Generative AI in Pre-Class Preparation for English Reading Classes: Quantitative Measures and Patterns of Use

研究論文	MEDIA, ENGLISH AND COMMUNICATION, No. 16 　　 2026 The Japan Association for Media English Studies

Generative AI in Pre-Class Preparation for English Reading Classes: Quantitative Measures and Patterns of Use

Tae Kudo (Kwansei Gakuin University), Yoshihiro Minamitsu (Osaka Electro-Communication University)

Abstract: Recent surveys suggest that generative AI (GenAI) has become part of many students' everyday learning environments. While its role in writing has received growing attention, its influence on reading remains underexamined. This study examined pre-class preparation for university English reading classes under non-AI- and AI-permitted conditions and explored students’ GenAI use. Participants were 83 non-English-major undergraduates at a private Japanese university. The study employed a classroom-based within-participant design. Quantitative comparisons revealed no statistically significant differences in quiz and final exam performance. Preparation time and perceived understanding also did not differ significantly. Questionnaire responses, usage logs, and reflection data showed that students used GenAI primarily for local comprehension support, including vocabulary checking, sentence-level meaning confirmation, and structural explanations. Uses involving metacognitive reflection were less common. These findings suggest that, in the short term, GenAI functioned mainly as a supplementary comprehension aid; its benefits may depend on learners' strategic and critical engagement.

Keywords: generative AI, English reading, AI literacy, pre-class preparation

1. Introduction

Recent advances in machine translation (MT) and generative AI (GenAI) have rapidly transformed the technological environment for university language learning. MT has already provided learners with immediate support for understanding foreign language texts. More recently, GenAI expanded this support beyond translation to include explanation, summarization, idea generation, and instant feedback. Recent large-scale surveys suggest that such tools are now widely used in higher education for brainstorming, summarizing texts, explaining concepts, and supporting coursework (Freeman, 2025; Ravšelj et al., 2025). These findings suggest that GenAI has already become part of many students' everyday learning environments.

A similar trend has been observed in Japan. Previous studies have demonstrated the adoption of MT by university students (Iyanaga, 2022; Oda, 2019; Sakai et al., 2024; Sato, 2023). Since its emergence, ChatGPT has also been widely adopted in educational settings (Mok, 2024; Ohmori et al., 2023). Sakai et al. (2024) suggested that MT was already familiar to many students and that familiarity with ChatGPT was increasing, but only a small minority had received instruction on how to use the latter. This point is important because the central issue in higher education may no longer be simply the use of AI-based language tools by students, but how this use should be understood and guided pedagogically.

However, research on the role of GenAI in language learning has developed unevenly across different skills. Recent reviews indicate that writing has received considerably greater attention than reading (Li et al., 2025; Lo et al., 2024). Representative empirical studies on writing have examined AI-supported planning, feedback, revision, writing development, and assessment (Hennessy, 2025; Li et al., 2024; Poláková & Ivenz, 2024; Song & Song, 2023; Su et al., 2023). While these studies indicate GenAI’s considerable potential for writing instruction, they also highlight the imbalance in existing literature.

The body of research on reading, compared to writing, remains smaller and less established. International studies suggest that MT and GenAI may support aspects of second language (L2) reading, but the findings are limited and mixed. For example, Park and Chon (2024) reported that MT could support some types of L2 reading comprehension, although the effects varied depending on learners’ proficiency and task type. Recent studies on GenAI have reported potentially positive effects in areas such as main-idea reading comprehension, academic reading achievement, reading anxiety, cognitive load, self-regulated strategy use, and engagement (Kim, 2024; Pan et al., 2025; Zhang et al., 2025). Taken together, these studies suggest that AI-based tools may support reading, but the empirical base remains much smaller than that for writing.

In Japan, reading-focused research on MT and GenAI has begun to emerge, but remains limited in scope. These studies have examined several aspects of L2 reading, including how learners use MT and how those uses relate to reading outcomes, as well as exploratory classroom practice, reading-related skills, and learners’ use of MT and GenAI (Kudo, 2023; Kudo et al., 2024; Minamitsu & Kudo, 2025; Prichard & Atkins, 2024). Related classroom-based studies also suggest how GenAI can support reading-oriented learning. For example, Miyahara (2024) reported the use of ChatGPT in a university reading course to support the development of vocabulary and grammar skills needed for reading, whereas Tabata (2023) showed that students used GenAI for reading-related purposes, such as simplified explanations, translation support, and English summaries of difficult texts.

Overall, the literature suggests that MT and GenAI are already included in many university students’ learning practices, both internationally and in Japan. However, in the context of reading, these tools raise a particular pedagogical concern: while they may help learners access difficult texts, they may also enable them to complete reading-related tasks without fully engaging with the text itself. The key issue, therefore, is how GenAI use may shape learners’ reading rather than simply support preparation. Although writing-related research has developed rapidly, empirical research on reading remains limited, and clear instructional frameworks for classroom use have not been fully established. The present study addresses this issue by examining the effects of permitting GenAI use during pre-class preparation for university English reading classes and describing how learners actually use GenAI tools in relation to assigned readings.

2. Research Objectives

This study was conducted with two primary aims. First, it examined whether allowing the use of GenAI during pre-class preparation for English reading classes influenced students’ performance and preparation-related measures. Specifically, it compared pre-class preparation under non-AI- and AI-permitted conditions to examine whether access to GenAI affected quiz and final examination performance. It also examined whether preparation time and perceived understanding differed between the two conditions.

Second, it explored how students used GenAI tools while preparing for English reading. Specifically, it aimed to identify the main purposes and patterns of use, such as vocabulary support, sentence-level meaning confirmation, structural explanation, summarization, and self-verification.

Accordingly, the study addressed the following research questions:

RQ1: Do quiz performance, final examination performance, preparation time, and perceived understanding differ between the non-AI- and AI-permitted preparation conditions?

RQ2: How do students use GenAI tools when preparing for English reading, and what purposes and patterns characterize their use?

3. Research Method

3.1 Participants

The study involved 83 undergraduate students majoring in fields other than English at a private Japanese university. All students had been placed into the same class based on a placement test, although their English proficiency still varied within the class. They were enrolled in a mandatory English reading course in which classroom activities centered on English text comprehension. In regular lessons, students primarily checked their understanding of the assigned texts, reviewed vocabulary and sentence structure, and received instructor explanations of difficult passages. They then answered the reading comprehension questions provided in the textbook. Extended discussions of topic background information and text summarization were not central components of the course. Prior experience with GenAI was not systematically assessed before the study and was, therefore, not controlled for. Data were collected with a focus on the preparatory activities associated with this course.

3.2 Research Design and Procedure

This study was conducted as a classroom-based exploratory study, using a within-participant sequential design. The same students experienced both conditions in the same order: first, the non-AI condition and, two weeks later, the AI-permitted condition. In this study, “pre-class preparation” refers to out-of-class work completed before the relevant class meeting, rather than in-class pre-reading activities.

In the non-AI condition, students read the assigned text independently before class. They could use dictionaries and reference materials provided by the instructor, but not GenAI or MT. In the AI-permitted condition, students were allowed to use GenAI tools as supplementary aids during preparation. In practice, students used a range of GenAI tools rather than a single platform; however, this study applied certain restrictions. MT was not permitted, and full-text translation of the reading passage was prohibited. Students were encouraged to use GenAI as a tool to check vocabulary, understand sentence structure, and support content comprehension, rather than as a substitute for their own reading and interpretation.

Before the AI-permitted assignment, the students received brief guidance on appropriate GenAI use. They were provided with examples of inappropriate use, such as using GenAI before first attempting to read and interpret the text, or accepting GenAI output uncritically. They were also shown how to create and submit shared links or screenshots to record their use of GenAI. In both conditions, they were first asked to read the text without external support before using permitted resources. These resources included dictionaries, reference materials provided by the instructor in both conditions, and GenAI tools in the AI-permitted condition.

The two texts used in this study were selected to be as similar as possible in terms of topic and difficulty. The texts were also similar in length, with 788 and 736 words in Texts A and B, respectively. As a rough indicator of readability, the Flesch Reading Ease scores were 52.45 for Text A in the non-AI condition and 54.63 for Text B in the AI-permitted condition.

Although the preparatory and classroom activities were conducted as part of the regular course, participation in the research, defined as permission to use students’ data for analysis, was voluntary. The students were informed that their decision to participate would not affect their course grades, and they could withdraw from the study at any time. All 83 students consented to the use of their data, and no student opted out of the study at any point. Cases with missing or incomplete data were treated as missing values in the relevant analyses. Informed consent was obtained through an online form administered through Google Forms.

3.3 Data Collection

Data were collected to address RQ1 and RQ2. The number of valid cases differed across the analyses because some students did not submit all required materials or submitted incomplete responses.

3.3.1 Reading-Related Performance

Reading-related performance was assessed using review quiz scores and relevant portions of the final examination. A review quiz was administered at the beginning of the following week’s class, without instructor explanation or classroom review of the assigned text. Each quiz comprised five multiple-choice items that assessed comprehension of the weekly reading material and served as an indicator of short-term understanding. At the end of the semester, the students took the final examination. For this study, only the portions of the examination corresponding to the non-AI- and AI-permitted reading materials were analyzed. These portions included items assessing comprehension of the target texts and related vocabulary, with some items requiring short-answer responses. Therefore, the scores were treated as broader indicators of reading-related performance on the target materials, rather than as pure measures of reading comprehension.

3.3.2 Preparation Time

As a measure of preparation time, the participants also reported the amount of time they had spent preparing, through post-preparation reflections after each assignment.

3.3.3 Perceived Understanding

Participants rated their perceived understanding of the text content on a 10-point scale before and after preparation. The present study used the post-preparation perceived understanding ratings for comparison across the conditions.

3.3.4 Purposes and Frequency of GenAI Use

In the AI-permitted condition, an eight-item questionnaire was administered to investigate why GenAI was used. Responses to each item were collected on a four-point Likert scale, indicating the frequency of use.

3.3.5 GenAI Usage Logs and Post-Task Reflections

To examine its actual usage, records of student interactions with GenAI during preparation were collected in the form of shared links or screenshots. Additionally, after completing the AI-supported preparation task, the participants were asked to reflect on how they used and adapted GenAI.

3.4 Data Analysis

To address the two research questions, this study employed quantitative analyses for RQ1 and descriptive and qualitative analyses for RQ2.

Regarding RQ1, paired-sample t-tests were conducted to compare the non-AI- and AI-permitted conditions in terms of review quiz and final examination scores, preparation time, and post-preparation perceived understanding ratings. To check the comparability of the materials and measures used in this study, the readability of the two texts was examined using the Flesch Reading Ease score. Furthermore, independent-sample t-tests were conducted on scores from previously administered quizzes and final examinations in the same format. These checks showed no statistically significant differences, suggesting that the materials and measures were broadly comparable for the purposes of the study.

Regarding RQ2, the questionnaire responses were first summarized descriptively to identify the overall trends in the purpose and frequency of GenAI use. The two authors then conducted an initial review of the GenAI usage logs and open-ended reflections and discussed broad recurring patterns. In this first stage, students’ GenAI use was categorized into cases where they appeared to hand the task over to GenAI and those where they used it as support. The authors developed subcategories through discussion. Each author independently classified the data; then, the classifications were compared and discussed until the final categories were agreed upon. The final description of usage patterns was based on questionnaire results, usage logs, and reflections.

4. Results

4.1 Results for RQ1

Table 1 Descriptive Statistics and t-Test Results for Quantitative Measures

Measure	n	Non-AI condition M (SD)	AI-permitted condition M (SD)	t	df	p	d
Quiz score	64	3.39 (1.34)	3.17 (0.98)	1.36	63	0.18	0.17
Final examination score	75	6.39 (1.85)	6.67 (1.64)	-1.28	74	0.21	0.15
Preparation time (minutes)	68	82.25 (32.88)	83.21 (43.18)	-0.23	67	0.82	0.03
Post-preparation perceived understanding	68	6.93 (1.64)	6.93 (1.56)	0.00	67	1.00	0.00

Note. M = mean; SD = standard deviation; n = number of complete pairs; df = degrees of freedom; d = Cohen’s d for paired samples, reported as an absolute value.

As indicated in Table 1, no statistically significant differences were found between the non-AI- and AI-permitted conditions for any quantitative measure. The quiz scores did not differ significantly between the two conditions, t(63) = 1.36, p = 0.18, with mean scores of 3.39 (SD = 1.34) and 3.17 (SD = 0.98), respectively, in the non-AI- and AI-permitted conditions. Final examination scores also did not differ significantly, t(74) = -1.28, p = 0.21; the mean scores were 6.39 (SD = 1.85) and 6.67 (SD = 1.64), respectively. Likewise, no significant difference was observed in preparation time, t(67) = -0.23, p = 0.82, with students spending an average of 82.25 (SD = 32.88) and 83.21 (SD = 43.18) minutes, respectively, in the non-AI- and AI-permitted conditions. Post-preparation perceived understanding did not differ between the two conditions, t(67) = 0.00, p = 1.00. The mean rating was 6.93 in both conditions (SD = 1.64 and 1.56 in the non-AI- and AI-permitted conditions, respectively). All of the data mentioned above were analyzed using langtest (Mizumoto, 2015).

4.2 Results for RQ2

To address RQ2, the questionnaire responses, GenAI usage logs, and post-task reflections collected in the AI-permitted condition were analyzed to describe how learners actually used GenAI during preparation. Although all participants experienced both conditions, the number of cases differed across analyses due to non-completion of the task and missing or incomplete GenAI-use records. Of the 83 students enrolled in the course, 77 completed the relevant preparation task. Among the 77 students in the AI-permitted condition, three did not use GenAI at all and seven used it minimally. Additionally, only 59 of the 74 students who used GenAI submitted records on GenAI use. Among the 59 submitted records, 44 used the free version of ChatGPT, eight, the paid version of ChatGPT, six, Gemini, and one, Copilot.

Table 2 summarizes the questionnaire results for GenAI use. These results show that GenAI was mostly used to check the meanings of words or phrases. Other common uses included checking the meaning of one or two sentences and asking for explanations of English sentence structure. Less frequent uses included seeking background or contextual information on the topics mentioned in the text and using GenAI to check their own answers or summaries.

Table 2

Questionnaire Results on the Purposes of GenAI Use

Purpose of use	4	3	2	1
To check the meanings of words or phrases	50	8	11	8
To generate illustrative sentences using words or phrases	12	8	19	38
To check the meaning of one or two sentences	15	24	18	19
To ask for explanations of English sentence structure	20	15	13	29
To ask for explanations of specific grammar points	9	12	22	33
To ask for paragraph-by-paragraph summaries	14	14	17	32
To check self-generated answers or summaries	14	13	11	39
To seek background or contextual information about topics	2	5	11	57

Note. 4 = frequently used, 3 = sometimes used, 2 = used a little, 1 = did not use it

To supplement the questionnaire findings, GenAI usage logs were classified by type of use. As presented in Table 3, the most frequent uses involved micro-level language support, particularly for checking the meaning of words or phrases. Additionally, many cases involved asking GenAI for explanations of the sentence structure within the preparation scope. The least frequent use involved students entering their own answers and asking GenAI to evaluate them.

Table 3

Classification of GenAI Use during Reading Preparation

Broad type	Specific use	Frequency
Micro-level language support	Checking the meaning of words or phrases	36
	Generating example sentences for words or phrases	8
	Checking the meaning of one or two sentences	13
Prompting GenAI for responses	Asking for explanations of the sentence structure (within the preparation scope)	23
	Asking for explanations of the sentence structure (beyond the preparation scope)	6
	Asking for explanations of specific grammar points	4
	Asking for paragraph summaries (within the preparation scope)	10
Submitting one’s own answers for GenAI evaluation	Checking one’s interpretation of one or two sentences	5
	Checking one’s analysis of the sentence structure (within the preparation scope)	6
	Checking one’s analysis of the sentence structure (beyond the preparation scope)	1
	Checking specific grammar points	0
	Checking one’s paragraph summary (within the preparation scope)	6

Note. “Preparation scope” refers to uses directly related to the assigned tasks for class preparation. Frequency indicates the number of students who used each type at least once. Repeated use by the same student was counted only once for each category. As individual students may demonstrate more than one type of use, frequencies may overlap across categories.

The analysis of the open-ended reflections identified five broad categories: “vocabulary or language knowledge supplementation,” “support for sentence structure/grammar understanding,” “top-down use,” “self-verification use,” and “self-regulatory use.” As some responses were assigned to multiple categories, the frequencies reported below are cumulative. There were 22 instances classified as vocabulary or language knowledge supplementation, 23 as support for sentence structure/grammar understanding, and eight as top-down use. Additionally, self-verification, wherein students checked their own ideas or answers against the AI output, was observed in 19 cases. This category included four subtypes: checking the appropriateness of students’ own interpretations or answers, checking sentence structure and logic, learning adjustment, and post-verification.

The reflections also revealed self-regulatory tendencies in some learners, including deliberately limiting the scope of GenAI use, prioritizing independent efforts, and feeling uncertain about the extent to which they should rely on GenAI. Some reflections further suggested that the GenAI responses were occasionally more extensive than desired by the students. Several students reported that GenAI provided explanations or Japanese translations beyond the specific points they had requested, which sometimes made it more difficult for them to identify exactly what they had or had not understood.

Finally, reflections from students who used GenAI sparingly or not at all suggested several reasons for its limited use. These included finding the assigned text relatively easy, being unsure of how to use GenAI effectively, being concerned about over-reliance on full-text translation, and believing that reading and interpreting the text on their own was more beneficial for learning.

5. Discussion

The present study examined the effects of permitting GenAI use during pre-class preparation for university English reading classes and explored how students actually used such tools during that process. Two main conclusions emerge from the findings.

First, no statistically significant differences were observed across the measured indicators under the conditions of this study. In the context of pre-class preparation, access to GenAI as supplementary support was not associated with statistically significant differences in the quiz or final examination performance. Preparation time and perceived understanding also did not differ significantly.

Simultaneously, the absence of any significant differences suggests that simply permitting access to GenAI does not automatically produce measurable learning benefits. GenAI may have the potential to support comprehension through vocabulary explanation, sentence analysis, and summarization of the text. However, this potential may not be fully realized unless learners use the tool strategically and in pedagogically appropriate ways. Thus, the present findings are consistent with earlier arguments that the educational value of AI-based tools depends less on their availability and more on how learners use them.

Second, the RQ2 findings showed that students mainly used GenAI to support bottom-up processing at the word and sentence levels. Overall, the participants appeared to use GenAI to address immediate points of difficulty, rather than to support top-down processing or metacognitive reflection on their reading. This may help explain why the quantitative results remained stable across conditions: GenAI may have functioned as a supplementary aid for resolving local comprehension problems without substantially changing learners’ overall preparation processes or broader performance.

The relatively infrequent use of GenAI to seek background or contextual information about the topics mentioned in the text and to summarize the text, should also be interpreted in relation to the course design. As the lessons mainly emphasized comprehension of the assigned text, vocabulary, and sentence structure, students may have had limited reason to use GenAI.

The qualitative data also revealed clear variations among participants. Some students used GenAI in reflective and self-regulatory ways; for example, by checking their own interpretations, comparing their ideas with GenAI outputs, or deliberately limiting their dependence on the tool. Others were unsure about how to use GenAI effectively or chose not to use it because they believed that reading and interpreting the text themselves was more beneficial for learning. These differences suggest that learners did not use GenAI in uniform ways. Rather, their use appeared to be shaped by their beliefs about learning, familiarity with the tool, and their ability to judge when and how GenAI support should be used.

Reflective comments also suggested that GenAI is not always helpful when it provides more information than learners need. In some cases, responses that went beyond the participant’s immediate question appeared to blur the boundary between what was and was not already understood. This suggests that the effective use of GenAI in reading may require not only access to support but also the ability to decide when to use such support and how to restrict GenAI responses to the information needed.

A small number of participants demonstrated distinct patterns of GenAI use. One participant used GenAI mainly for ongoing self-verification by presenting their own answers and requesting evaluation rather than direct explanation. Another participant began the interaction by assigning GenAI the role of an “AI English reading coach,” providing a learner profile and clear instructions on how support should be provided. These participants used GenAI as a guided learning partner throughout their tasks. Although these cases were exceptional, some students appeared to be able to use GenAI in more deliberate and pedagogically sophisticated ways than others, suggesting that AI literacy may have differed considerably across learners.

These findings have significant pedagogical implications. If students are permitted to use GenAI without clear guidance, its role may remain limited to ad hoc and surface-level support. Conversely, if teachers provide more explicit instruction on how to use GenAI for reading, students may be better able to use it in productive and critical ways. Such instruction might include asking for concise vocabulary explanations, requesting structural support without relying on full-text translation, comparing one’s own interpretation with the GenAI output, and evaluating the reliability and usefulness of responses generated by GenAI. It may also be important to help learners manage the scope of AI support so that responses remain focused on specific points of difficulty.

Taken together, these findings suggest that GenAI may be best understood not as a substitute for learners’ own reading and interpretation, but as a supplementary scaffold. In this study, the analyses did not reveal any statistically significant declines associated with GenAI use, although the present research design may not have been sufficiently sensitive to identify small adverse effects. Simultaneously, the absence of statistically significant gains suggests that benefits should not be expected automatically. Its educational value may depend on how it is integrated into classroom practice.

These interpretations should be considered in light of several limitations. First, the study was conducted within a single institutional context with a specific group of Japanese university students, which may have limited the generalizability of the findings. Additionally, students’ preparation behavior may have been shaped by the design of the reading course. As the lessons did not place strong emphasis on the discussion of topic background or summarization of the text, the infrequent use of GenAI to seek background or contextual information or to summarize the text should not be interpreted as a general tendency across reading classes. Second, as the study was conducted over a relatively short period and involved only a limited number of texts, the long-term effects of GenAI use on reading comprehension development were not examined. Third, each condition was paired with a different text, and all the students completed the conditions in the same fixed order: the non-AI condition first and the AI-permitted condition two weeks later. Consequently, the effects of condition could not be separated from possible text and order effects. Fourth, the study focused primarily on measurable indicators, such as quiz performance, final examination performance, preparation time, and perceived understanding, but did not directly examine cognitive processing while preparing for reading classes. The quantitative results should also be interpreted cautiously because they may have been influenced by factors other than GenAI use during pre-class preparation, especially in the case of final examination performance, which could reflect in-class engagement, instructor explanations, and subsequent review. Fifth, the study described patterns of GenAI use but did not statistically evaluate the relationships between specific usage patterns and the measured indicators. Finally, although usage logs and reflections provided valuable information about learner behavior, not all participants submitted complete records of GenAI use.

Therefore, future research should examine sustained and systematically guided use of GenAI in reading instruction. This need is also consistent with recent reviews suggesting that, although GenAI may be helpful for language learning, empirical research on reading remains comparatively limited and more systematic investigation of pedagogical integration is needed (Li et al., 2025; Lo et al., 2024). Specifically, future studies should investigate whether explicit instruction in AI literacy and prompt design helps learners use GenAI strategically and reflectively.

Further research could also compare different types of GenAI support, such as vocabulary-focused assistance, structural explanation, comprehension monitoring, and reflective feedback, to determine which forms of support are most beneficial for reading comprehension development. Additionally, closer examination of individual and tool differences may help explain why some learners use GenAI effectively, whereas others use it only minimally or not at all. Such limited use may partly reflect uncertainty about how to use GenAI.

6. Conclusion

This study compared the non-AI- and AI-permitted conditions in preparation for university English reading classes and explored how learners actually used GenAI tools during the preparation process. The quantitative results showed that differences between the two conditions in quiz performance, final examination performance, preparation time, and perceived understanding did not reach statistical significance. These findings suggest that under the short-term conditions of this study, permitting access to GenAI as a supplementary tool was not associated with clear differences in any of these measures.

Simultaneously, the RQ2 findings revealed that learners used GenAI in diverse ways, and most uses focused on vocabulary support, sentence-level meaning confirmation, and sentence-structure explanation. Uses involving reflection and self-verification were observed in some cases, but these were uncommon. Therefore, the results indicate that learners’ GenAI use was concentrated on bottom-up processing at the word and sentence levels, rather than on top-down processing or metacognitive reflection during reading preparation.

Overall, these findings do not provide evidence that access to GenAI automatically improves performance in reading preparation. No statistically significant differences were detected in performance or perceived understanding. However, the findings do not establish equivalence between the conditions or rule out possible adverse effects. The benefits of GenAI may depend on whether learners receive appropriate guidance on its effective and critical use.

From a pedagogical perspective, this study highlights the importance of designing classroom practices that position GenAI as a scaffold for comprehension, rather than as a tool for bypassing the reading process. Supporting learners in the strategic, critical, and reflective use of GenAI may be essential for it to meaningfully contribute to reading instruction in higher education.

Notes

Part of this study was presented at Language Education Expo 2026, held at Chuo University on March 1, 2026, under the title “Daigakusei wa eigo rīdingu yoshū de seisei AI o dō katsuyō shite iru no ka: puronputo to furikaeri kijutsu kara miru seisei AI katsuyō no jittai” [How Do University Students Use Generative AI in Preparation for English Reading? ―Actual Patterns of Use Revealed Through Prompts and Reflection]. This work was supported by JSPS KAKENHI Grant Number 24K04148.

The authors would like to thank the two anonymous reviewers for their careful reading and constructive comments, which helped improve the clarity and quality of this article. Any remaining errors are the authors’ own.

References

Freeman, J. (2025). Student generative AI survey 2025 (HEPI Policy Note 61). Higher Education Policy Institute.

Hennessy, E. (2025). Enhancing L2 writing with ChatGPT for Japanese EFL students. In B. Lacy, M. Swanson, & P. Lege (Eds.), Moving JALT into the future: Opportunity, diversity, and excellence (pp. 32–42). JALT. https://doi.org/10.37546/JALTPCP2024-04

Iyanaga, K. (2022). Nihonjin daigakusei no kikai honyaku shiyō no jittai chōsa to kongo no eigo kyōiku e no dōnyū ni kansuru kōsatsu [A survey of Japanese university students’ use of machine translation and considerations for its future introduction into English education]. Memoirs of Kyoto Tachibana University, 48, 1–19.

Kim, R. (2024). Effects of ChatGPT on Korean EFL learners’ main-idea reading comprehension via top-down processing. Language Research, 60(1), 83–106. https://doi.org/10.30961/lr.2024.60.1.83

Kudo, T. (2023). Machine translation and reading in a foreign language: A preliminary trial lesson with university students. Language and Culture, 26, 57–71.

Kudo, T., Minamitsu, Y., & Sumi, S. (2024). Exploring the impact of machine translation on students’ reading skills and vocabulary knowledge: Can it enhance English reading proficiency? Kwansei Gakuin University Researches in Higher Education, 14, 31–43.

Li, B., Tan, Y. L., Wang, C., & Lowell, V. (2025). Two years of innovation: A systematic review of empirical generative AI research in language learning and teaching. Computers and Education: Artificial Intelligence, 9, Article 100445. https://doi.org/10.1016/j.caeai.2025.100445

Li, J., Huang, J., Wu, W., & Whipple, P. B. (2024). Evaluating the role of ChatGPT in enhancing EFL writing assessments in classroom settings: A preliminary investigation. Humanities and Social Sciences Communications, 11, Article 1268. https://doi.org/10.1057/s41599-024-03755-2

Lo, C. K., Yu, P. L. H., Xu, S., Ng, D. T. K., & Jong, M. S. Y. (2024). Exploring the application of ChatGPT in ESL/EFL education and related research issues: A systematic review of empirical studies. Smart Learning Environments, 11, Article 50. https://doi.org/10.1186/s40561-024-00342-5

Minamitsu, Y., & Kudo, T. (2025). English reading comprehension support through machine translation and generative AI: A study of tool usage patterns. Research in the Humanities, 28, 1–11.

Miyahara, A. (2024). Exploring ChatGPT usage in English language education: Cultivating critical thinking with AI. Teacher Education Center Annals, Gifu Shotoku Gakuen University, 23, 53–60.

Mizumoto, A. (2015). Langtest (Version 1.0) [Web application]. https://langtest.jp/

Mok, J. (2024). Students’ use of generative artificial intelligence (ChatGPT) at Rikkyo University. Journal of Foreign Language Education and Research, 5, 21–39. https://doi.org/10.14992/0002001290

Oda, T. (2019). Machine translation and foreign language education in Japan. The Journal of Humanities and Natural Sciences, 145, 3–27.

Ohmori, F., Saito, J., Matsuba, R., & Kita, T. (2023, June 8). Zenkoku no daigakusei no ChatGPT riyō jittai ga hajimete akiraka ni― Daigakusei no ChatGPT riyō jōkyō to nōryoku keisei e no eikyō ni kansuru chōsa kekka (sokuhō)― [The first survey of ChatGPT usage by university students nationwide: Survey results on the use of ChatGPT by university students and its impact on skill development (preliminary report)]. DBER Center. https://dber.jp/wp-content/uploads/2023/06/chatgpt_report.pdf

Pan, M., Lai, C., & Guo, K. (2025). Effects of GenAI-empowered interactive support on university EFL students’ self-regulated strategy use and engagement in reading. The Internet and Higher Education, 65, Article 100991. https://doi.org/10.1016/j.iheduc.2024.100991

Park, G. J., & Chon, Y. V. (2024). From no aid to neural aids: Exploring the influence of machine translation on L2 reading problems. English Teaching, 79(4), 33–54. https://doi.org/10.15858/engtea.79.4.202412.33

Poláková, P., & Ivenz, P. (2024). The impact of ChatGPT feedback on the development of EFL students’ writing skills. Cogent Education, 11(1), Article 2410101. https://doi.org/10.1080/2331186X.2024.2410101

Prichard, C., & Atkins, A. (2024). Examining readers’ use of machine translation through eye tracking. System, 125, Article 103415. https://doi.org/10.1016/j.system.2024.103415

Ravšelj, D., Keržič, D., Tomaževič, N., Umek, L., Brezovar, N., Iahad, N. A., Abdulla, A. A., Akopyan, A., Aldana Segura, M. W., AlHumaid, J., Allam, M. F., Alló, M., Andoh, R. P. K., Andronic, O., Arthur, Y. D., Aydın, F., Badran, A., Balbontín-Alvarado, R., Ben Saad, H., . . . Aristovnik, A. (2025). Higher education students’ perceptions of ChatGPT: A global study of early reactions. PLOS ONE, 20(2), Article e0315011. https://doi.org/10.1371/journal.pone.0315011

Sakai, S., Ohkatsu, H., Kato, S., & Tsuchiya, K. (2024). Sumaho neitibu no ChatGPT to kikai hon'yaku ni taisuru ishiki ni tsuite no bunseki [An analysis of smartphone natives’ attitudes toward ChatGPT and machine translation]. Chiba University of Commerce Review, 62(1), 1–14.

Sato, M. (2023). English language education coexisting with machine translation: An analysis of resistance and changes in the significance of learning. Journal of the Japan Association for Developmental Education, 18, 81–90. https://doi.org/10.18950/jade.2023.07.20.01

Song, C., & Song, Y. (2023). Enhancing academic writing skills and motivation: Assessing the efficacy of ChatGPT in AI-assisted language learning for EFL students. Frontiers in Psychology, 14, Article 1260843. https://doi.org/10.3389/fpsyg.2023.1260843

Su, Y., Lin, Y., & Lai, C. (2023). Collaborating with ChatGPT in argumentative writing classrooms. Assessing Writing, 57, Article 100752. https://doi.org/10.1016/j.asw.2023.100752

Tabata, T. (2023). Eigo jugyō ni okeru seiseikei AI no katsuyō: Jiji eigo jugyō de no jissen hōkoku [The use of generative AI in English classes: A practical report from a current-affairs English course]. SELT Journal, 3, 103–109.

Zhang, Y., Lai, X., Yi, S., & Lu, Y. (2025). Does ChatGPT-based reading platform impact foreign language paper reading? Evidence from a quasi-experimental study on Chinese undergraduate students. Education and Information Technologies, 30, 9737–9754. https://doi.org/10.1007/s10639-024-13190-0

書誌情報

掲載誌：MEDIA, ENGLISH AND COMMUNICATION, No. 16 (2026, pp. 43-58) 発行元：一般社団法人　日本メディア英語学会 表　題：英語リーディングの予習における生成AI：量的指標と利用パターン 著　者：工藤多恵（関西学院大学）、南津佳広（大阪電気通信大学） 要　約：近年の調査では、生成AIは多くの学生の学習環境の一部となっており、ライティングにおける生成AIの活用に関する研究は広がる一方で、リーディングにおける活用については未だ限定的である。本研究では、大学英語リーディング授業の予習において、生成AIの使用を認めた場合と認めなかった場合の効果を比較するとともに、学生が実際にどのように生成AIを活用していたのかを明らかにした。対象は、日本の私立大学で英語を専攻としない83名の学生で、同一の学生が生成AIを使用する条件と使用しない条件の両方で予習課題に取り組んだ。クイズの結果、期末試験の結果、予習時間、英文理解への自信といった量的データのいずれにおいても、両条件間に統計的に有意な差は認められなかった。質的データからは、学生が主に語彙確認、文レベルの意味確認、構文の説明といった限定的な支援を目的とした生成AIの使用がわかった一方で、より主体的な活用は少なかった。以上の結果から、生成AIはリーディングにおける補助的な足場となりうるものの、その有効性は、学習者がこれらのツールを戦略的かつ批判的に活用できるかどうかに左右される可能性があることが示された。 キーワード：生成AI、英語リーディング、AIリテラシー、予習 論文カテゴリー：研究論文 CCライセンス：CC BY-NC 4.0 受理日：2026年8月10日 © 2026 Tae Kudo and Yoshihiro Minamitsu