Relief or displacement? How teachers are negotiating generative AI's role in their professional practice

Synthesis: Dangol and colleagues interviewed 22 teachers in a large U.S. public school district that had been encouraging generative AI since late 2023, most of them working without a formal district-wide classroom policy. The teachers were neither adopting nor resisting the technology wholesale. They were negotiating it: trading relief from bureaucratic workload against a feared displacement of the intellectual and relational work that makes teaching a profession. The paper names two tensions — relief versus displacement, and mediating student use while building one's own expertise — and locates them in district policy, peer networks, professional norms, and relational commitments. Its central contribution is to show that the "adopt and train" playbook misreads the problem. Teachers already exercise professional judgment about where AI belongs; the question for designers and administrators is whether institutional conditions support that judgment or leave it to individual initiative.

Key Findings

1. Teachers negotiated rather than adopted or resisted. Thematic analysis of 22 interviews produced 10 top-level codes and 45 subcodes across four rounds of coding, and the resulting account describes situated decisions rather than a binary uptake. Teachers moved between embracing, conditional, and critical stances depending on the task, the class, and what they believed the tool could do.
2. District permission was the precondition for experimentation. Formal endorsement, professional development offerings, and the assurance that use would not be reprimanded changed teachers' calculations. One middle school science teacher with 11+ years of experience described the district push as license to continue: she would not "get in trouble for using it." Salary-advancement courses that carried credits toward pay increases tipped others toward trying genAI when time was scarce.
3. Relief from workload was the strongest pull. Teachers described 80-hour weeks of grading, planning, and paperwork, and framed genAI as a way to protect the energy they bring to students. A high school CLDE teacher with 11+ years of experience credited it with "work-life balance" and a decrease in stress of "80 fold," describing the tool as a thought partner in an isolating job.
4. Supporting multilingual learners was a distinct adoption driver. With 42% of the district's students classified as second-language learners, teachers used genAI to translate documents, generate sentence stems, and produce visual supports. One STEM teacher described classes containing four languages while speaking only English, and newcomers arriving "every week."
5. Reluctance had four separable sources. Lack of time to learn new routines and to iterate on prompts; the stigma of "cheating" voiced by students, parents, and colleagues; a preference for established, low-tech practice; and environmental concern. The last was a minority position but explicit — one teacher reported cutting back after learning about water use, another called the environmental impact "horrifying."
6. Teachers drew boundaries to protect judgment and relationships. They treated genAI output as a starting point requiring review, rejected unreviewed lesson plans as unacceptable, and separated assessment of growth from assessment of mastery: AI was permissible while students were practicing, not when they had to demonstrate a skill unaided. Several redacted names and personal details before pasting student work, and some preferred to grade alone rather than expose student data.
7. Sustainable integration was described as a system property, not a teacher trait. Teachers asked for clear district rules, approval lists for tools, professional development led by people who understand classrooms, student AI literacy curricula, and protection of the relational core of learning. Because access to these supports varied by school within one district, the authors argue that genAI can produce within-district inequities when implementation depends on individual initiative.

A district that moved before it wrote rules

The study site, anonymized as BPS, is a large urban-suburban district serving more than 38,000 students across 59 schools as of the 2024–25 school year, including charter and magnet programs. Its student body comes from over 130 countries and speaks more than 160 languages; 74.6% qualify for free or reduced-price lunch, 14.5% receive special education services, and 4.9% are identified as gifted and talented. Since early 2023 the district had encouraged genAI use and adopted MagicSchool AI district-wide, a teacher-oriented platform with guided workflows, while many teachers also used general-purpose systems such as ChatGPT and Gemini. Interviews ran from March through May 2025, before any formal district-wide classroom policy existed.

Recruitment ran through the district's Director of Educational Technology, followed by purposive sampling from a screening survey. The 22 participants spanned elementary, middle, and high school across subjects. Teaching experience ranged from 1–2 years to over 11 years, with 63.6% reporting more than a decade in the classroom. Each interview lasted approximately one hour and was conducted remotely over Zoom. Analysis followed Reflexive Thematic Analysis, with two coders open-coding transcripts and refining a codebook over four rounds.

Why teachers picked it up

Adoption was social before it was technical. Teachers described being pulled in by a colleague's demonstration, a district technology contact, or a teacher they followed on TikTok, then becoming local advocates themselves — one ran a professional development session for his school after a ten-second elevator pitch convinced a coworker. Peer validation also countered the suspicion that genAI use was lazy. A middle school social studies teacher with 11+ years of experience distinguished the two: using genAI meant "skipping the steps that are just bureaucratic nonsense," not "plugging in and using no thought."

The relief teachers described was concrete: brainstorming, differentiation, translating documents, drafting family communication, generating sentence stems, building behavior intervention plans, and producing formative assessments. A middle school CLDE teacher framed the alternative starkly — a job that consumes 80 hours a week leaves little capacity to "be the best teacher you can be when you're in the classroom with kids." Others described genAI as a creative opening, using it to let students interview a historical figure through a chatbot or to build avatar-led micro-professional-development videos.

Why teachers held back

Time was the most common barrier, and it operated twice: teachers lacked planning time to learn a tool, and prompting itself could consume more effort than the task warranted. One elementary teacher abandoned prompts that required repeated editing, and a social studies teacher found it failed on graph-based assessment items it could not read.

Stigma was the second barrier, and it came from multiple directions. Students accused teachers of cheating when they noticed ChatGPT-written materials; parents, one teacher said, "would be very upset" if they knew; colleagues treated genAI use as a sign of lowered standards. The stigma reached professional identity — a high school CLDE teacher asked what her role amounted to "if AI's doing all this work." A third group simply preferred what they knew, describing established plans as reliable and experimentation as unnecessary. Environmental concern formed a smaller fourth strand.

The boundaries teachers drew

The most consequential finding is that boundaries were pedagogical, not technical. Teachers decided what to use genAI for by asking what they were measuring: one allowed AI during growth work but withdrew it at mastery, so students had to "show me that you can do it without that support." A special education teacher taught fourth graders to interrogate authorship — rephrasing their own ideas was acceptable, generating them was not — while conceding that many would use the tools regardless.

A second boundary concerned voice: teachers rewrote outputs so materials still sounded like them. A third concerned trust, with facts cross-checked and genAI treated as "a majority source" rather than a trusted one. The fourth concerned privacy: teachers redacted student names, avoided uploading scanned documents, and used district accounts so their usage could be audited. That is collaborative sensemaking in the strict sense, with trust calibrated against disciplinary standards.

What sustainable integration would require

Teachers did not ask for more tools. They asked for rules, examples, and time. They wanted to know what their district permitted, which tools had been vetted, and whether AI counted as plagiarism — questions that a "go explore it" mandate left unanswered. They wanted professional development from former teachers and a one-page guide to when genAI helps. They wanted students taught to challenge what genAI produces, with ethics and authorship treated as part of the curriculum rather than an afterthought. And they insisted that integration preserve the relational work of teaching: one social studies teacher estimated that only about 2% of students can teach themselves from printed material, while most need a social environment.

Read together, these are governance demands rather than training demands. Opportunities also varied by school within the same district, which converts an individual gap into an equity problem: where guidance is absent, the teachers with the most time and collegial support capture the benefit.

What this means for practice

• Instructors. Write down your own boundary rule before adopting a tool, stated in terms of what you are assessing: genAI during practice, not at mastery. It is the boundary teachers in this study converged on independently, and it is defensible to students and parents in a way that a blanket permission or ban is not.
• Administrators. Publish an approved-tool list and a plain-language statement of what is permitted, then fund the time to learn it. The study's clearest institutional finding is that endorsement without guidance pushes decisions down to individual teachers, and that the resulting variation tracks school resources.
• Designers and professional development leads. Build for review rather than replacement: surface uncertainty, flag claims worth verifying, and keep the teacher's edit in the loop. Teachers in this study did not want outputs to accept; they wanted drafts to interrogate.
• District leaders. Pair any rollout with student-facing AI literacy and ethics instruction. Teachers reported student dependence and authorship confusion that no teacher-facing tool addresses.

Limitations

• All 22 participants came from a single U.S. public school district, so the perspectives may not transfer to districts with different resource levels, governance structures, or genAI approaches.
• Interviews were conducted remotely over Zoom, which limited observation of contextual cues and the informal rapport that in-person fieldwork affords.
• The sample was purposively selected through a district administrator's recruitment email, which likely over-represents teachers already interested in or using genAI.
• The study is cross-sectional and does not compare teaching before and after adoption; it describes situated decisions under evolving technical, institutional, social, and ethical constraints.
• All data are self-reported accounts of practice; the researchers did not observe classrooms, audit actual tool use, or measure student outcomes.
• Interviews took place from March through May 2025, during a policy vacuum that has since closed, so the negotiation described may date quickly as district rules and tools change.

Citation

Dangol, A., Kotiyal, S., Wolfe, R., Bowers, A. J., Vigil, A., Yip, J., Kientz, J. A., Shahid, S., Yeh, T., Cho, V., & Davis, K. (2026). Relief or displacement? How teachers are negotiating generative AI's role in their professional practice. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26). ACM.