5 Things Educators Wish AI Platforms Would Stop Doing

It starts with genuine optimism. A lecturer at a polytechnic in Singapore loads a new AI tool before the semester, spends a weekend configuring it, and tells her students it will be available to help them review course material. Three weeks later, she checks in. One student used it to draft a cover letter for a part-time job. Another asked it about cryptocurrency. Several got answers that had nothing to do with the module at all.
"It was helpful," one student tells her cheerfully. "Just not for the class."
She turns off the tool before midterms.
This is not a rare story. Educator adoption of AI is accelerating. By 2025, more than 60 percent of higher education institutions in the Asia-Pacific region reported piloting at least one AI-assisted learning tool. But adoption and satisfaction are not the same thing. The teachers and trainers who interact most closely with these tools, who watch students use them in real time, who field the confused questions, who deal with the aftermath of a bad AI interaction, carry a specific set of frustrations that rarely make it into product roadmaps or press releases.
What follows is not a takedown of AI in education. Most educators who've grown frustrated with AI tools still believe in what the technology could be. That is exactly why the gap between what they were promised and what they got stings as much as it does. Here are five things they wish AI platforms would stop doing.
1. Pretending to Know the Course When They Don't
The first problem is also the most fundamental. Most general-purpose AI tools, even those marketed for education, respond to whatever a student asks, drawing from a vast and largely ungoverned body of training data. Ask about Chapter 4 of a specific module, and you might get an answer that sounds authoritative but was assembled from a different textbook, a different curriculum, or a source the educator has never seen and wouldn't have approved.
For teachers, this is not a minor inconvenience. It is a professional problem. Educators spend significant time deciding what their students should learn, in what sequence, and with what emphasis. A tool that bypasses that structure doesn't save time. It undermines it.
A 2024 survey by the EdTech Hub found that 71 percent of teachers who had tried AI tools for student support felt the tool's responses were "not consistently aligned" with their course content. Many described correcting student misconceptions that had come directly from the AI.
What solving it actually looks like: an AI that works exclusively within the materials the educator has approved (lesson notes, course documents, specific readings) and declines to speculate beyond them. When Noodle Factory's Walter is loaded for a course, it grounds itself in what the educator uploads. If the answer isn't in the approved content, Walter says so rather than reaching for something that sounds plausible.
2. Keeping Educators in the Dark
The second frustration is about visibility. Most AI tools give students a private conversation with the platform and give educators nothing. No summary of what questions were asked. No flag when a student is circling the same concept repeatedly. No signal that half the class asked about the same topic the night before an exam, which would be, for any observant teacher, extraordinarily useful information.
This invisibility serves no one well. Educators cannot improve their teaching without feedback. Students who are quietly confused stay quietly confused. The AI becomes a black box that sits between teacher and learner, and nobody can see inside it.
Research from the University of Melbourne published in 2024 found that formative feedback loops, knowing in real time what students are struggling with, are among the highest-leverage interventions in improving learning outcomes. AI tools that generate no educator-facing data actively close that loop.
What solving it actually looks like: a dashboard. Not a data dump, but a meaningful summary: what topics students asked about most, which students have been engaging, where confusion clusters. Educators should be able to see their AI as an extension of their own observation, not a replacement for it.
3. Wandering Off-Topic
This one sounds almost comical until you have lived through it. Students are inventive, and they quickly learn the contours of any tool. Give them an AI with no topic guardrails and they will, entirely reasonably, use it for everything: homework in other subjects, personal advice, creative writing projects, arguments with their friends. The tool becomes a free-for-all, and the educator who introduced it ends up answering for outputs she never sanctioned.
The off-topic problem is partly a guardrail failure and partly a scope failure. Many platforms were designed to be as broadly useful as possible, because breadth is a selling point in consumer markets. But a classroom is not a consumer context. A student asking an AI to help draft a resignation letter for her part-time job is not being served educationally. She's just getting a free writing assistant.
The OECD's 2024 report on AI in education flagged "contextual drift" (AI tools being used for purposes outside their stated educational scope) as one of the primary reasons educators disengage from AI platforms after initial adoption.
What solving it actually looks like: an AI that holds its lane. When a question falls outside the course's scope, the tool should acknowledge the question warmly and redirect, not because it's incapable of answering, but because staying on topic is what serving the student's learning actually requires.
4. Being Built for Admins, Not Teachers
Many AI tools sold into educational institutions are not really designed for the people who teach. They are designed for the person who buys them: a department head, a chief digital officer, a procurement committee. The interface is clean. The compliance documentation is thorough. The onboarding requires an IT team.
Then the tool reaches a teacher. And the teacher, who has forty minutes between classes and no IT background, cannot figure out how to upload her notes, cannot understand why the AI keeps giving different answers, and cannot find any documentation written in plain language.
This is a design choice, not an accident. Enterprise EdTech has historically been sold top-down, and tools reflect the priorities of buyers rather than users. A 2025 report from EDUCAUSE found that ease of use by individual instructors, not administrative features, was the single most cited factor in whether a tool was actually adopted at the classroom level. Institutions were buying. Teachers were not using.
What solving it actually looks like: a setup process that an educator can complete alone, in an afternoon, without submitting a support ticket. Uploading a PDF and setting a learning objective should not require a training day. Educators are not the barrier to AI adoption; confusing interfaces are.
5. Ignoring the Rubric
Perhaps the most telling frustration is the most specific. When an educator designs an assessment, she is making decisions about what matters: what counts as a good answer, what level of detail is expected, what concepts must be demonstrated. These decisions live in a rubric. A rubric is, in a meaningful sense, a map of the educator's pedagogical intent.
Most AI tools have never seen it.
Ask a student a practice question through most AI platforms, and the AI will evaluate the response against its own internal standard, which may or may not resemble the educator's. A student gets feedback that says her answer was "clear and well-structured," but the rubric requires her to apply a specific framework from the course. She hasn't. The AI didn't know to check.
A 2024 study from the National Institute of Education in Singapore found that formative AI feedback was most effective when it was explicitly aligned to instructor-defined criteria, and least effective when students received generic positive reinforcement that didn't map to their actual performance on formal assessments.
What solving it actually looks like: an AI that can be given the rubric as a reference point. Not a grading machine (assessment judgment belongs with educators) but a practice tool that primes students against the right criteria, so that when they reach the formal assessment, the feedback they received along the way was pointing in the right direction.
The Standard Is Higher Now
None of these frustrations are inevitable. They are design choices: decisions made about what to optimise for, what to make visible, what to restrict. And the educators who have felt let down by AI tools are not anti-technology. They are, in many cases, the people who argued internally for AI adoption, who stayed up configuring the platform, who wanted this to work.
They are also the people best positioned to say what good actually looks like. Educators who have tried these tools and found them wanting are not a problem for EdTech to manage. They are a specification for what the next generation of tools needs to be.
The bar is not that high. Know the course. Show us what students asked. Stay on topic. Be usable without an IT team. Work from the rubric. These are not futuristic demands. They are the same things a good human teaching assistant would do on the first day.
Further Reading
What Makes an AI Tutor Actually Teach? — The design principles that separate AI tools educators can trust from those that just sound helpful — course-grounding, scaffolding, and knowing when not to answer.
Is Your Institution Ready for Agentic AI? A Readiness Checklist for Universities — Before the next generation of AI tools arrives, here's how to assess whether your institution has the infrastructure, policies, and culture to use them well.
What SEA Educators Actually Think About AI in the Classroom — The gap between what the surveys say about AI optimism in Southeast Asia and what educators on the ground are actually experiencing.
Sources
EdTech Hub. (2024). Teacher Perspectives on AI-Assisted Learning Tools in Asia-Pacific. edtechhub.org
University of Melbourne. (2024). Formative Feedback and Learning Outcomes: A Review of Digital Interventions. unimelb.edu.au
OECD. (2024). AI in Education: Emerging Insights and Policy Implications. oecd.org/education
EDUCAUSE. (2025). 2025 EDUCAUSE Horizon Report: Teaching and Learning Edition. educause.edu
National Institute of Education, Singapore. (2024). Rubric-Aligned AI Feedback and Student Performance in Formative Assessment. nie.edu.sg


