AI as a Critique Partner

In the previous post, I focused on using AI as a tutor rather than an answer machine. AI can explain concepts, ask questions, provide hints, and generate opportunities for practice. But for that interaction to support learning, the student has to remain active.

The goal is not simply to reach the answer. The goal is to become more capable of reaching an answer. That same principle applies when students use AI for feedback.

AI can review a draft, examine a proposed solution, identify possible weaknesses, raise questions, and suggest alternatives. It can respond almost immediately, which gives students opportunities to receive feedback while they are still engaged in the work.

That has real potential. But feedback is not automatically learning. What matters is what students do with it. This is why I prefer to think of AI as a critique partner rather than an editor. An editor often moves directly toward improving the product. It corrects sentences, reorganizes ideas, rewrites unclear passages, and attempts to create a more polished result. A critique partner has a different purpose.

A critique partner helps the student see the work more clearly. It identifies places where the reasoning is difficult to follow. It asks whether the evidence supports the conclusion. It points out assumptions, tensions, missing perspectives, and decisions the student may need to reconsider.

It does not simply make the work better. It helps the student decide how the work should become better. That distinction matters because AI can move from feedback to rewriting very quickly. A student might ask, “How can I improve this paragraph?” AI may respond with a fully revised paragraph. The new version may be clearer. It may use stronger transitions, more precise vocabulary, and a more professional tone. But the student has been given the solution before identifying the problem.

They may replace the original paragraph without understanding what was weak, why the revision works better, or whether the new version still communicates what they intended to say. The product improves. The student may not. A critique-oriented interaction would look different.

The student might ask: “Identify the two places where my reasoning is least clear. Do not rewrite the paragraph. Explain why a reader might have difficulty following it and ask me questions that could help me revise it.”

Now AI is being asked to make the problem visible rather than make the problem disappear. The student still has to interpret the feedback. The student still has to make decisions. The student still has to revise. That work is where much of the learning occurs.

This approach can extend well beyond writing. A student proposing a network design could ask AI to identify possible points of failure. A student recommending a business strategy could ask what assumptions would need to be true for the strategy to succeed. A student analyzing a historical event could ask which perspectives are missing from the analysis. A student developing a lesson could ask where the activity may fail to align with the intended learning outcome. A student troubleshooting a computer could provide a proposed diagnosis and ask what evidence would contradict it.

In each case, AI is not being asked to take over the task. It is being asked to apply pressure to the student’s thinking. That pressure can be valuable because we do not always see the weaknesses in our own work.

Once we have developed an idea, we know what we intended to communicate. We may fill in missing connections without realizing they are absent from the page. We may become attached to an early solution. We may interpret evidence in a way that supports the direction we already chose. A critique partner can interrupt that pattern. It can ask questions we did not think to ask. It can adopt the perspective of a skeptical reader, a customer, a supervisor, a user, or someone affected by the decision.

It can generate counterarguments. It can identify conditions under which a recommendation might fail. It can help us see the distance between what we intended and what we actually produced. But AI critique comes with an important limitation.

The tool is often highly responsive to the way the student frames the request. If a student asks, “Why is this a strong argument?” AI will probably identify strengths. If the student asks, “What is wrong with this argument?” AI will probably identify weaknesses. If the student says, “I think this is the best solution,” AI may generate reasons to support that belief. If the student says, “I no longer think this will work,” AI may generate reasons the same idea should be rejected. The response can change with the framing even when the underlying work has not changed.

Research on large language models has identified a related behavior often called sycophancy. In this context, sycophancy refers to a model’s tendency to align its response with the user’s stated view rather than consistently prioritizing the strongest or most accurate conclusion. In other words, AI may sometimes tell us what we appear to want to hear.

That creates a problem when the tool is being used as a critique partner. A student may ask AI to evaluate an idea but unintentionally signal the answer they want. “I think this is a strong thesis. What do you think?” “This seems like the best solution. Do you agree?” “I believe this evidence proves my conclusion. Can you explain why?” The AI response may feel like independent confirmation even though the student’s framing helped produce it.

This can create a kind of feedback loop. The student expresses a belief. AI validates the belief. The student interprets the validation as evidence. The original belief becomes stronger. But no genuinely independent evaluation occurred.

Students need to learn how to request critique that does not begin by asking for agreement.

They might say: “Evaluate this argument without assuming my conclusion is correct.” “Identify the strongest reason to reject this recommendation.” “Describe what evidence would weaken my position.” “Apply the assignment criteria and identify where the work does not yet meet them.” “Generate two competing interpretations and explain what evidence would help distinguish between them.” “Do not praise the work. Focus on questions I need to answer before revising.”

These prompts do not guarantee an objective or accurate critique. AI may still misunderstand the task, overemphasize minor issues, or invent weaknesses that do not matter. But they can create more productive resistance. That resistance is important because the most helpful feedback is not always the feedback that feels best to receive.

AI-generated praise can make students feel confident. That confidence may be useful when a student is hesitant or discouraged. But generic encouragement is not the same as evidence that the work is strong. “This is a thoughtful and well-structured response” may sound reassuring, but what specifically makes it thoughtful? Which parts are well structured? Where does the reasoning become weaker? Does the response actually meet the assignment criteria?

Students need feedback tied to something they can examine. A specific passage. A decision. A piece of evidence. A stated criterion. A visible gap between the goal and the current work. Without that connection, AI feedback may sound meaningful without giving the student a clear basis for action.

The opposite problem can occur as well. AI can produce a long list of suggested improvements whether the work needs them or not. Because the system is designed to be helpful, it may find something to revise even when a section is already effective. A student may interpret the number of suggestions as evidence that the work is poor. Or they may assume every suggestion should be applied. That can lead to revision without judgment.

The student changes the introduction because AI suggested a different opening. They replace words because AI proposed more formal alternatives. They reorganize the paper because AI generated another structure. They add qualifications, examples, and transitions because the tool recommended them.

The result may be more polished. It may also be less focused, less authentic, or less representative of what the student intended to communicate. Good feedback does not eliminate the need for choice. It creates the need for better choice. Students should not ask only, “What did AI tell me to change?” They should ask: What problem is this suggestion trying to solve? Do I agree that the problem exists? Would the suggested change actually improve the work? What might be lost if I make it? Does the feedback align with the purpose and criteria of the assignment? Is the suggestion addressing meaning, or only surface appearance? What decision will I make? These questions help students engage with feedback rather than simply comply with it.

That distinction is sometimes described as feedback literacy: the ability to understand, evaluate, and use feedback effectively. Feedback literacy has always mattered. Students have never benefited simply because an instructor wrote comments on a paper. They benefit when they interpret those comments, connect them to the work, decide how to respond, and apply what they learned to future situations.

AI makes this ability even more important because feedback is now abundant. Students can request feedback at any point. They can ask for it repeatedly. They can receive suggestions on every sentence, every decision, and every stage of a task. The challenge is no longer simply obtaining feedback. The challenge is deciding which feedback deserves attention. That requires students to understand the goal of the work before asking AI to evaluate it. If students do not know what they are trying to accomplish, they have no stable basis for deciding whether the feedback helps.

AI may then become the source of both the criteria and the revision. It decides what good work looks like and then transforms the work to match that definition. The student becomes a spectator to the improvement process. A stronger approach begins with explicit criteria. What is the assignment asking the student to demonstrate? Who is the intended audience? What constraints need to be respected? What qualities matter most? What decisions has the student already made? Students can provide that information when asking AI for critique. They can ask the tool to apply specific criteria rather than a vague idea of improvement.

For example:

“Review this recommendation according to these three criteria: cost, reliability, and future upgradeability. Identify where I have provided enough evidence and where I need more. Do not recommend different components yet.”

Or:

“Read this discussion post using the assignment requirement that I connect the course concept to a real situation. Identify whether I made that connection clearly and ask one question that would help me deepen it.”

Or:

“Review my troubleshooting process. Do not tell me the correct diagnosis. Identify the point where I made an assumption without enough evidence.”

These requests keep the learning outcome at the center. They also limit AI’s role. The tool is not being given permission to transform the entire product. It is being asked to help the student examine a particular part of their thinking.

Students can then create a feedback decision record. The record does not need to be long. It might include three simple categories:

·      Accepted.

·      Modified.

·      Rejected.

For each meaningful suggestion, the student briefly explains the decision. I accepted this feedback because it identified a claim that lacked evidence. I modified this suggestion because the original example was useful, but the connection needed to be clearer. I rejected this recommendation because it would change the intended audience and purpose of the work. This makes the student’s judgment visible. It also reminds students that rejecting feedback can be a thoughtful decision. AI feedback should not carry automatic authority.

A student who rejects every suggestion without consideration is not engaging productively. But neither is a student who accepts every suggestion simply because it came from a system that sounds knowledgeable. The learning appears in the evaluation. That evaluation becomes especially useful when students compare AI feedback with other sources.

How does the AI response align with the assignment rubric? How does it compare with peer feedback? Does it match what the instructor emphasized? Does testing support the AI’s critique? Does the original source confirm the issue? Would a person from the intended audience interpret the work the same way? AI should be one voice within the feedback process. It should not automatically become the final voice.

Human feedback remains important because people understand context in ways AI may not. An instructor knows what students have learned, what the assignment is intended to develop, and what misconceptions have appeared across the class. A peer may respond as a real reader rather than a simulated one. A supervisor may understand organizational needs that were never included in the prompt. Human feedback can also communicate care, expectations, and belief in the learner.

AI can imitate the language of encouragement. It cannot replace the significance of another person choosing to engage seriously with a student’s work. At the same time, AI can extend the feedback process.

A student might use AI before submitting a draft so that instructor feedback can focus on deeper issues. They might use it after receiving human feedback to explore a concept they did not understand. They might practice responding to critique before participating in peer review.

The goal does not have to be choosing between AI feedback and human feedback. The goal can be determining what each contributes. AI offers speed, availability, repetition, and the ability to simulate multiple perspectives. Human feedback offers context, relationship, disciplinary judgment, and genuine accountability.

Productive feedback environments may use both. But students still need ownership of the work. That ownership is not preserved merely because they wrote the first draft. It is preserved when they continue making the decisions that shape the final version. They need to know why something changed. They need to understand what the change accomplishes. They need to be able to defend the decision. They need to recognize when a revision improves the work and when it merely makes the work sound more like AI.

This is why AI may be most valuable as a mirror, a questioner, or a source of productive friction. It can help students see what they have done. It can make possible weaknesses visible. It can test the strength of an idea. It can introduce a perspective the student has not yet considered. But it should not quietly become the author, decision-maker, or owner of the revision.

A useful critique partner does not leave the student with a perfect product. It leaves the student with better questions. What am I trying to communicate? Where is my reasoning weakest? What evidence do I still need? What assumptions should I reconsider? Which feedback will I act on? What decision remains mine?

Those questions keep the student inside the work. And that is the larger goal of learning alongside AI. Not simply to produce better answers, but to develop students who can evaluate, revise, and take responsibility for what they produce.

AI can contribute to that process. It can challenge. It can question. It can expose weaknesses. It can suggest possibilities. But the student still has to decide. Across this series, that decision has appeared repeatedly.

Students have to decide how to frame a prompt, what thinking to offload, how to verify a response, how much trust it deserves, what role AI should play in collaboration, how to use it as a tutor, and which feedback to accept or reject.

These are all parts of a larger capacity. Discernment.

That will be the focus of the final post in this series: knowing when, why, and how to use AI and when the better decision may be not to use it at all.

Continuing the Conversation

Series 1: AI Is Exposing Existing Problems ✓ Completed
Series 2: What We Do About It ✓ Completed
Series 3: Cultivating Human Thinking ✓ Completed
Series 4: Learning Alongside AI
Current Post (7 of 8): AI as a Critique Partner
Next Up: Discernment: Knowing When and How to Use AI