The AI Learning Hub Journal

When Not to Use AI

Five situations where the right move is to keep AI out of itnone of these is about the tool being bad — each is about what this particular task needs1When the point is that you build the skillThe task exists to change you, not to produce a file — outsourcing it skips the changeyou are the deliverable2When you cannot check it and it mattersFluent and wrong look identical from outside, so unverifiable plus consequential is a bad pairno way to catch a mistake3When the input is private, or not yours to shareSomeone else's messages, medical details, ID documents, or anything under a confidentiality rulenot your data to paste4When a real person should be in itAn apology, a hard conversation, a message that only counts because you wrote it yourselfthe sender is the point5When it is simply not allowedExam rules, school policy, a workplace or client agreement — the rule holds whether you agree or notknow the rule firstTHE THREE QUESTIONS THAT COVER ALL FIVEWhat is this task for?If the point is the learning,doing it yourself is the point.Could I tell if it were wrong?If not, and it matters, get asource or a person involved.Whose is this, really?Someone else's information,or someone else's decision.Most everyday uses are fine — these five are the ones worth stopping forThe question is never whether AI is allowed in general, but whether it belongs in this taskIf you would not be comfortable saying you used it here, that is the signal
Judgement, not a ban list — the same tool is right in one task and wrong in the next

When the Struggle Is the Point

Some tasks exist to change you rather than to produce an artefact. Practising a language, working through a maths problem set, drafting your first argument in an unfamiliar subject, learning to code — the difficulty is the mechanism. Removing it removes the benefit and leaves you with a finished object and no new capability. Musicians do not get better by listening to recordings of the piece. There is a real version of this that is worth protecting: sit with the problem for a while, get properly stuck, then ask for a hint rather than the answer. Effort first, help second, is a rule that survives contact with real deadlines.

  • If the task is meant to build a skill, shortcutting it defeats the entire purpose
  • Get stuck first, then ask for a hint — not the solution
  • Practice, drafting and problem sets are the highest-value places to stay unassisted
  • The output was never the point in these cases; you were

When It Is Not Your Voice to Outsource

Some writing is valuable precisely because it came from you. A personal statement, a message to someone you have hurt, a reflection on your own experience, a note in a card, a text to a friend going through something. Generated text in these contexts reads as slightly hollow even when people cannot say why, and being caught faking sincerity costs more than writing something clumsy but real. There is a middle path — write it yourself and ask for feedback on clarity — that keeps your voice while fixing your grammar. The rule of thumb: if the value is in the fact that you wrote it, write it.

  • Personal statements, apologies and condolences lose their function when generated
  • Admissions readers and interviewers see enormous volumes of generic text and notice
  • Acceptable middle: your words, AI feedback on clarity and structure
  • If it matters that it came from you, then it has to come from you

When It Is Genuinely the Wrong Tool

Some cases are just bad fits. Real-time facts — prices, scores, timetables, whether something is open — belong to a source that actually knows, not to a model recalling training data. Anything with serious personal consequences, like medical symptoms, legal questions, or a mental health crisis, needs a qualified human, and a fluent answer is actively dangerous because it feels authoritative. Confidential information about other people should not be pasted in at all. And decisions that depend on your values are not information problems: an AI can lay out considerations, but choosing what you care about is not something you can delegate.

  • Live, changing facts: use the source that owns the data, not a model
  • Health, legal and crisis situations: a fluent answer is not a qualified one — go to a real person
  • Other people's private information should never be pasted into a chat box
  • Value-laden choices can be informed by AI but not made by it

Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.