Tuesday evening, half past eight. An English teacher still has fourteen essays to mark and opens ChatGPT: "Give feedback on this argumentative essay by a year 10 pupil." She pastes the whole document. At the top are the pupil's name, class and date, and in the introduction he writes about his parents' divorce.
The rule of thumb is short: the task is fine, the pupil is not. Asking for feedback on a text, turning a grade list into an overview, tightening a care note, all of that can be done with a chatbot. What has to go before you paste is everything that points to the pupil: name, class, pupil number, date of birth, and anything about care, health or home. Whether it is allowed with this account at all depends on what the school has decided about the tool.
Why is a prompt different from an email to a colleague?
An email to the learning support coordinator stays inside the school, under the school's own agreements. A prompt goes to a vendor. What happens to it then depends on the account. On a free or personal account the input can be kept and used to improve the model, unless you have switched that off, and there is no data processing agreement underneath. On a school account with an agreement, different terms apply, and the school knows them.
This is data about minors, often about health or home circumstances. Under the GDPR a stricter regime applies to it, and a prompt is processing, not a notepad. The Dutch Data Protection Authority reported receiving several data breach notifications in which employees had entered personal data into an AI chatbot, and advises organisations to make agreements on which tools are allowed and which data may go in. The general list of what never belongs in a prompt is in what you may and may not put in an AI prompt. This piece is about the three tasks a teacher has most often.
Three tasks, and what comes out
| Task | What stays in | What comes out before you paste |
|---|---|---|
| Feedback on an essay | The text itself, the level, the assessment criteria | Name, class, date, and passages where the pupil writes about themselves or their family |
| Converting or analysing a grade list | The grades and the subjects, with a number per pupil that only you can translate back | Names, pupil numbers, dates of birth, and the class if it is small |
| Tightening a care note | The structure of the note, the kind of concern in general words | Everything: name, class, diagnosis, family situation, the date of the conversation |
The third row is the honest one. A care note consists of data you cannot leave out without the note disappearing. If you want it shorter, use the environment the school has set up for that under an agreement, or do it yourself. A free chatbot is not an option for that task, however well it summarises.
Why is removing the name not enough?
Because a class is small. "A year 10 pupil who has had an adjusted timetable since the autumn break and whose parents are separating" points to exactly one person, without a name, for everyone who knows the class. A date of birth plus a class does the same. Dates are the underrated group: the date of a conversation or an incident is known to the colleagues who were there. Why combinations of harmless fields single someone out is in why "take the names out" is not enough.
For the essay that means: not only the name at the top, also the paragraph about home. For the grade list: a class of eleven with one fail is, with the subject attached, already traceable for the department.
How do you still get the task done?
Essay: remove the header and replace the name in the text with "the writer". If there is a personal passage, leave it out and ask for feedback on structure and argument only. The feedback is just as useful, because it is about the text.
Grade list: in a copy, turn the names into numbers, keep the key at school, and paste the columns with grades. Ask for the overview, then put the names back yourself.
Care note: not in a free account. If the school has an environment with a data processing agreement, use that, and still without a citizen service number. More techniques for cleaning a prompt without losing the task are in the prompt redaction guide.
What if the school has not decided on the tool yet?
Then the instruction that belongs to "no decision yet" applies: public information only, and report the tool to whoever is responsible. That is not a ban. It is the honest state of affairs, and it saves you guessing. Which tools the school has assessed and under which condition belongs in the school's AI policy, and that policy only works if it reaches you at the moment you open the tool, not in a document from last school year.
This is what that can look like. BeeSensible shows the school's decision when you open an AI tool in the browser, with the approved alternative as the first button, and marks sensitive data while you type in the supported apps. You choose what happens to it: mask, remove, or leave it. Nothing is blocked, and what you type is not stored.
