SUBMIT A SUPPORT REQUEST
Teachers have always created great resources. The question now is who benefits when AI helps make them.
There is already a well-established market for teacher-created educational resources. Platforms like TES allow teachers to publish and sell lesson materials directly to other educators, keeping the income themselves. It is a legitimate side income for many, and for some a significant one. The resources are created in their own time, using their own expertise, and the platform handles the transaction. The school sees none of it.
That model works. But it does raise a question worth sitting with: when a teacher spends their Sunday building a brilliant set of history resources that 4,000 other teachers then buy, what does the school that employed and shaped that teacher’s practice actually get out of it? The answer, currently, is nothing. There is a different model possible one where the intellectual output of a school’s staff generates income that flows back to the school. That is a conversation for another post. What we want to explore here is what AI now makes possible in the resource-creation space, and what the legal and practical realities look like for anyone thinking about using it.
What AI can now generate for the classroom
The range of content AI tools can now produce at speed is genuinely broad. Text-based resources, worksheets, lesson plans, differentiated tasks, reading comprehension passages have been within reach of language models for a couple of years. But the more interesting shift is in visual and three-dimensional content.
Tools like 3D AI Studio (3daistudio.com) allow a teacher to generate a 3D model from a text prompt or a photograph in a matter of seconds. Type “cross-section of a human heart” or “Roman amphitheatre” and you get a manipulable 3D object rather than a flat image from a search engine. Canva has integrated AI-powered 3D generation directly into its interface, which many schools are already using for general design work. Adobe Photoshop’s generative fill and generative expand features allow educators who already have design skills to extend, alter, or create images with a written prompt useful for producing bespoke visuals that match a school’s specific context rather than relying on stock photography.
For primary schools in particular, the ability to produce custom illustrations or characters that reflect the school’s community, settings that are locally recognisable, diagrams that match exactly what was discussed in a lesson is valuable in a way that generic clip art never was. The barrier to entry is low. You do not need to be a designer. You need to know what you want and be able to describe it.
Using AI for classroom resources is broadly fine. Selling those resources commercially is where it gets complicated.
The distinction that matters
The legal question nobody is asking clearly enough
There is an important distinction between generating AI content for your own classroom use and generating it with the intention of selling it to others. For internal use a worksheet, a display, an explanatory image for a Year 4 science topic the practical and legal risk is low. You are using a tool to support your teaching. No different, in spirit, from using Canva or Word.
The moment you intend to sell that content, the picture changes. Under current UK law, AI-generated works occupy an ambiguous position. The Copyright, Designs and Patents Act 1988 does contain a provision, section 9(3), that attributes authorship of computer-generated works to the person who made the arrangements for the creation of the work. In principle, this could protect AI-generated content. In practice, the courts have not tested this thoroughly in the context of modern generative AI, and legal opinion is divided on whether the provision was ever intended to apply to outputs from large language models or diffusion models.
The more pressing issue is what the AI tool itself was trained on. If the model was trained using copyrighted images without a licence and many were then there is a reasonable argument that the output contains derivative elements of those works. The UK government ran a consultation on this between December 2024 and February 2025, and as of March 2026 has chosen to maintain the current legal position while gathering further evidence, rather than introducing a broad commercial text and data mining exception. That means the legal landscape for selling AI-generated content remains genuinely unclear. It is not banned. It is not definitively protected. Platforms like TES are beginning to require sellers to confirm that submitted resources comply with the terms of any AI tools used including whether those tools permit commercial use of their outputs.
The practical advice is straightforward: before you sell anything generated with an AI tool, read that tool’s terms of service carefully. Some explicitly permit commercial use of outputs. Others do not. 3D AI Studio, for instance, publishes guidance on this directly on their site. Canva’s terms have specific clauses governing AI-generated content and what you can do with it commercially. Photoshop’s generative features are licensed under Adobe’s commercial terms, but the position on derivative liability from training data is not settled. This is not a reason to avoid these tools. It is a reason to know what you are using and why.
What about putting AI-generated content on a school website?
This is a question that comes up more than you might expect. The short answer is: yes, you can use AI-generated images and content on a school website, and there is no blanket rule against it. The more useful question is how Google treats it, because that has practical consequences for any school that cares about being found in search results.
Google’s stated position is that it does not penalise AI-generated content as a category. What it penalises is low-quality content produced at scale with the primary purpose of manipulating search rankings; what it calls spam. The March 2024 core update specifically targeted this type of output and removed a significant volume of it from search results. Google’s ranking systems reward original, useful content that demonstrates genuine expertise, experience, authoritativeness, and trustworthiness otherwise known as the E-E-A-T framework. A school page about its SEND provision, written with real knowledge of the school’s context and updated regularly, will outperform a generic AI-generated page that could have been written about any school anywhere. The issue is not AI. The issue is whether the content is genuine and useful to the person reading it.
For school websites specifically, the risk of AI-generated content looking spammy is low if it is used to support and extend real content rather than replace it. An AI-generated illustration used on a science page is not a problem. A page full of AI-generated filler text that says nothing specific about the school, its values, or its children is a problem and not just for Google. It is a problem for parents trying to understand whether a school is right for their child.
Skills files, MCP, and the ability to update content automatically
One area we work in directly at iTCHYROBOT is the use of structured instruction files called skill files that we combined with MCP (Model Context Protocol) integrations to allow AI to interact with live systems. In practical terms, this means an AI model can be given a set of rules about how to write, what to include, what to avoid, and how to format content, and then use those rules to push content directly to a website, a document system, or any other connected platform. They can be used to extract content and review it. Imagine opening Claude and asking it to look at your website and see if any of the policies still have Mrs Smith as the Head teacher who left last year on them. Or is the SENDCo listed as the correct person on all pages. Well this is the world we now live in and the world that at iTCHYROBOT we work in day in, day out.
A note from the human author (Rob)
As a test this post was created using exactly the workflow described above. As technical director I wrote a base article in Claude desktop before invoking the skills file to audit and improve what was written. That included looking at trusted sites and sources I had provided such as where to find the relevant legal information of copyright. The result is an article that has passed through defined editorial standards, the category structure, the tag vocabulary, and the Gutenberg block templates used on this site. The end result is a properly formatted standardised article that can be reviewed and published. An MCP server connects the AI session to the WordPress REST API. The result is a consistent, on-brand post that meets the same editorial bar as anything written manually. Well in all honesty probably a better formatted documented. This can be without requiring someone to open a browser, log in, and format a post by hand. However, I still believe in us humans having the final veto so I am editing this post and ensuring it is written the way I want with the info presented before I hit publish! The same approach can be applied to keeping a school website updated: news items, policy page refreshes, staff pages, term dates. The AI does not replace the person who knows the school. It removes the friction between knowing what needs to be said and getting it published.
I do not intend on publishing a full how-to guide on this as the setup requires some technical groundwork and varies depending on what systems a school already has in place and what you would like to integrate and automate. This is a field of active research for us in iTCHYROBOT at present and we do like a challenge so if you have a use case whether it is keeping your website current, generating differentiated resources at scale, or thinking about whether AI-generated content could work for your school let me know your thoughts and lets see what is possible in this generative AI led future we are heading into.
All the best
Rob (rob@itchyrobot.co.uk)
