Microsoft and Google will train your teaching staff to use Copilot or Gemini for Education competently in an afternoon, and to their credit, the training is genuinely well made: clear screens, sensible prompts, a certificate at the end. What it does not do, because it is not designed to, is teach anyone when to trust what the tool just produced.
That distinction sounds academic until a school actually sits with it. Vendor training is, by its nature, product training: here is the interface, here are the buttons, here is a good prompt for a lesson plan or a parent email. It is thorough on the mechanics and almost entirely silent on the judgment, because judgment is not a feature the vendor can ship, and teaching it properly means admitting the tool is sometimes wrong in ways that are hard to spot.
The gap is not about skill, it's about who owns the risk
A vendor has no reason to dwell on the ways their own product misleads a busy teacher, so the training understandably moves quickly past it. But a teacher who has completed the course and now uses AI to draft a rubric, summarise a student's writing progress, or flag a possible case of plagiarism is making judgment calls the training never touched: is this output accurate for my specific class, does it reflect my school's actual standards, and what happens if I am wrong. The vendor is not negligent here, they are simply optimising for adoption, which is a different goal from optimising for a teacher's professional judgment under pressure.
Ask whoever ran your last AI training session one question: did it include a worked example of the tool getting something wrong, and what a teacher should do about it? If the answer is no, the training taught the interface, not the judgment, and that gap is worth closing separately, in-house, with your own subjects and your own standards as the material.
What schools notice, eventually
The schools that notice this gap tend to notice it the same way: not during the training itself, which usually goes well and generates good feedback, but a few months later, when a teacher raises a case where the AI-drafted feedback on a student's essay was fluent, confident, and simply wrong about what the student had actually argued. At that point the question stops being "can our staff use the tool" and becomes "do our staff know when not to trust it," which is a training question nobody sold them, because it is not a training question a tool vendor is well placed to answer.
"Honestly, I'm pretty sick of all these training sessions," a teacher said to me. "I get that there are great tools I can learn to use to make my classes better, but I just don't see it. How do I get AI to help me in the actual learning process of my students? It's more than just knowing the menu system, or where or what to click. I wish the training was more on how exactly to get students to where they should be in their growth and thinking."
That complaint is not about AI getting an answer wrong. It is the same gap from the other side of the desk: a teacher who has sat through the interface training and still cannot say how any of it moves a specific student's thinking forward, because nobody trained that part.
Why this matters more in a Malaysian school context
Malaysian schools are adopting vendor AI tools quickly, often through existing Microsoft or Google education licences that already include the training modules, which makes the rollout fast and the cost close to zero. That is a reasonable way to get started. But it means the judgment layer, if it is taught at all, has to be added deliberately and separately, by the school itself, because the fast and free path stops exactly where the harder and more important part begins.
Want your staff trained on the judgment, not just the tool?
I run short briefings for school leadership and teaching staff that sit alongside whatever vendor training you already have, focused specifically on when to trust AI output and when to stop and check.
Book a BriefingWhat This Means for You
Vendor training is not wrong to include, and there is no reason to skip the free afternoon that gets staff comfortable with the interface. But it is only half the job, and the half it leaves out, knowing when the tool's confident answer deserves scepticism, is the half that actually protects your students. That half has to be taught by someone who owns the outcome, not someone selling the licence.