported learning, the process must become part of the evidence.

This is the central idea of Visible Agency. Learner agency is the goal. Visible Agency is the evidence.

Visible Agency is the observable evidence that learners remained active, thoughtful and responsible when AI was part of the learning process. It asks teachers to look for more than the completed task. It asks them to look for the questions students pursued, the choices they made, the judgements they explained, the revisions they justified and the ownership they demonstrated.

AI-supported learning should make the learner’s thinking, judgement and responsibility more visible. For the broader framework, see Visible Agency: How to Design AI-Supported Learning Without Outsourcing Student Thinking.

Why final products are no longer enough evidence of learning

Final products have always been incomplete evidence. A correct answer does not always show understanding. A strong presentation does not reveal every decision that shaped it. A working prototype does not show every design trade-off. A polished paragraph does not show every uncertainty behind it. A coded solution does not show whether the learner understands why it works. A persuasive visual does not show how the learner judged its accuracy, relevance or ethical use.

Teachers have always had to look beneath the surface of student work. Formative assessment depends on making evidence of student understanding visible enough for teachers and learners to act on it; Black and Wiliam’s work on assessment for learning emphasises the importance of classroom processes that reveal what students understand and help guide next steps (Black & Wiliam, 1998).

AI makes this more urgent because the surface can now be improved quickly. A student may use AI to strengthen structure, clarify language, generate examples, suggest revisions, organise ideas, debug code, create images, model possibilities, analyse data, plan an investigation or rehearse an explanation. Used well, this can support learning by helping students see possibilities, notice gaps and improve the quality of their work.

When AI improves the product, teachers need stronger evidence of the thinking behind the improvement.

If a student submits a clear response, a strong design, an effective presentation or a working solution, the teacher still needs to know what the student understood, decided, questioned, checked and changed. Without that evidence, it becomes difficult to know whether AI supported the student’s thinking or quietly replaced parts of it. Research on generative AI and assessment argues that educators need ways to understand how vulnerable tasks are to AI completion, especially when tasks are intended to assess critical thinking or other higher-order cognitive work (Zaphir et al., 2024).

This is why final-product assessment needs to be strengthened with process evidence. Process evidence means identifying the moments where student thinking matters and making those moments visible enough for feedback, assessment and responsibility.

The question is not simply, “What did the student produce?” The stronger question is, “What thinking does this work reveal?” That question becomes essential when AI is part of the process. For a related discussion, see How to Stop AI From Replacing Student Thinking.

What Visible Agency means

Learner agency is the broader goal. It describes the learner’s capacity to act with intention, make meaningful choices, monitor progress, respond to feedback and take increasing responsibility for learning. Research on generative AI and agency suggests that AI may support learner agency through personalisation and assistance, while also raising risks around autonomy, equity and changing forms of control in digital learning spaces (Roe & Perkins, 2024).

Visible Agency is the evidence that this agency remained present. That evidence becomes especially important in AI-supported learning because students may receive more help than teachers can easily see. AI may support idea generation, drafting, explanation, comparison, feedback, revision, design, planning, coding, modelling, visual creation, data interpretation or reflection. Some of that support may strengthen learning. Some of it may reduce the need for students to think. Teachers need a way to tell the difference.

Visible Agency gives teachers and students a practical language for that difference. It asks what the learner did with the support. It asks whether the student remained intellectually present. It asks whether their questions, choices, judgements, revisions and ownership are visible in the work.

Visible Agency is the evidence that the learner did not disappear inside the support.

Agency cannot be assumed from a finished product. It has to be made visible through the learning process. When students use AI, they should be able to show what they were trying to understand, what they chose, what they judged, what they changed and what they are prepared to defend.

That is what makes agency visible.

The Five Forms of Visible Agency

Visible Agency becomes practical through five forms of evidence: questions, choices, judgements, revisions and ownership. These are the traces of learning that help teachers see whether students remained active, thoughtful and responsible.

Together, they form the core evidence framework for AI-supported learning.

1. Questions

Questions show the direction of the learner’s thinking. When students use AI, the questions they ask matter because a question reveals what the learner is trying to understand, test, clarify, improve or challenge. It shows the purpose behind the interaction.

A completion-focused process often begins with requests such as “Write this for me,” “Give me the answer,” “Make this sound better,” “Create my slides,” or “Solve this problem.” A stronger process begins with inquiry: “What are three possible ways to explain this idea?” “What is missing from my argument?” “What assumptions am I making?” “How could this design be improved?” “What might be wrong with this data interpretation?” “What would make this solution more efficient?” “What evidence would strengthen this claim?”

The question does not need to be complex. It needs to show intention. In Visible Agency, the learner shows what they were trying to understand, test or improve. This matters because the quality of the question shapes the quality of the support. If students learn to ask better questions, AI becomes less of a shortcut and more of a thinking partner. Generative AI literacy frameworks emphasise that responsible AI use includes effective prompting, understanding the interaction with generative systems, and using AI critically rather than passively (Zhang & Magerko, 2025).

2. Choices

Choices show agency in the process. AI can generate options quickly. It can suggest structures, examples, explanations, phrases, images, sources, questions, models, pathways, revisions and solutions. The presence of options does not automatically create learning. The learning sits in how students choose among them.

Students need to show what they used, changed, ignored or pursued. They need to show where they directed the work instead of simply accepting the tool’s direction. A student might choose one explanation over another because it fits the evidence better. They might reject a suggested phrase because it sounds too general. They might adapt an image prompt because the first result misrepresented the idea. They might ignore a coding suggestion because they do not understand it well enough to use it responsibly. They might choose a design option because it better meets the needs of a particular audience.

These choices reveal the learner’s role. In Visible Agency, the learner shows what they used, changed, ignored or pursued. That evidence matters because AI-supported work can otherwise appear seamless. A final product may not show where the student made decisions. Making choices visible helps teachers see that the learner remained active in the process. Human agency in AI-supported education depends in part on human oversight and AI-human complementarity, where learners and teachers retain meaningful control rather than allowing the tool to determine the work on their behalf (Viberg et al., 2026).

3. Judgements

Judgements show the quality of discernment. AI can sound confident even when it is incomplete, generic or wrong. It can produce ideas that are fluent but shallow, plausible but inaccurate, visually impressive but misleading, technically functional but poorly understood, or helpful but misaligned with the learning purpose.

Students therefore need more than access to AI. They need judgement. Judgement is visible when students explain how they evaluated quality, accuracy, relevance or usefulness. They might compare an AI response against a source, a rubric, class criteria, their own draft, a stronger example, a data set, a design brief or a real-world constraint. They might identify what was accurate, what was missing, what needed checking, what was ethically questionable or what did not fit the task.

This is where AI-supported learning becomes more than production. The student is not simply receiving support. The student is evaluating support. In Visible Agency, the learner shows how they judged quality, accuracy, relevance or usefulness.

AI literacy is commonly framed around the capacity to understand, use, evaluate and critically reflect on AI systems, which supports the need for students to make judgement visible when AI contributes to learning (Long & Magerko, 2020; Zhang & Magerko, 2025). For a deeper exploration of this part of the framework, see How to Design AI Tasks That Require Student Judgement

4. Revisions

Revisions show the development of thinking. A revision is evidence that something changed. The student saw a gap, recognised a weakness, clarified an idea, improved a connection, strengthened evidence, refined meaning, adjusted a design, corrected an interpretation or improved a solution.

When AI supports revision, teachers need to see more than the improved final version. They need to see what changed and why. A student might explain that they revised a paragraph because the original claim was too broad. They might show that they changed a visual because the first version created the wrong impression. They might identify that AI helped them find a flaw in their code, but they still had to understand why the fix worked. They might explain how feedback helped them improve a presentation, redesign a prototype or refine a data interpretation.

The important evidence is not that revision happened. It is that the learner understood the purpose of the revision. In Visible Agency, the learner shows what changed and why the change improved the work or thinking. This protects the learning value of feedback because AI can suggest changes, but students still need to understand which changes matter and why they improve the work. The Education Endowment Foundation describes metacognition and self-regulated learning as involving planning, monitoring and evaluating one’s learning, and these processes are strengthened when students are taught to think explicitly about how they learn within subject tasks (Education Endowment Foundation, 2025).

5. Ownership

Ownership shows whether the learning has become the learner’s own. When AI is part of the process, students need to be able to stand behind what they submit. That does not mean they produced every word, idea, image, solution or structure alone. It means they understand the work, accept responsibility for the decisions behind it and have absorbed the thinking deeply enough to defend it.

This is an important test of AI-supported learning. A student may be able to describe how AI was used. They may be able to identify which suggestions they accepted or rejected. They may even be able to explain the purpose of the task. But ownership goes further than explanation.

Ownership means the student can defend the position, reasoning and final choices as their own.

They can sit inside the work without needing the tool beside them. They can explain why the argument holds, why the evidence matters, why the image represents the idea, why the design decision fits the purpose, why the solution works, why the revision improved the meaning and why the final version represents their understanding.

They are standing behind thinking they have made their own. In Visible Agency, the learner shows what they understand, defend and stand behind. This is where AI transparency becomes responsibility. The emerging literature on GenAI and agency repeatedly points to the need to preserve learner autonomy and responsibility when AI systems provide increasingly capable forms of support (Roe & Perkins, 2024; Viberg et al., 2026).

For a deeper exploration of this part of the framework, see How to Design AI-Rich Tasks That Still Require Student Ownership.

How to make AI use transparent without creating unnecessary paperwork

Transparency matters, but it can easily become too heavy. If teachers ask students to document every prompt, every response, every change and every AI interaction, the process may become paperwork rather than learning evidence. Students may spend more time recording the use of AI than thinking about the learning.

Transparency should reveal the moments where student thinking mattered. It should be brief, purposeful and connected to learning, helping teachers see the learner’s role in the work without burying the process in unnecessary records.

A student might identify one question they asked AI, one suggestion they accepted, one suggestion they rejected, one claim they checked, one revision they made or one decision they can defend. That may be enough. The key is to make the evidence proportional to the task. A major inquiry, essay, design project, media product, coding task or research investigation may need more visible process. A short classroom task may need only one sentence, one annotation or one oral explanation.

Visible Agency is learning evidence. It helps students become more aware of their own process. It helps teachers give better feedback. It helps schools move towards making student thinking visible for learning. This aligns with formative assessment principles: the purpose of collecting evidence is to inform learning, feedback and next steps, not simply to record performance (Black & Wiliam, 1998).

Transparency becomes useful when it reveals the learner’s role in the work.

How teachers can assess visible process without overburdening themselves

Teachers do not need more marking. If Visible Agency becomes another layer of workload, it will not be sustainable. The aim is to reduce uncertainty, not increase burden. Visible process should help teachers make better judgements with lighter, sharper evidence.

That might mean asking students for a brief process note instead of a long reflection. It might mean sampling one piece of evidence rather than marking every step. It might mean using a one-minute conference, an annotated draft, a highlighted revision, a short exit ticket, a design note, a code comment, a comparison between versions, or a brief oral defence of a final decision.

The evidence can be small if it is well chosen. Instead of asking students to submit a full AI log, a teacher might ask:

• What was one question you asked to improve your thinking?
• What was one AI suggestion you chose not to use?
• What was one claim you checked?
• What changed between your first version and final version?
• What part of this work are you prepared to stand behind?

These prompts are manageable because they target the evidence that matters. They do not ask teachers to inspect the whole process. They ask students to make a meaningful part of the process visible.

Teachers can also rotate the focus. One task might focus on questions. Another might focus on judgement. Another might focus on revision. The form of evidence should match the learning purpose. If the purpose is inquiry, questions may matter most. If the purpose is research, judgement and evidence checking may matter most. If the purpose is drafting, revision may matter most. If the purpose is design, choices and revisions may matter most. If the purpose is presentation or publication, ownership may matter most.

Visible Agency becomes manageable when teachers choose the evidence that best fits the task.

Classroom prompts that reveal student thinking

The simplest way to make student thinking visible is to ask better process questions. These prompts can be used in writing tasks, inquiry projects, research tasks, presentations, design challenges, media products, coding tasks, data investigations, reflections and AI-supported discussions. They are prompts teachers can choose from depending on the learning purpose.

Questions

• What were you trying to understand, test or improve?
• What question did you ask before using AI?
• How did your question change during the process?
• What did you ask AI to help you think about?
• What question would you ask next?

Choices

• What did you choose to use from the AI support?
• What did you choose to ignore or change?
• What option did you reject, and why?
• What decision shaped the direction of your work?
• Where did you keep your original thinking instead of using the suggestion?

Judgements

• What AI suggestion was most useful, and why?
• What AI suggestion needed checking?
• What did you compare the AI response against?
• What was accurate, incomplete, misleading or too general?
• How did you decide whether the response was good enough?

Revisions

• What changed between your first version and final version?
• What feedback helped you improve the work?
• Which revision made the biggest difference?
• Why did the change improve the quality of the work?
• What did you understand more clearly after revising?

Ownership

• What part of this work are you prepared to stand behind?
• What thinking has become yours in the final version?
• What support did you use, and how did you use it responsibly?
• What decision are you most responsible for?
• How would you defend your final work without AI support in front of you?

These prompts can be used as exit tickets, conference questions, annotations, peer discussion prompts, oral explanations or quick written reflections. The point is to make the learner visible.

How visible process strengthens student responsibility

Visible process is also valuable for students. When students know they will need to explain their questions, choices, judgements, revisions and ownership, they use AI differently. They are less likely to treat AI as a completion tool and more likely to treat it as a thinking support. They begin to notice their own role in the process.

Responsibility grows through attention. Students are more likely to regulate their learning when they are asked to plan, monitor and evaluate the strategies they use, which is why metacognitive and self-regulated learning approaches are strongly connected to visible process and student responsibility (Education Endowment Foundation, 2025; Nicol & Macfarlane-Dick, 2006).

If students only submit final products, they may not need to think carefully about how they used support. If they are asked to make a meaningful part of the process visible, they become more aware of their decisions. They begin to ask: What am I trying to understand? Why am I using this suggestion? How do I know this is accurate? What changed in my thinking? Can I defend this work as my own?

These agency questions help students see themselves as responsible participants in the learning process.

This is why Visible Agency matters. It gives teachers evidence, but it also gives students a way to become more conscious of their own learning. AI-supported agency needs student responsibility because the presence of support does not remove the need for ownership. It increases the importance of ownership.

Why visible process is a leadership issue

For school leaders, the challenge is to build a shared language for AI-supported learning. Without that language, conversations about AI can become too narrow. They may focus on cheating, detection, policy or tool access. Those issues matter, but they do not provide a full learning framework.

Visible Agency offers a more productive conversation. It helps leaders and teachers ask whether students are still asking meaningful questions, making choices, judging the quality of AI support, revising with understanding and taking responsibility for the final work. These questions shift the conversation from control to evidence.

They also create a practical bridge between AI use, learner agency, assessment, feedback and task design. Teachers are being asked to make the learning process visible enough to protect student thinking and responsibility.

Consistency matters. If every teacher invents a different transparency requirement, students experience confusion and staff experience unnecessary workload. If schools develop a shared language around Visible Agency, teachers can make better decisions with less friction.

The aim is to align the principles: questions, choices, judgements, revisions and ownership. These five forms give schools a shared way to talk about what should remain visible when AI is part of the learning process. AI literacy research supports the need for learners to understand, evaluate and use AI responsibly, but schools also need practical learning-design language that translates those capacities into classroom evidence (Long & Magerko, 2020; Zhang & Magerko, 2025).

Frequently asked questions

Final thought

AI-supported learning should make the learner more visible. It should give teachers and students better ways to see the learning process: better questions, clearer choices, stronger judgements, more purposeful revisions and deeper ownership.

That is the promise of Visible Agency. It asks schools to design learning so that student thinking remains visible when AI is part of the process. Because the final product no longer tells the whole story, the learner must remain visible.

Where to go next

This article is part of the Visible Agency series.

• For the broader framework, read Visible Agency: How to Design AI-Supported Learning Without Outsourcing Student Thinking.

• For the learning demand side of the work, read How to Stop AI From Replacing Student Thinking.

• For the judgement side of the work, read How to Design AI Tasks That Require Student Judgement.

• For the metacognitive side of the work, read How to Use AI to Strengthen Metacognition.