The risk of students using AI to cheat tends to get a lot of attention – with good reason.
A student can simply copy and paste a prompt into a chatbot and receive a polished paragraph, a five-paragraph essay, a lab summary or a reading response almost instantly. Teachers may then be left wondering whether the work reflects the student’s thinking and actual work or what the chatbot generated.
An estimated 84% of high school students surveyed said they had used generative artificial intelligence for schoolwork in 2025, according to College Board, a nonprofit that administers the SAT and AP exams.
As an assistant professor of school psychology studying artificial intelligence in K–12 education, I think the question is not only whether students are using AI to cheat, but whether there is evidence that learning actually happened.
I recently surveyed public school educators and administrators about how generative AI is affecting schools to better understand the answer to this question.
My study, conducted from spring 2025 to spring 2026, included 303 educators and other school professionals in Wisconsin – teachers, administrators, IT staff and technology directors, as well as school psychologists and counselors. I also surveyed another 132 professionals at schools across the country.
The results are not nationally representative, but they offer a snapshot of how some K–12 professionals are thinking about AI and student learning.
While a large number of respondents were concerned about AI bias, misinformation and data privacy, the most common worries were about academic dishonesty and plagiarism.
In Wisconsin, approximately 65% of respondents identified these issues as a concern, compared with 74% who did so on a broader, national level.
But respondents also pointed to a deeper issue: How do teachers know what students actually understand when AI can generate essays, summaries or math steps in seconds?
In the Wisconsin sample, 47% of respondents who answered this question said that “difficulty in assessing student learning when AI is used” is a concern.
That figure increased to 53% in the national sample.
When asked “What impact, if any, have you noticed AI has had on student behavior, mental health, or engagement?” respondents selected from a provided list of options. Among those options, 29% of Wisconsin respondents and 40% of respondents in the national sample selected “increased student reliance on AI,” while 19% and 33%, respectively, selected “reduced critical thinking or problem-solving.”
Teachers have long known that a student’s finished assignment is not perfect evidence of learning. A parent might help too much. A student might copy from a friend. A student might complete the work but not understand it well enough to explain it later.
Generative AI makes that problem more visible and more complicated.
Take a common homework task, such as writing a paragraph explaining the theme of a short story. In the past, teachers looked at students’ writing to understand whether they read the story, thought about the theme and could explain it in writing.
Now, this kind of homework prompt may produce a result that appears organized, accurate and polished. But it is becoming harder for teachers to understand whether students actually understood the story, identified the theme and articulated it independently, or whether students simply entered a prompt into an AI tool.
Some teachers do use AI-detection tools to determine whether students’ work is original.
In a 2025 national survey of sixth- through 12th-grade public school teachers, 43% reported used these kinds of apps regularly, while another 27% had tested or experimented with them.
But these tools can make mistakes in both directions. One study of 14 AI-detection tools found false-positive rates as high as 50% and false-negative rates as high as 100%, depending on the tool. The same study found that about 20% of AI-generated texts were misclassified as human-written; that rose to about 52% when AI-written text was manually edited and 71% when it was machine-paraphrased. Other researchers found that detectors falsely flagged nonnative English writing as AI-generated at an average rate of 61.3%.
I don’t think that means schools should abandon writing assignments or homework altogether. But educators may need to be more intentional about what each assignment is supposed to measure.
Some teachers are already making those kinds of changes, including asking students to show or explain their process, or asking them to include oral components to their written work or write more in class.
Some teachers are also giving students paper-and-pencil tasks when they need to see students’ independent thinking.
If the goal is writing fluency, teachers may need to see students write. If the goal is reading comprehension, they may need students to explain, apply or defend their thinking.
Many schools are still deciding how to approach AI. In my survey, only 33% of Wisconsin respondents and 29% of national respondents said their district had a formal AI policy.
Teachers and students alike could benefit from clarity on how and when they can use AI.
Researchers who developed the Artificial Intelligence Assessment Scale, a tool that helps educators spell out when and how students can use AI on an assignment, have argued that educators should identify what level of AI use makes sense based on the learning outcomes being measured.
This mindset is useful because not all assignments are the same. One assignment might require no AI use because the teacher needs to see independent writing.
Another might allow AI for brainstorming but require students to submit original notes and a final reflection. Another might ask students to critique an AI-generated answer and explain what is accurate, incomplete or misleading.
The educators in my survey were not simply rejecting AI. Many reported using AI themselves for planning, communication, documentation, differentiation, administrative tasks and student-support activities.
Their concerns were more specific.
They were worried about academic dishonesty but also about assessment, student reliance, critical thinking, misinformation and privacy. Those concerns point to a practical challenge schools now face: how to preserve meaningful evidence of learning when AI can produce polished academic work.
The goal is not to catch every possible misuse of AI. That is likely impossible. The goal is to design learning tasks where teachers can still answer the question that matters most: What does this student actually understand?
Assistant professor of psychology and education, University of Wisconsin-Stout Polytechnic
Brett DeJager does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.
University of Wisconsin-Stout Polytechnic provides funding as a member of The Conversation US.