A 29-point gap that is too large to be noise
When a survey of 222 high school teachers finds 83.7 percent of English teachers worried about AI-driven academic integrity problems compared to 54.9 percent of math teachers, the 29-percentage-point gap is wide enough to rule out ordinary survey variation as the explanation. Something structural about how English and math assignments differ is driving this split, and understanding that structural difference matters more for policy design than the topline concern numbers themselves.
The survey, conducted by three Central Washington University faculty members in May 2026, captured responses from teachers actively grading student work during a period when generative AI tools were already widely accessible to students. That timing makes the gap a reflection of lived classroom experience with AI-assisted cheating attempts, not a hypothetical concern about future risk.
Why the assignment format is the real variable here
English instruction leans heavily on essays, open-response analysis, and discussion-style writing, exactly the assignment formats generative AI tools handle most convincingly, producing fluent, structurally sound prose that can be difficult to distinguish from genuine student work without careful scrutiny. Math instruction, by contrast, leans on step-by-step procedural problem-solving, where an AI-generated answer often either gets the specific numeric steps wrong in ways that are easy to spot, or requires a level of problem-specific customization that is harder for a student to generate convincingly without understanding the underlying method themselves.
That format difference, more than any difference in how seriously English versus math teachers take academic integrity generally, is the more plausible driver of the 29-point gap. The gap largely traces back to English assignments presenting a structurally larger attack surface for AI-assisted cheating than typical math assignments do, rather than to English teachers simply being more vigilant or more anxious about the issue.
Why a uniform school-wide AI policy mismatches this reality
Many schools have responded to AI cheating concerns by adopting a single academic integrity policy applied uniformly across every department, a practical approach from an administrative standpoint but one this survey data argues is a mismatch with the actual risk profile of different subjects. A policy calibrated to the exposure level of English or humanities assignments may be unnecessarily restrictive for math classrooms, while a policy calibrated to math's lower exposure may leave English classrooms under-protected.
The more defensible approach this data points toward is department-level policy customization within a school-wide integrity framework, similar to the faculty-level autonomy model gaining traction in higher education AI governance discussions. English departments likely need more specific guidance on acceptable AI use in drafting and research assistance, given how much more exposed their assignment formats are, while math departments can reasonably operate with a lighter-touch policy tuned to their different risk profile.
What this means for assessment design, not just policy
The deeper implication of this gap goes beyond policy wording into assessment design itself. If essay-based assignments are structurally more exposed to AI-generated submissions than procedural math problems, the sustainable fix is not only tighter enforcement, it includes redesigning English assessments to incorporate elements AI handles less convincingly, in-class writing components, oral defense of written arguments, or iterative drafts with visible revision history that are harder to fabricate wholesale.
Math departments, facing a lower baseline exposure level according to this data, can reasonably prioritize different interventions, potentially focusing integrity efforts on take-home problem sets specifically rather than redesigning in-class assessment formats that already carry lower AI-cheating risk. Treating both subjects with the same assessment redesign playbook wastes effort on the department facing the smaller actual problem.
The gap this survey does not resolve
This survey measures teacher concern, not actual cheating incidence, which is a meaningful limitation worth flagging directly. It is possible English teachers are more worried because they are detecting more actual incidents, or it is possible they are more worried because essay-based work simply makes them more suspicious of submissions that would have passed without scrutiny in a pre-AI classroom. The survey data alone cannot distinguish between these two explanations, though both point toward the same practical conclusion about where assessment redesign effort should concentrate.
A useful next research step, which this survey does not provide, would be pairing teacher concern data with actual confirmed AI-cheating incident rates by subject, to determine whether the 29-point concern gap tracks a genuinely larger integrity problem in English classrooms or primarily reflects a detection-difficulty difference. Schools making resource allocation decisions based on this data should treat it as a reasonable planning signal rather than a precise measurement of actual cheating prevalence, and should revisit that allocation once better incidence data becomes available rather than treating this survey as the final word.
District research offices are well positioned to close that gap relatively cheaply, since many already collect academic integrity case records as part of routine discipline reporting. Cross-referencing those existing records against subject area would give schools a genuine incidence measure to sit alongside this survey's concern measure within a single budget cycle, rather than waiting for a dedicated follow-up study to answer a question the data they already hold could largely resolve.
The broader signal for curriculum and edtech vendors alike
For curriculum developers and edtech vendors building AI-detection or assessment-integrity tools, this subject-specific gap argues against a one-size-fits-all product design. A detection or assessment tool tuned primarily around essay-format English assignments, the format where the problem is most visible in this data, risks being poorly calibrated for math and STEM subjects with a fundamentally different submission format and risk profile.
School technology leaders evaluating integrity tools should ask vendors directly how their product's accuracy and false-positive rates vary across subject areas and assignment formats, rather than accepting a single overall accuracy figure that may mask meaningfully different performance between an English essay and a math problem set. This survey's own subject-specific gap is a useful prompt for that more granular vendor evaluation conversation.
