Studies comparing how students respond to feedback from instructors, peers, and AI tools consistently find a clear hierarchy of trust, with instructor feedback rated as the most authoritative and reliable source by a wide margin. Students in these studies describe using AI feedback for general structural suggestions and peer feedback mainly for readability and encouragement, while reserving their deepest trust and most careful attention for comments that come directly from a teacher. This pattern holds even in courses where AI feedback is generated using the same rubric criteria a teacher would apply.

This trust gap is not simply about accuracy, since AI feedback in controlled studies often identifies real issues in student writing just as reliably as human feedback on structural and mechanical dimensions. Instead, the gap seems to reflect something about authority and relationship: students know a teacher understands the specific assignment, the classroom context, and their individual growth over time in a way that a general-purpose AI tool cannot. A comment that says an argument needs stronger evidence lands differently coming from a teacher who has been reading a student's work all semester than from a tool encountering the essay for the first time.

For schools weighing how heavily to lean on AI-generated feedback, this research points toward a specific design principle rather than a reason to avoid AI tools altogether. Feedback that students believe came directly from their teacher, because the teacher reviewed and personalized it before it was sent, carries the same trust and authority as any other teacher comment, even if AI helped generate the first draft of that feedback. The moment feedback is delivered as unmediated AI output, without visible teacher involvement, it loses much of the weight that makes students actually act on it.

What This Means for Feedback Design

Practically, this suggests that AI-generated first-pass feedback works best as a draft the teacher edits and personalizes rather than a message sent to students with minimal review. A teacher who takes AI-drafted comments and adds a specific reference to a student's earlier work, adjusts the tone to match how they normally communicate, or flags one particular strength worth calling out, transforms generic feedback into something a student will recognize as genuinely theirs. That small investment of teacher time preserves the trust relationship that makes feedback actually influence revision.

• Review and personalize AI-generated feedback before it reaches a student, rather than sending it unedited
• Add at least one specific, individualized comment that only a teacher familiar with the student could write
• Keep feedback delivery consistent with how a teacher normally communicates, so students recognize the source
• Use AI feedback to handle structural and mechanical observations, freeing teacher time for higher-value comments
• Explain to students how AI is used in the feedback process, since transparency tends to preserve rather than undermine trust

The Value of Combining Feedback Sources

Rather than treating teacher, peer, and AI feedback as competing options, research on formative feedback suggests students benefit most when these sources work together, each contributing a different kind of value. Students report using AI feedback for broad structural suggestions early in the revision process, then relying on teacher feedback for targeted, assignment-specific improvements, and finally turning to peer feedback for readability and formatting checks closer to a final draft. This layered approach lets each feedback source do what it does best rather than asking any single source to do everything.

Teachers designing a writing unit can build this layering intentionally rather than leaving it to chance. An early draft might go through an AI-assisted first pass focused on organization and evidence, a middle draft might receive teacher feedback on argument quality and depth, and a final draft might go through peer review focused on clarity and mechanics. Structuring the sequence this way makes the role of each feedback source explicit to students, which research suggests improves how seriously they take each round.

Why Revision Behavior Still Depends on the Teacher

Even when AI feedback identifies the same issues a teacher would flag, some studies find students revise less thoroughly in response to AI feedback alone than they do in response to teacher feedback. This gap is particularly pronounced on sections that require deeper conceptual revision rather than surface editing. The gap in revision behavior, not just in trust ratings, is a strong argument for keeping teacher involvement visible throughout the feedback and revision cycle rather than treating AI feedback as a substitute for teacher engagement at any stage.

The practical takeaway is not that AI feedback fails to help students improve their writing, since the evidence on structural and mechanical gains from AI feedback is generally positive. Rather, the improvement is strongest when AI feedback supports and extends a teacher's presence in the feedback loop rather than replacing it. This reinforces why a human-in-the-loop model produces better outcomes than either a fully automated system or purely manual grading on its own.

Applying This in the Classroom

Teachers adopting AI grading tools should think carefully about how the feedback is framed to students, since the framing appears to matter as much as the content itself. Presenting AI-assisted feedback as something the teacher reviewed and stands behind, rather than something generated and sent automatically, preserves the trust relationship that drives genuine revision. A short note explaining that a comment was reviewed by the teacher, even briefly, can make a meaningful difference in how seriously a student engages with it.

This is ultimately a case for transparency rather than concealment. Students do not need to be unaware that AI played a role in generating first-pass feedback; they need to trust that a teacher's judgment shaped what actually reached them. Building that trust explicitly into the classroom workflow, rather than assuming it will happen automatically, is what turns AI-assisted feedback into something students genuinely act on rather than something they skim and dismiss.