Teachers used generative AI as a self-directed professional-learning scaffold, but the evidence remains self-reported

Li, Wang and Bonk examine how K-12 teachers use generative AI for self-directed professional development outside a formal course. Their explanatory mixed-methods study recruited 298 U.S. teachers through Prolific in May and June 2024. Eligibility required prior use of ChatGPT for teaching or professional development, so the sample represents experienced adopters rather than teachers in general. Participants completed a 27-item Likert survey grounded in Garrison's self-directed learning model. Five teachers later completed interviews and shared ChatGPT or Gemini interaction records.

Teachers reported using ChatGPT most often to find pedagogical or lesson ideas, organize a lesson, generate quizzes, customize materials and explain subject content. Survey responses indicated strong perceived ownership of professional learning, alignment with personal standards and monitoring of progress. Interviewees described the tool as a way to get an initial structure quickly, explore alternatives and adapt a draft to their particular learners. They also described checking output against professional knowledge rather than accepting it unchanged.

The study is informative about process, not effectiveness. The quantitative analysis is mainly descriptive, and participants already had AI experience. Teachers who disliked or abandoned the technology are less likely to appear in such a sample. Only five people completed interviews after recruitment attrition, making the qualitative evidence detailed but narrow. The researchers did not compare users with non-users, observe classroom practice, score generated lessons, measure time saved or assess student learning.

These boundaries change the appropriate claim. The evidence supports saying that participating teachers perceived generative AI as a flexible professional-learning scaffold and illustrated how they reviewed and contextualized its output. It does not show that ChatGPT caused better teaching, reduced workload by a measured amount or improved student achievement. Confidence in using a tool is also different from confidence that its output is accurate, equitable and instructionally appropriate.

For professional learning design, the reported behaviours suggest useful routines. A teacher can ask for several approaches, identify the assumptions behind each, compare them with curriculum and learner evidence, revise a chosen plan and record why the revision is better. Peer review can surface subject errors, accessibility problems or lowered cognitive demand. School leaders can provide protected accounts and exemplars while preserving teacher choice and avoiding usage quotas that reward activity over quality.

For Hong Kong schools, a pilot should include Chinese and English materials, local curricula and varied subject areas. Evaluation can measure the full time required to prompt, check and revise; independently score the resulting materials; and observe whether changes survive classroom use. Teachers who do not adopt the tool should be included so barriers and negative experiences remain visible.

The paper's value lies in documenting an emerging form of informal teacher learning while keeping the evidence modest. Generative AI may make professional exploration more immediate, but professional judgment, peer critique and outcome measurement remain the mechanisms that turn a fast suggestion into defensible teaching practice.

Professional development should therefore evaluate the teacher's reasoning trail, not merely the generated artifact. A short reflection on rejected suggestions, revised assumptions and observed classroom results can make expert judgment visible and support better peer learning.