Artificial intelligence (AI) describes “a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments” (U.S. Department of State, 2020, para. 2). Advancements in machine learning and natural language processing drive the use of AI rapidly becoming a prevailing trend across various fields (Adjekum et al., 2024), which improves workflow and boosts productivity. AI is reshaping various sectors, fostering innovation, and sparking discussions about ethics and ownership of work creation.
As AI becomes more prevalent, there is growing concern about students’ misuse of AI in academic writing, particularly regarding plagiarism, which affects not only higher education (Evangelista, 2025) but also P-12 education (Hsiao, 2024). This study stems from P-12 teachers’ concerns to systematically explore their perceptions of AI and plagiarism, and to assist in identifying solutions. The findings provide a foundation for educating and training teacher candidates in an online teaching endorsement program in the southeastern United States.
As recognized, plagiarism involves the unauthorized appropriation and presentation of one’s ideas or expressions, regardless of the form in which another individual originally created them (as cited in Hexham, 1999, para. 2). Different types of plagiarism exist, many of which arise from the incorrect use of sources and citations. Since AI can produce human-like content, teachers face greater challenges in distinguishing between student-written and AI-generated content (Alexander et al., 2023).
Several AI detectors have been developed to identify AI-generated content, including Originality.ai, GPTZero, Scribbr's AI detector, QuillBot's AI detector, and ZeroGPT. These tools can be helpful; however, after careful examination, the detection results for different combinations of human-written and AI-generated content varied across detectors, raising concerns about their accuracy (Hsiao, 2024, 2025). Currently, no AI detection tool is 100% accurate. Detection results can vary depending on factors such as the length or complexity of the writing, the extent of editing, or the specific AI model used to create it. Thus, relying solely on a detector to identify AI-generated content is not recommended.
To better distinguish between student-written and AI-generated content, teachers must be aware of common indicators of AI-generated content that aid in plagiarism detection. These indicators may include generic content, impeccable writing, the absence of personal experience and context, inconsistent style and tone, unusual phrases or words, missing or inaccurate citations, and specific AI-generated structures or formats (Baule, 2023; Chugh, 2023; Hsiao, 2024, 2025). In addition, reconsidering assessment strategies that do not rely solely on academic writing and implementing alternative assessments, such as creative projects, interactive exercises, and peer reviews/feedback, can help evaluate student learning outcomes more comprehensively (Elgersma, 2024; Hsiao, 2024, 2025).
The study took place in an online teaching endorsement program in the southeastern United States. The program prepares teacher candidates to plan, design, and deliver online instruction for students in grades P-12. Candidates also explore various aspects of online teaching and learn how to address challenges that arise. Online cheating and plagiarism are among the issues addressed in the program. These issues have become particularly important since the advance of AI. The study was conducted to explore them in greater depth.
To explore candidates’ perceptions of AI, plagiarism, and alternative assessments, the initial study invited 101 candidates from six course sections during the summer and fall semesters of 2025 to participate by completing a survey. As a result, a total of 18 candidates responded, yielding a response rate of 17.82%. The 37-item survey contains items concerning (1) demographic Information, (2) experience of students’ use of AI in writing, (3) perception of AI and its impact on plagiarism, (4) awareness of common indicators of plagiarism, (5) adoption of alternative assessments, and (6) potential integration of AI in online instruction. Most of the items are 5-point Likert-scale questions, and four open-ended questions are included. Reliability across all sections ranged from .609 to .914, indicating acceptable to strong internal consistency.
The participants were 18 females aged 25 to 54 years. They primarily taught math (27.78%), science (22.22%), or multiple subjects (27.78%) at the middle (38.89%) and high school (27.78%) levels. Most (72.22%) reported having an advanced level of confidence in their technological pedagogical knowledge. They reported using AI occasionally (27.78%) or sometimes for specific tasks (50%). Although most of their schools (72.22%) have policies on AI use and plagiarism, many of these policies (27.78%) are not highly restrictive.
Participants rated the frequency with which students use AI for writing assignments based on their daily observations. Their ratings were generally low. Across five of six questions, the mean scores indicated that students’ use of AI in writing assignments was infrequent, leaning toward the “rarely” end of the scale. Regarding AI misuse in writing assignments, participants reported a moderate level of AI misuse among their students. The mean score of 3.3 on a 5-point frequency scale suggests that such AI misuse occurred occasionally in their P-12 classrooms.
With a sample size of 18, the Kruskal–Wallis H test was used to detect differences among demographic groups because its nonparametric nature makes it more appropriate when the normality and equal-variance assumptions cannot be met. The test result showed that a significant difference existed between subject areas in students’ use of AI tools to conduct research, χ2(4) = 13.057, p = .011, with a mean rank score of 14.90 for Math, 12.00 for Science, 6.50 for Social Studies, 6.50 for “Other Subjects,” and 4.50 for “Multiple Subjects.” Mean rank scores indicate that students’ use of AI tools for conducting research may be more prevalent in Math and Science courses than in the other subject areas.
Participants generally agreed that AI-generated text should be treated as a secondary source (M = 3.89, SD = .58) and that proper citation and acknowledgment are important (M = 3.94, SD = .73). Otherwise, using AI to generate text without attribution is considered plagiarism. Participants also agreed that their students’ use of AI in writing assignments increases the risk of plagiarism (M = 3.83, SD = .92).
Participants were neutral to mildly agreeable about not classifying thoroughly evaluated, reworded, and appropriately cited AI-generated text as plagiarism (M = 3.5, SD = .62). This rating was lower than the above-mentioned ratings, reflecting the ongoing challenge of determining the originality of AI-generated content. Regarding the difficulties in detecting AI-related plagiarism (M = 3.22, SD = 1.22) and the accuracy of AI detectors (M = 3.11, SD = .83), participants expressed neutral opinions. They may consider that current AI detectors cannot accurately identify all instances of plagiarism, or that some instances can be recognized through human judgment.
The Kruskal–Wallis H test revealed significant differences in participants’ perceptions of AI and its impact on plagiarism according to their confidence levels in technological pedagogical knowledge. One was about their perception of whether AI-generated text should be treated as a secondary source, χ2(2) = 6.515, p = .038, with mean rank scores of 10.96 for participants with an advanced level in technological pedagogical knowledge, 10.50 for an expert level, and 4.50 for an intermediate level. Another significant result was participants’ agreement that using AI-generated text without proper citation or acknowledgment should be considered plagiarism, χ2(2) = 7.422, p = .024, with mean rank scores of 11.23 for participants at an advanced level, 5.63 for an intermediate level, and 2.50 for an expert level. Mean ranks for these two items indicate that participants with greater confidence in technological pedagogical knowledge may better understand AI’s benefits and limitations and view it as a support tool, while keeping the original work primary.
Based on the ratings of awareness of common indicators of plagiarism, it appears that inconsistencies in writing style (M = 3.89, SD = .76) and unusual phrases or word choices in students’ writing assignments (M = 3.94, SD = .54) often raise participants’ suspicions. Participants also generally agreed that impeccable writing (M = 3.67, SD = .84), the absence of personal experience or context (M = 3.67, SD = .59), and noticeable changes in tone (M = 3.67, SD = .59) can help them identify potential plagiarism.
Participants showed slight agreement that missing or inaccurate citations (M = 3.50, SD = .79) and specific AI-generated structures or formats (M = 3.56, SD = .71) can serve as indicators of potential plagiarism. The lowest rating was associated with generic content (M = 3.17, SD = .62), where participants remained neutral. This neutrality may be related to the nature of the writing assignments or whether specific contextual details are required.
The Kruskal–Wallis H test revealed significant differences in participants’ awareness of common indicators of plagiarism, by frequency of personal AI use. One was about using the absence of personal experience and context as an indicator to detect instances of plagiarism, χ2(2) = 6.475, p = .039, with mean rank scores of 12.17 for sometimes use, 8.25 for often use, and 5.7 for rare use. Another significant result was participants’ agreement that using specific AI-generated structures or formats as an indicator to detect instances of plagiarism, χ2(2) = 6.805, p = .033, with mean rank scores of 14.25 for often use, 9.44 for sometimes use, and 5.80 for rare use. Mean ranks for these two items indicate that frequent personal AI users may be more sensitive to common indicators of plagiarism, given their familiarity with AI-generated content and the limitations of AI tools.
Participants agreed that the adoption of alternative assessments prompts students to create personalized or context-specific work (M = 4.11, SD = .47), thereby helping to prevent plagiarism. They also agreed that using such alternatives helps evaluate student learning outcomes in more diverse ways than relying solely on traditional writing assignments (M = 4.00, SD = .77), and connects learning to real-life situations, making it more challenging to replicate assignments (M = 4.17, SD = .51). It also helps promote problem-solving skills beyond just writing (M = 4.00, SD = .69).
Participants recommended alternative assessments they have implemented in online instruction to help prevent plagiarism. The most frequently mentioned assessments were creative projects, interactive exercises, peer reviews/feedback, and presentations. Other recommended assessments included debates, role-playing, portfolios, concept maps, simulations, storytelling, journals, and case studies. One participant even suggested using labs. Adopting alternative assessments that align with the nature of the subject matter may be more effective at preventing plagiarism than relying solely on traditional writing assignments.
Participants generally indicated a willingness to support the integration of AI tools into online instruction (M = 3.67, SD = .84). At least eight participants indicated the need to train students to use AI effectively. They also emphasized that teachers require training. The appropriate activities they considered for integrating AI tools into online instruction included brainstorming ideas, summarizing resources, editing, outlining, grammar checking and proofreading, providing feedback, finding sources, creating images, and supporting personalized practice.
When integrating AI into online instruction, participants reported that they would consider implementing strategies to prevent plagiarism, such as (1) requiring students to complete writing assignments based on their personal experiences, (2) providing students with training on appropriate AI use, (3) establishing an academic integrity policy and agreement, and (4) requiring students to use appropriate citations.
To conclude, participating online teacher candidates demonstrated a cautious perception toward the use of AI-generated text regarding plagiarism. They acknowledged that AI detectors are not 100% accurate. Applying human judgment, using common indicators, and adopting alternative assessments can help online teachers more effectively detect and prevent plagiarism. They also emphasized the importance of training in AI use and proper citation practices. Subject area, confidence in technological pedagogical knowledge, and frequency of personal AI use may influence online teacher candidates’ perceptions of AI use relating to plagiarism and warrant further investigation.
EdTech Archives