Introduction: Something Everyone Wondered About, but No One Had Verified
Using Generative AI for lesson preparation is becoming commonplace. Structural analysis of materials, brainstorming questions, drafting blackboard plans — ChatGPT, Claude, Gemini, NotebookLM. As more teachers master these tools, a certain question has begun to drift through the staffroom air.
"Is it okay to input textbook text into AI?"
If you search online, you'll find plenty of personal interpretations and hearsay. The Ministry of Education's "Guidelines for the Utilization of Generative AI in Elementary and Secondary Education (December 2024, Ver. 2.0)" provides a direction for use, but individual copyright judgments ultimately depend heavily on the rights holder's perspective. In other words, no matter how many interpretations by random people on the internet you read, you won't find the answer to this question.
Some say, "It's fine because of Article 35," while others say, "Inputting is reproduction, so you need permission." Both arguments have merit, but neither comes from the rights holders themselves. Textbooks belong to the publishing companies and the many authors who provide their works to them.
So, I thought, I have no choice but to ask the parties involved.
With that in mind, I sent an inquiry email with the same content to the three publishers of the textbooks I use in my classes — Tokyo Shoseki, Mitsumura Tosho, and Kyoiku Shuppan. As a result, I received surprisingly polite and sincere responses from all three companies.
And to jump to the conclusion: the answers were three companies, three ways. In this article, I will summarize as accurately as possible the questions I sent, the gist of each company's response, the common lines and differences that emerged, and how we should behave in our daily work from tomorrow.
First, Organizing the Premises: Three Rules Involved in This Issue
Before reading the responses from each company, let's briefly organize the rules that appear (please be sure to check the primary sources for accurate content).
① Copyright Law Article 35 (Limitations on Rights for the Purpose of Teaching)
This provision allows for the reproduction and public transmission of copyrighted works without permission to the extent deemed necessary in the "course of a lesson" at a school. However, cases that "unreasonably prejudice the interests of the copyright holder" are excluded. Public transmission is predicated on the payment of compensation to SARTRAS (Society for the Administration of Remuneration for Public Transmission for School Lessons), which is usually handled collectively by the establisher (municipality). Specific operations are organized in the "Operational Guidelines for the Revised Copyright Law Article 35." The important thing is that these operational guidelines do not yet have specific descriptions regarding the use of Generative AI.
② Copyright Law Article 30-4 (Utilization Not for the Purpose of Enjoyment)
This is a provision gaining attention in the AI era, stating that works can be used without permission if the use is not for the purpose of "enjoying" the expression of the work, such as for information analysis. However, it does not apply if the purpose includes savoring and utilizing the content (= enjoyment).
③ Agency for Cultural Affairs "Views on AI and Copyright," etc.
This is an organization presented by the Agency for Cultural Affairs in 2024, summarizing the thinking on copyright at each stage of AI input and generation.
④ MEXT "Guidelines for the Utilization of Generative AI in Elementary and Secondary Education (Ver. 2.0)"
This shows the basic thinking for utilizing Generative AI in schools, but it is a "guideline for usage" and does not grant permission to input individual copyrighted works. Clearing the utilization guidelines and clearing copyright are two separate hurdles that need to be checked independently.
And another practical keyword is "opt-out." This refers to settings that prevent input content from being used as training data for the AI. In this inquiry, I clearly stated as a premise that "services where opt-out settings are possible will be used with those settings enabled."
Furthermore, RAG-type tools like NotebookLM (a mechanism where you upload materials and have it answer based on that content) have unique points of contention. Even if not used for training, the uploaded materials are "accumulated" within the tool, and notebooks can be "shared" among multiple people. This "accumulation" and "sharing" could affect the judgment under Article 35 — a perspective clearly pointed out in Kyoiku Shuppan's response, and a part where I personally lacked resolution until I inquired.
The Questions I Sent to the Three Companies
I sent the following three points to the three companies (specifying that the use would be after performing opt-out settings).
[1] Inputting text into Generative AI for lesson research
Whether I, as a teacher, may input the text of materials published in textbooks (text only, not including illustrations or photos) into AI for the purpose of lesson research to utilize it for structural analysis, consideration of question examples, and consideration of blackboard plans.
[2] Creation of slide materials based on lesson research
Whether I may create slides to be presented to students in class (consisting of partial quotes or summaries of the text + original questions and diagrams) based on that lesson research.
[3] In-class use of slides
Whether using the created slides in the course of a lesson, targeting only the students in my own class, within the classroom and on learning terminals, is acceptable as being within the scope of Copyright Law Article 35 (no disclosure or distribution outside the school, and no sharing on SNS).
"Using it only for the lessons of my own class, within the necessary scope" — these are questions narrowed down to the minimum use case that teachers on the ground most want to know.
Supplement: Why I Asked with "Text Only" and "Opt-out Premise"
When creating the inquiry text, I was conscious of three things. I'll write them down as I think they will be helpful for those making similar inquiries in the future.
First, I narrowed the point of contention to "text of the body."
The pages of a textbook contain illustrations, photos, and diagrams in addition to text, and each has a different rights holder. The act of putting a page into AI as an image or PDF could potentially touch the rights of all of them. Therefore, in the question, I specified "text only, not including illustrations or photos" to minimize the points of dispute. In fact, Tokyo Shoseki's response clearly showed caution regarding the large number of rights holders for diagrams and photos, so I felt this narrowing down was the right choice.
Second, I wrote it on the premise of opt-out settings.
If you ask, "Can I put it into an AI that might use it for learning?" the answer is almost certainly NO. By setting the safest possible setting that can actually be chosen as a premise, it becomes easier for the rights holders to consider.
Third, I divided the use cases into three specific stages.
No one can answer a vague question like "Can I use AI?" By breaking down the actions into ① input for lesson research → ② creation of slides → ③ use in one's own class, and clearly stating the boundaries like "no disclosure or distribution outside the school, and no sharing on SNS," it becomes easier for the responding side to give a clear yes or no. I think making it easy for the other side to answer is a shortcut to drawing out a sincere response.
Mitsumura Tosho's Response: Clearly "No Problem" with Conditions
The response was the most concise of the three and could be directly applied to operations on the ground. The gist is as follows:
For [1] to [3], there is no problem as long as conditions are met. Those conditions are:
- It must be within the scope of Copyright Law Article 35, predicated on registration with SARTRAS.
- Loading into Generative AI must be limited to the scope necessary for the lesson.
- Settings to prevent data from being accumulated for learning (opt-out) must be performed.
- Please refrain from using the entire text or all pages of the textbook.
- Please refrain from sharing or broadcasting among teachers (as it would deviate from the scope of Article 35).
What is noteworthy is that they clearly state "sharing among teachers" is NG. The better the teaching materials made with AI, the more you want to share them with your grade level. However, what Article 35 protects is strictly "the course of that lesson," and the exchange of teaching material data between teachers is outside that scope — being able to confirm this line clearly in the words of the rights holder was a major gain.
Tokyo Shoseki's Response: Organization by Tool and Specific Examples of "NG if Exceeded"
Tokyo Shoseki's response was very specific, listing thoughts for each Generative AI tool and examples that exceed the scope. The gist is as follows:
Thinking by tool:
- For use with NotebookLM, if it is within the scope permitted under various laws and guidelines, no permission procedure is required.
- For use with ChatGPT and Gemini, if settings such as opt-out are made to prevent the AI itself from learning, and it is similarly within the permitted scope, no permission procedure is required.
As laws and guidelines to be referred to, SARTRAS's "Operational Guidelines for the Revised Copyright Law Article 35 (FY2021 Edition)" and the Agency for Cultural Affairs' "Views on AI and Copyright" and "AI and Copyright Checklist & Guidance" were cited.
And five specific examples that were deemed to "exceed" the scope:
- Copying or scanning the entire textbook or a large part of it (beyond the scope of one lesson unit) (the same applies to converting textbooks to PDF for Generative AI use).
- Using (uploading, etc.) scanned data for purposes other than the lesson in question.
- Further archiving the data (including data accumulation in Generative AI tools).
- Sharing teaching material data among teachers, etc.
- Publishing teaching material data (including independently developed applications, etc.) in a state where an unspecified large number of people can access it on the internet, or in a format that can be viewed semi-permanently.
Regarding my three questions, the answer was that "if the Generative AI tool used meets the aforementioned conditions, it can be used within the scope permitted under laws and guidelines," and for the "extent deemed necessary," they asked me to judge after checking SARTRAS's operational guidelines.
In addition, a very important note was added that for diagrams, illustrations, photos, and videos in textbooks, there are many rights holders other than the company (illustrators, painters, photo agencies, photographers, etc.), so for use that falls under the exceeded examples, prior permission from all those rights holders is required. My question's limitation to "text only" was precisely because I was conscious of this point.
Kyoiku Shuppan's Response: Most Cautious — "Cannot Grant Permission"
And Kyoiku Shuppan's response was the most cautious of the three. It was a sincere and weighty response that even showed the detailed consideration process. I will summarize the gist:
- The text of textbooks contains many copyrighted works for which third parties other than the company hold the rights, and they are not in a position to grant permission for all the text.
- The use of textbook text in Generative AI is considered to fall under "reproduction," "adaptation," and "public transmission" under the Copyright Law, and in principle, permission from each copyright holder is required.
- Copyright Law Article 30-4 is considered inapplicable. This is because the current usage of lesson research, slide creation, and classroom use is considered to include the purpose of "enjoying" the content of the textbook.
- Regarding Article 35 as well (while stating it as the company's own view), it is thought to be inapplicable. There is a possibility that "accumulation" of data through additional learning, "sharing" through AI, and data accumulation/sharing among multiple people in RAG-type tools could be judged as "exceeding the extent deemed necessary" or "unreasonably prejudicing the interests of the copyright holder."
- There is the Textbook Copyright Association as a contact point for rights processing, but since opinions on use in Generative AI are divided even among rights holders, there may be cases where permission cannot be obtained or high usage fees are incurred.
- Based on the above, even for copyrighted works for which the company holds the rights, they cannot grant permission regarding use in Generative AI.
In other words, practically speaking, the safe conclusion is not to input the text of Kyoiku Shuppan textbooks into Generative AI. In the response, it was also repeatedly and politely stated that whether to choose the limitation of rights provisions is up to the user's judgment and responsibility, and that whether it falls within the applicable scope is ultimately a judicial decision.
What I Saw by Lining Up the Three Companies
The stances of the three companies at a glance:
Before going into details, let's organize the overall picture.
- Mitsumura Tosho: "No problem" for [1] to [3] with conditions. Conditions are SARTRAS registration, necessary scope, opt-out, no full text, no sharing among teachers.
- Tokyo Shoseki: "No permission procedure required" if within the permitted scope of laws and guidelines. Organized NotebookLM and ChatGPT/Gemini (opt-out required) by tool, and clearly stated 5 NG examples that exceed the scope.
- Kyoiku Shuppan: Showed its own view that both Article 30-4 and Article 35 are "thought to be inapplicable" and "cannot grant permission regarding use in Generative AI." Effectively, an organization that one should refrain from inputting text.
"Conditional OK," "Conditional OK," "Effectively NG." The answers to the same questions were this divided.
The "Minimum Line" That Was Common
Although the intensity of the stance varied, there was a common line in the responses of the three companies.
First, opt-out (settings to not allow learning) is a major premise. This is the starting point, not an indulgence that makes everything OK as long as you do it.
Second, using the full text or a large part is out of the question. Even for companies that say it's okay to use, the scope is "the extent necessary for the lesson" or "one lesson unit." The act of converting a whole textbook to PDF and having AI read it should be considered out according to any company's response.
Third, sharing among teachers and publishing on the internet is NG. Article 35 protects up to "that teacher's course of that lesson." Placing teaching material data made with AI on a grade-level server, distributing it on SNS, or incorporating it into an app and publishing it — all the directions that make things more convenient are outside the scope.
Fourth, the final judgment and responsibility lie with the user. Every company clearly stated or suggested that whether to use the limitation of rights provisions is the user's judgment and responsibility, and the appropriateness is ultimately a judicial decision. It's not that "anything is fine because the publisher said OK."
The Biggest Difference: Whether Article 35 "Works" for Generative AI
On the other hand, what was decisively divided was the interpretation of whether Copyright Law Article 35 applies to input into Generative AI.
Mitsumura Tosho said, "There is no problem as long as it's within the scope of Article 35, predicated on SARTRAS registration." Tokyo Shoseki said, "No permission procedure is required if it's within the permitted scope of laws and guidelines." And Kyoiku Shuppan said, "Article 35 is thought to be inapplicable (own view)."
For the same law, asking the same question, the views are this divided. This is not because each company is careless, but because descriptions of Generative AI do not yet exist in SARTRAS's operational guidelines, and it is a transitional period where legal interpretation has not been solidified. In addition, textbooks are a collection of many third-party copyrighted works, and the composition of rights holders and the thinking on risk held by each company also differ. The fact that "the answer differs depending on the publisher" was the biggest lesson learned from this inquiry.
Daily Practice from Tomorrow: This Is What I Will Do
Based on the responses from the three companies, I have decided to operate as follows. I hope it serves as one example for your reference.
① Change the response for each textbook publisher.
Instead of a general theory like "Is AI utilization OK?" I will think in terms of "What about this textbook for this subject?" Since views are divided, it is logical to follow the thinking of the publisher of the textbook you are using. If you are unsure about a textbook, inquire. Based on this experience, each company will answer sincerely.
② Always check the opt-out settings.
Before using an AI service, check if it is in a setting or plan where the input content is not used for learning. If it's a service contracted by the school, start by checking the contract type with management or the ICT person in charge.
③ Input only "one lesson unit/necessary scope."
Even for textbooks from publishers that are not cautious, I will only input the necessary parts of the materials being handled. I will not load the full text or input PDFs that combine multiple units. Since page images containing illustrations and photos have more rights holders than text, I will not make them the subject of input in the first place.
④ Do not share or publish teaching material data made with AI.
I will not share prompts or generated products containing the text with colleagues or post them on SNS. If I feel like showing someone, "Isn't this question plan good?" I will limit it to the original part, excluding the textbook text part.
⑤ For textbooks from cautious publishers, consult without inputting the text.
For textbooks with cautious views like Kyoiku Shuppan, I will refrain from inputting the text itself. Instead, I will give the AI only "the name of the material, a memo summarizing the scene and theme in my own words, and the aim of the lesson" to consult on questions — a method of borrowing only the AI's thinking power without reproducing the copyrighted work that is the text. The accuracy will drop, but I think it's a realistic compromise to receive the benefits of AI while respecting the rights holder's intentions.
And above all, don't forget to check the school/municipality's Generative AI utilization guidelines and consult with management. Even if you clear copyright, rules for service and information management exist separately.
Case Study: Is This Usage Safe? Out?
With the responses of the three companies in mind, let's consider five common situations on the ground. The following is strictly my understanding based on each company's response and is not a definitive judgment, but please use it as a reference for your line of thought.
Situation ①: Inputting the text of one material to be handled tomorrow into an AI with opt-out settings already made to brainstorm question examples.
This is exactly my question [1]. According to the responses of Mitsumura Tosho and Tokyo Shoseki, it is considered permissible as long as conditions (SARTRAS registration, necessary scope, opt-out) are met. However, if it is a Kyoiku Shuppan textbook, the response in line with the company's view is to refrain from inputting the text itself.
Situation ②: Converting all the textbooks for a grade into PDF and uploading a year's worth to NotebookLM.
I think this is difficult according to the responses of any of the three companies. Tokyo Shoseki clearly states "reproduction beyond the scope of one lesson unit (the same applies to converting textbooks to PDF for AI use)" and "data accumulation in AI tools" as exceeded examples, and Mitsumura Tosho also says, "Please refrain from using the entire text or all pages."
Situation ③: Sharing a worksheet made with AI that includes quotes from the text in a grade-level folder with a teacher in the next class.
This is leaning towards OUT. Mitsumura Tosho clearly states "sharing/broadcasting among teachers" and Tokyo Shoseki clearly states "sharing teaching material data among teachers, etc." as being outside the scope. Article 35 protects up to "that teacher's that lesson." If sharing, it should be limited to the original part, excluding the textbook text part.
Situation ④: Sharing tips for prompts containing the text on X or note to contribute to everyone's AI utilization.
This is OUT. Tokyo Shoseki's response clearly states "publishing in a state where an unspecified large number of people can access it in an internet environment" as an exceeded example, and says the same applies when incorporating it into independently developed applications, etc. If broadcasting, it is necessary to process it into a form that does not include the text. For me, as someone who publishes teaching material apps, this was a sentence to take to heart.
Situation ⑤: Not inputting the text, but giving the AI only "the name of the material, a scene summary written in one's own words, and the aim" to consult on questions.
Since the copyrighted work that is the textbook text is not being reproduced, I believe this is a method that can be operated outside the current point of contention. It is a realistic alternative that can be used even for textbooks from publishers with cautious views like Kyoiku Shuppan (of course, it's a different story if the summary follows the expression of the text too closely).
What I Didn't Ask This Time — Remaining Points of Contention
This inquiry was narrowed down to the situation where "a teacher uses it for lesson research and their own class's lessons." Conversely, the following points of contention have not yet been verified.
When students input textbook text into AI.
What about activities where students themselves put the text into AI for summarization or dialogue in the course of a lesson? While students' use can be included in the "use in the course of a lesson" of Article 35, each company's view when Generative AI is involved cannot be directly read from this response.
In the case of teacher's manuals or compliant teaching materials.
The rights relationships and price structures differ between the textbook itself and manuals/workbooks. Whether manual data can be put into AI needs to be verified as a separate question.
Use in school administration.
Cases where text derived from textbooks is mixed in AI use for school administration that is not "the course of a lesson," such as creating report card comments or class newsletters. Since Article 35 is a provision for the purpose of lessons, the playing field changes in the first place.
After the revision of SARTRAS operational guidelines.
If descriptions of Generative AI are added to the operational guidelines, the current situation of "three companies, three ways" may change significantly. I will continue to watch this and will write another article if there are any movements.
There are still many things I don't know. But I think the value of this inquiry was being able to draw a map of "what is known and what is not known."
Important Request
There are several important notes regarding this article.
This article is not legal advice.
I am an elementary school teacher and not a legal expert. The introduction of each company's response is a summary by the author who strove not to lose the intent of the original text, and the final responsibility for interpretation lies with me.
Each company's response is as of the time of the inquiry (June 2026).
Every company clearly stated that it is a "response under the current laws and guidelines." If SARTRAS's operational guidelines correspond to Generative AI, the situation may change significantly.
Please do not use this article as a "basis for OK."
When actually using it, please be sure to check the primary information yourself: the publisher of the textbook you use, SARTRAS's operational guidelines, and the Agency for Cultural Affairs' materials. And if you are unsure, please inquire.
Finally, I would like to express my heartfelt gratitude to the representatives of Tokyo Shoseki, Mitsumura Tosho, and Kyoiku Shuppan who, despite their busy schedules, repeatedly considered and gave polite responses to an inquiry from a single teacher. The responses from the three companies, although their stances differed, were all permeated with a sincerity to protect the many rights holders involved in textbooks.
Conclusion: Don't Run Through the Gray with "It's Probably Fine"
To be honest, it took courage to send the inquiry email. I wondered if they would think, "Don't ask such trivial things." But what came back were three sincere responses.
We are in a position to teach children about copyright. An adult who teaches "Let's value the rights of the person who made it" shouldn't be running through a gray zone with "it's probably fine" in their own teaching preparation. Now that we have a powerful tool called AI, I think our attitude of facing rights head-on is being questioned.
If you ask, you'll get an answer. If you verify, you can see the range where you can use it with peace of mind. And if the person who verified shares that result (in a way that considers rights), the burden on the next person who verifies will decrease. I believe that the utilization of AI in educational settings can only take root healthily on top of such a steady accumulation of verifications.
I hope this article serves as a reference for the first step for teachers who have the same hazy feelings. And if you obtain a response by inquiring with the publisher of the textbook you use — please tell me that "fact that you verified."
If you found this article helpful, I would be encouraged if you could support me by sharing or following 🍀 I will continue to broadcast information on education x ICT from the field and my own apps. 👉 https://x.com/rohimotoiwaka/status/2042738992324710659?s=20
#EducationICT #GenerativeAI #Copyright #ElementarySchoolTeacher #HappySchoolWithGASApps
