Writing comments on report cards, progress reports, and other student assessments requires teachers to compress weeks of learning into a few sentences a family can understand. At the end of a long school day, an exhausted teacher might decide to enlist artificial intelligence (AI) to help them make their comments sound warm, clear and professional.
For example, an AI chatbot might take a note that reads, “Did not answer during the whole-class math discussion but later explained the solution correctly to a partner” and offer a revision like this: “She is developing confidence in sharing her mathematical thinking and would benefit from continued encouragement to participate.” The sentence AI produced is polished and plausible. But the tool has made a claim about the student’s confidence that the teacher’s notes do not support, and the evidence in the original note—that the student could explain her solution accurately to a partner—has disappeared.
The facts have not been wildly distorted, but the meaning assigned to them has shifted. This kind of error in AI output is harder for a human to catch than a wrong name or an invented score because every detail sounds believable. A teacher who asked only for warmer language may not notice that the revised comment invites an understanding of the child that goes beyond their own observation.
Reviewing AI output must consist of more than checking grammar, tone, and recognizable facts. We must also ask whether the new sentence reflects what the classroom evidence shows.
Translation, Not Interpretation
Teachers have always needed to compress complicated classroom experiences into shorthand. Phrases such as “rarely participates,” “lacks confidence,” or “struggles with reading” are efficient, but they can erase the path between what happened and what we concluded. For instance, they don’t tell us when a student stopped participating, what behavior led us to infer low confidence, or which aspect of reading appeared to be difficult. Once those observations disappear from a teacher’s notes, no prompt can recover them later.
AI can revise a comment to make it clearer to families, but it’s not trained to make assumptions about a student’s motivation, confidence, effort, or ability.
A note that says, “identified the main idea accurately after partner discussion; immediate written responses remain inconsistent” may sound less elegant than “struggles with writing but is comfortable sharing ideas with a partner.” But it preserves something crucial — what the student did, under what conditions, and what remains unknown.
Chatbots are best treated as translators, not diagnosticians. AI can revise a comment to make it clearer to families, but it’s not trained to make assumptions about a student’s motivation, confidence, effort, or ability beyond the evidence provided. When the reason behind a student’s action is unclear, the responsible choice is to communicate the unknowns rather than fill the gap with a plausible, but potentially inaccurate, explanation.
Teachers can craft prompts that establish boundaries for AI revisions. For example, a teacher might say something like this:
“Help me turn these notes into a clear, respectful report card comment for the family. Keep what the student did, when and how they did it, and any help they received. Please don’t add information or explanations that aren’t in my notes, or make anything sound more certain than it is. Include only the next steps I’ve given you. If anything is missing or unclear, let me know separately.”
However, crafting more specific prompts is not sufficient. Teachers have a professional responsibility to evaluate a chatbot’s revisions to make sure the text it generates is accurate and complete.
Checking the Comments
Three questions can make the general guidance to “check accuracy” much more concrete for teachers using AI tools:
- What did the tool add?
- What did it remove?
- What did it make more certain?
Seemingly innocuous words that an AI tool might add to teacher-generated text can enlarge a claim: consistently requires a pattern, independently requires knowledge of what help was available, cannot implies much more than one unsuccessful attempt, and struggles with can turn a difficulty observed in a single setting into a general description of the learner.
In addition, AI might eliminate important context that tells readers more precisely what the student was able to do and under what conditions. For example, a student who retold a story accurately using picture cues might be described in an AI-generated comment as “independently retelling stories in sequence with growing confidence.” The wording is encouraging, but inaccurate: the use of picture cues has disappeared, while “independently” and “confidence” have entered the account without evidence.
Once a polished interpretation enters the record, any uncertainty behind it can be difficult to recover.
A more accurate comment might say, “She retold the story accurately using picture cues.” This preserves both the accomplishment and the support that was present, without adding unsupported claims about independence or confidence.
How Comments Turn to Action
Report card comments do not remain private drafts. Families may act on them, students may absorb them, and other teachers may use them to form their own impressions of a student’s needs. Once a polished interpretation enters the record, any uncertainty behind it can be difficult to recover.
For example, if a math student is described as needing confidence, a natural response would be providing encouragement and more opportunities to practice speaking publicly. However, if the original notes revealed that she can express her reasoning to a partner, that raises different questions about what kind of support she needs. It could be that, when speaking to a partner, she had more time to prepare an explanation or she might have learned something between the two observations. Instead of working on the student’s confidence, a teacher might offer brief preparation time before asking for a response.
To ensure that an AI-generated comment points in the right direction, place it beside the original notes and ask one final question: “Would both accounts lead another reasonable adult to understand the student in essentially the same way and offer the same kind of support?”
If not, the tool has done more than “improve” the wording. It has rewritten the teacher’s judgment. The student will never see the prompt that produced the AI-generated text, yet those words could shape her future.
For teachers, AI can make an exhausting writing task lighter and help educators communicate more clearly. But we cannot allow an AI tool to assume responsibility for deciding what a child’s performance means when the evidence leaves room for more than one answer.
