Research Article
Exploring the Impact of Generative AI on Curriculum Design and Instruction: A Study With Preservice Teachers
Synthesis: Gold, Winn and Marshall Arnold (2026) report an interpretive study of 70 preservice teachers in the senior methods block at the University of Dayton who attended a 90-minute seminar on using GenAI for lesson planning, co-taught by four methods faculty in science, social studies, mathematics and literacy. Data came from a presurvey, an exit ticket, prompt histories and an end-of-semester survey completed by 59 of the 70 after roughly 150 hours of field work. The authors describe a shift from seeing the tool as a lesson-planning shortcut to seeing it as support for differentiation, assessment and parent communication. The number of participants reporting no or limited GenAI use fell from 15 to 8, and 67 of the 70 refined prompts across two to four rounds per lesson. The paper's claim is that structured professional development raises GenAI literacy and confidence, and that teacher preparation should therefore teach the tool rather than leave it to chance.
Key Findings
- The perceived uses widened, and lesson planning stayed central. Lesson planning and activity ideas were the most-cited uses before the seminar and remained so afterward, while assessment support, classroom management and parent communication all gained mentions. The authors read the pattern as tentative familiarity turning into concrete strategies rather than a change of subject.
- Skepticism fell by half. Participants reporting no or limited GenAI use dropped from 15 to 8 between the presurvey and the exit ticket. The authors treat that as the clearest signal that a short, hands-on session can move students who were unsure, not only those already interested.
- Prompt refinement was the mechanism, and it was near-universal. 67 of the 70 participants used two to four rounds of prompting per lesson, starting from broad requests such as a grade-level science lesson and adding instructional models, differentiation and technology integration. Documented refinements included adding the 5E model, adding group work with student roles, and modifying a lesson for English language learners or students with individualized education programs.
- Subject matter clustered in the core elementary areas. 60 lesson requests fell in science, mathematics and literacy, covering read-aloud comprehension, place value and ecosystem modeling. That clustering reflects the methods courses the seminar was attached to as much as student preference.
- Output was edited, not submitted. Several participants moved generated text into Google Docs for further rewriting, and the authors read this as evidence that the tool functioned as a planning collaborator rather than a finished product. It is the behavior they treat as the marker of GenAI literacy.
- The field placement extended use beyond planning. In the end-of-semester survey, lesson planning remained the main use, with professional communication — letters and emails to families — and resource creation such as rubrics and worksheets named repeatedly. Participants described gains in creativity, efficiency and differentiation, and one framed the benefit as time returned to students.
- Critical awareness rose alongside use. Exit tickets surfaced concerns about student misuse, accuracy of generated content and overreliance, which the authors connect to the privacy, bias and equity cautions in the literature. They present this as part of the learning rather than resistance to it.
- The seminar itself was not uniformly useful. Some participants found it clarifying; others said it covered material they already knew. The authors take that split as an argument for differentiated professional development rather than a single compulsory session.
- The authors' conclusion is a curriculum recommendation. They argue that teacher preparation programs should model GenAI use in their own courses, build hands-on lesson-planning tasks into methods coursework, and keep reflective discussion of ethics alongside the technical training.
How the study was done
The design is interpretive and constructivist, built around one open question: how preservice teachers use GenAI to support lesson planning. After institutional review board approval, 74 preservice teachers in the senior methods block attended a seminar titled Using Generative AI to Support Student Teachers in Lesson Planning. Of those, 70 agreed to participate and completed the presurvey and exit ticket, and 59 completed the follow-up survey at the end of the semester. Participants were enrolled in preschool-to-fifth-grade (n = 64) or intervention specialist (n = 10) licensure programs. No self-reported demographics were collected; the authors note that the group reflected the national teaching workforce in being mostly White women, without reporting the distribution.
The seminar followed the 5E inquiry model: open exploration of GenAI platforms, sharing of early learning, instruction in prompt design and scaffolds, a second round of exploration using those practices, and a close on challenges and ethics. The presurvey asked how students used AI, what they would never use it for, and where they saw potential in teaching; the exit ticket asked about emotions, intentions, new knowledge and remaining questions; the final survey asked how AI had actually been used during the semester. Chat histories were submitted as shared links and coded with focused coding, while survey responses were coded, reduced to themes and counted. Codes came from the literature, from the seminar and from participants' own wording. The authors state plainly that 90 minutes could not change practice and that they were measuring how participants conceptualized their relationship to the tool, not observed classroom behavior.
Context matters for reading the results: in fall 2024, when the data were collected, the university was only beginning its GenAI conversations, and it did not establish a preferred campus tool or require syllabus statements until fall 2025.
What this means for practice
- Teacher educators. A single 90-minute session with open exploration before instruction moved self-reported intentions and breadth of use. That is a cheap intervention, but the authors' own caution applies: it changes what students say about the tool, not what they do in a classroom, so a seminar should be paired with the field placement rather than treated as the whole treatment.
- Methods instructors. The prompt histories are the assignment worth copying. Requiring students to submit the conversation turns prompting into an assessable artifact and gives the instructor evidence of whether refinement happened or the first output was accepted.
- Program designers. Teach the three uses students discovered late — assessment support, classroom management and family communication — explicitly. The presurvey showed those were the least visible to students, and they were also where the exit ticket showed the largest proportional gains.
- Faculty developers. Differentiate the session. Experienced students found the content redundant, and a single track wastes their time while under-serving the rest; a short diagnostic before the seminar would separate the two groups.
- Department chairs. Expect the ethical questions to arrive with the enthusiasm. Participants raised misuse, accuracy and overreliance without being prompted to, which suggests ethics content belongs in the same session as the technical content rather than in a separate policy briefing.
Limitations
- Single institution, single cohort, one semester. There is no comparison group and no random assignment, so the reported changes cannot be separated from the effects of the methods courses, the field placement or simple maturation across a semester.
- The intervention was 90 minutes. The authors acknowledge that this cannot produce durable practice change, and the follow-up measures what students said they did, not what they did.
- Most outcomes are self-reported, and the reporting stays descriptive. Prompt histories provide one behavioral trace, but the study reports no observation of planning sessions and no independent rating of the lesson plans produced. Figure 1 presents mention counts without statistical testing, and the paper does not report inter-coder agreement or a coding audit trail for the thematic analysis.
- The sample composition is unexamined. The final survey had 59 respondents against 70 at baseline, and no analysis of who dropped out is reported, so attrition bias is unaddressed. No participant demographics were collected either, which prevents any analysis of whether the gains differed by program, background or prior experience.
- The sample is preservice, not practicing, teachers, and the elementary and intervention-specialist focus limits transfer to secondary or disciplinary preparation.
- The study measures no student outcome. Nothing here shows whether lessons planned with GenAI were better for elementary learners, which is the claim a teacher preparation program ultimately needs.
Citation
Gold, L. A., Winn, V., & Marshall Arnold, J. (2026). Exploring the Impact of Generative AI on Curriculum Design and Instruction: A Study With Preservice Teachers. Journal of Teaching and Learning with Technology, 14(Special Issue), 35–47.