Writing the Rules for Generative Machines: Tensions and Entanglements in Preservice Teachers' Classroom AI Policies

Synthesis: Nash and Burriss (2026) offer a qualitative instrumental case study of how 27 preservice English language arts teachers in the final year of a secondary teacher preparation program wrote their own classroom AI policies for 6-12 English classrooms. Framed by postdigital theory, sociocultural literacy theory, and a process view of writing, the study treats the 27 course policies as discursive texts that expose how early-career teachers negotiate competing discourses about generative AI in English instruction. Grounded-theory coding of the policies and their accompanying reflections produced five findings: policies (1) reflected ambivalence and ambiguity about AI-mediated ideation, (2) located the central work of thinking in text production and disallowed AI composition, (3) cautiously invited AI for checking, citing, and feedback, (4) largely failed to address reading, and (5) largely permitted AI for research and inquiry. The authors surface a persistent contradiction: participants unanimously sought to protect students as critical thinkers, yet allowed AI for the ideation and revision tasks they themselves described as thinking, while locating "real" thinking in final written products. A noticeable technodeterminism runs through the policies, with even skeptical teachers arguing that AI integration was unavoidable for students' futures. The study's key contribution is methodological — policy writing as a window into writing pedagogy, AI Literacy, and the shifting boundaries of what it means to read, think, and compose.

Key Findings

• Nearly every policy permitted some AI use, but almost always on teacher terms: 26 of the 27 policies allowed generative AI for at least some tasks, overwhelmingly restricted to teacher-specified tasks, times, and places; only one participant prohibited all classroom AI use, on ethical grounds tied to labor practices and copyright.
• Ideation and brainstorming were the most broadly permitted use — and the most ambiguous: 22 of 27 policies allowed students to generate ideas, brainstorm, or prewrite with AI, while only one explicitly operationalized its stated commitment to protecting thinking by banning AI for idea generation. Limits were rarely operationalized: Willa allowed AI "to get your thinking started" but declared "this is where the line should be drawn" without specifying where.
• The work of thinking was located in text production: 22 of 27 policies disallowed, or left unclear, the use of AI to compose sentences, paragraphs, or papers. Willa wrote that "the actual writing is the most crucial part of the assignment," reflecting a view of writing as intellectual performance assessment rather than a layered thinking process.
• Checking, citing, and feedback were cautiously invited: 16 participants permitted AI for grammar checking and proofreading (Andrea allowed Grammarly "in any context"; Theresa reserved "subjective" style feedback for the teacher), while 16 prohibited producing large AI-generated text. Two raised Privacy concerns about submitting student writing to models, and one disallowed AI citation generation entirely.
• Reading was largely unexamined: 22 of 27 policies did not mention reading at all. Of the five that did, most framed AI as a conversation partner for clarifying challenging texts; only two disallowed AI for reading, citing the productive struggle and comprehension work that summarization could short-circuit.
• Research and inquiry use was permitted with verification cautions: 11 participants allowed AI-supported searching with explicit demands that students assess credibility; two disallowed it. Several noted that AI-generated misinformation makes students responsible for submitted content, and one planned a lesson on validating AI output and recognizing bias.
• Technodeterminism shaped adoption against participants' own reservations: Even the most critical PSTs argued that AI would "not go away," that teaching about it was their duty, or that it mattered for students' future jobs — a stance that, the authors argue, both responds to and reproduces an AI-centric future.
• Policies were riddled with internal contradictions: Many policies simultaneously prohibited submitting AI text and held students responsible for AI-generated content they submitted, mirroring the conflicting messaging in broader school, district, and societal discourse rather than resolving it.

Study Design & Method

• Qualitative instrumental case study using grounded theory methods (Strauss and Corbin 1990), with open coding and recursive re-readings; the aim was recurring themes in a localized context, not formal theory.
• 27 preservice English language arts teachers in the final (capstone) year of a secondary English teacher preparation program at a mid-sized rural public university in the eastern United States, a state with no state-level AI guidance at the time.
• Frameworks: postdigital theory (technology as already woven into practice), sociocultural and material literacies theory, and composition research framing writing as a process that supports thinking.
• Data: the primary sources were participants' course AI policies and their written reflections on designing them; secondary sources included reading responses, annotated AI-use work, and course materials (slides, lesson plans, handouts, readings).
• Analysis: two researchers coded independently in synchronous sessions, built an Excel matrix of allowed, disallowed, and unclear uses, then reconciled into 21 categories across 27 data sets — 567 classifications with 98% agreement (555/567), with consensus discussion and member checking for the remainder.
• Limitations: a single two-class context, no empirical evidence from participants' actual teaching placements, and policies written speculatively for anticipated classrooms rather than enacted ones; the authors caution against generalization.

What this means for practice

• Instructors. Write reading into your classroom AI policy: 22 of the 27 policies studied did not mention reading at all, leaving AI-supported summarization and comprehension ungoverned while writing received detailed rules.
• Instructors. Operationalize the boundary instead of gesturing at it — 22 of the 27 policies allowed AI ideation while only one explicitly banned it, and language such as "this is where the line should be drawn" gives students no rule to follow. Treat ambiguity as a policy defect rather than a neutral choice: contradictory rules (no AI text, but you are responsible for AI text you submit) leave students unable to comply.
• Faculty developers. Make classroom AI policy authoring a core activity in teacher preparation; it surfaces beliefs about writing, reading and Critical Thinking more sharply than abstract discussion of generative AI. Pair it with support for deconstructing writing into its component parts, since AI separates writing-as-product from writing-as-process: help teachers ask where thinking, learning and value actually live across brainstorming, drafting, revising and editing rather than locating thinking only in the final text.
• Instructors. Interrogate the assumption that AI integration is unavoidable: participants who privately doubted its value still adopted it, so establish explicit grounds on which a use can be declined and treat technodeterminism as a discourse to examine.
• Researchers. Treat teacher-authored policies and reflections as discourse data, and replicate the design in states and districts that do issue AI guidance.

Limitations

• The study is a single two-class context: 27 preservice English language arts teachers in one capstone cohort at a mid-sized rural public university in a state with no state-level AI guidance at the time.
• The policies were written speculatively for anticipated classrooms, because participants were not teachers of record, and the study gathered no empirical evidence from their actual teaching placements.
• Coding produced 567 classifications across 27 data sets with agreement on 555 (98%) between two coders, and the authors caution explicitly against generalizing to other preservice contexts.
• Because policy writing was a course assignment, what participants wrote may reflect course expectations as much as the rules they would enact in their own classrooms.

Citation

Nash, B. L., & Burriss, S. K. (2026). Writing the rules for generative machines: Tensions and entanglements in preservice teachers' classroom AI policies. Reading Research Quarterly, 61(3), e70136.