Before publishing: ten recommendations

Artificial intelligence (AI) can make mistakes with complete naturalness, and an error in a simulation or a quiz ends up as mistaken learning. Before publishing, the material has to be used as students would use it, checking the concepts, the data and the answers it accepts as correct. This review cannot be delegated, since responsibility for what is taught lies with the person who publishes it.

Students’ names, marks, voices or images are personal data. In Spain, the Spanish Data Protection Agency states in its guide for schools that teachers must use the tools provided by the school or the education authority, and that content a teacher publishes on their own, outside the school, is their own responsibility. Other countries have different rules, but the precaution is the same. The simplest way to comply is for the material not to ask for data. When a tool needs to identify students, such as a gradebook, the data must stay on the teacher’s device or in the systems the school has decided to use. Particular care is needed with platforms that easily add user accounts and databases, since the data are then stored on third-party servers.

If nobody understands how a resource works, it cannot be fixed when it fails, and in practice it stops being open. Meeting this point does not require knowing how to program, since it is enough to be able to describe briefly what the application does, what it stores and whether it communicates with any external service.

A resource that embeds content from another website, or loads parts from third-party servers, stops working when those services change or close. The same happens with the platform where the material was created, since the shared link lasts as long as the company decides.

Materials generated with AI tend towards the showy, and decorative effects are often an obstacle for some students. An accessible resource can be used without a mouse, can be understood without relying on colour and reads well on a small screen.

The images, texts, sounds and pieces of software that are incorporated have authors and licences, even if the AI placed them there. The attribution must go inside the material itself, so that it travels with it when it circulates out of context.

With AI, work moves very fast, and a few weeks later nobody remembers why each decision was taken. Keeping that record makes it possible to resume the work, to explain it to someone else and not to undo by mistake something that had a reason.

In a resource created with vibe coding, the code is the work of the AI, and usually nobody has reviewed it line by line. The people who reuse it need to know this in order to decide how far they can trust it. It is therefore advisable to state which tool was used to create it and, above all, what the person who publishes it has checked, such as the correctness of the content, how it works or how data are handled.

Works are protected by copyright automatically, so a resource without a licence cannot be safely reused even if it has been published. A free licence tells other people that they may use it, adapt it and share it, and on what conditions. Code and content need different licences, and what the AI generates raises questions of authorship that are dealt with in its chapter.

The licence gives permission, but the material must also be obtainable in a form that allows work to be done on it. That way someone else can adapt it to their classroom without asking anyone for anything.