top of page

Can AI Help Teachers Give Feedback Students Actually Use?

2 hours ago
4 min read
A teacher and two students discuss annotated drafts together at a classroom table. Illustrative image generated with AI.

A student turns in a clear history paragraph with one problem: the evidence supports a different claim from the one in the opening sentence. An AI tool could generate a page of comments in seconds. The teacher has a smaller and more important decision to make. Which comment will help this student see the mismatch and work out what to do next?


That is the standard I would use before introducing AI into classroom feedback. A first draft of a comment can reduce the work of getting started. The educational value depends on whether the comment accurately reflects the student's thinking, fits the learning goal, and leads to a decision the student can make.



What the November review found


Stanford's November 10 review of the evidence on AI in education added 32 papers to the 20 identified in its earlier review. The studies varied in setting, tool, and what they measured. A study designed to detect an effect in one classroom or university cannot settle what will happen in a different grade or subject.


Two examples involving teaching assistants in U.S. universities help explain the potential. In one, AI drafts made feedback more likely and comments longer, while students rated usefulness similarly. In another, AI suggestions helped teaching assistants provide more feedback that covered more of the rubric. Students produced higher-scoring revisions, but their later post-test performance did not show a statistically significant improvement compared with the study’s comparison condition. A better revision and a skill a learner can use again are different outcomes worth measuring separately.


The review also describes a German secondary-school study of writing feedback that detected no significant differences in revisions or subsequent writing performance between students receiving AI-generated and teacher feedback. Results across age groups and settings leave room for careful experimentation. They also leave teachers with the responsibility to check what any individual comment is doing for the learner.


Start with one decision the student can make


Consider that history paragraph. A useful comment might ask the student to underline the sentence that contains the main claim, then identify the evidence that most directly supports it. The student might revise the claim or look for better evidence. Both choices require reasoning. Supplying a polished replacement paragraph would remove that choice.


The idea works beyond writing. In a science explanation, ask the student to mark where an observation ends and an inference begins. In mathematics, ask which line of a solution assumes something that still needs to be shown. The comment becomes an invitation to inspect a particular part of the work. It gives the teacher something specific to look for when the student returns.


Read the work before accepting the suggestion


AI can misread a student's meaning, overlook a correct unconventional method, or offer a confident critique that the rubric does not support. The teacher who reads the work first can compare a suggested comment with their own diagnosis. For a small classroom trial, I would keep the rubric beside the assignment and ask whether the draft identifies the right issue, points to actual evidence, and uses language the student can act on.


A comment can be technically correct and still badly timed. If a student is struggling to structure an argument, six notes about punctuation might divert attention from the next big decision. Different students may need different kinds of feedback on the same assignment. That judgment comes from knowing the work, what has already been taught, and what the learner is ready to try.


Use only school-approved tools and follow the school's rules for student work. A teacher can test the approach with a fictitious sample or a teacher-created example before deciding whether the tool has a place in a real feedback workflow. Students should also know what role the tool has and how they can ask about a confusing comment.


Give feedback somewhere to go


Build a few minutes for revision into the next lesson. Ask students to point to where they acted on feedback and explain the choice in a sentence. A student can also explain why a suggested change did not fit. That conversation can surface misunderstandings earlier than a pile of corrected assignments.


A later, related task adds a separate check. In the history example, ask students to match a new claim to evidence and explain the connection without the original comment in front of them. Review a small sample of revisions and new tasks. Record how much teacher time the process actually required, including time spent correcting suggestions. The result can inform the next unit without pretending that a classroom check is a research trial.


Ask students what they thought a comment meant before deciding whether it worked. A teacher may read a note as a question, while a student reads it as an instruction to make a particular change. That difference matters when the goal is to build judgment.



Keep a clear reason for using it


A sensible first experiment is one class, one assignment, and one rubric criterion. Compare the suggested comments with the comments you would want students to receive, then look at how students respond and what they can do later. If the comments are less specific than a short conversation or create extra review work, that is useful information for the next decision.


Feedback earns its place when a student understands what to try next and can explain why. That is the result to look for before widening the use of any tool.


Illustrative image generated with AI.


What’s next in education—and what does it mean for your school?

Explore AI, school leadership, and the future of learning—with practical ideas delivered to your inbox.

  • twitter
  • linkedin

© 2020–2026 What Is Next in Education. All rights reserved.

bottom of page