FAQ

Open ChatGPT, Claude, or Gemini in a browser tab and use it for one real task you'd do this week — a lesson outline, a parent email, a rubric draft. Don't read a book about it first; the only way to build judgment is by feeling the gap between what it does well and where it goes sideways. After ten or fifteen of those small tasks you'll have more practical intuition than most consultants.

For 80% of educator work, a single capable chatbot is plenty — ChatGPT, Claude, or Gemini will all do the job. Specialized tools (Diffit, MagicSchool, Khanmigo) wrap the same underlying models in workflows; useful, but optional. Master one general tool first; then add specialized ones only when you have a specific friction they remove.

Pick whichever your institution already has a license or data agreement for — that decision matters more than which model is two percentage points smarter this month. If you have no constraint, Claude tends to be strongest for long-form writing and reasoning, ChatGPT for breadth and ecosystem, Gemini for anything tied to Google Workspace. They are closer to each other than the marketing suggests.

A chatbot answers whatever you ask. An AI tutor is a chatbot wrapped in a pedagogical scaffold — it's instructed to ask questions rather than give answers, track a learner across sessions, and stay inside a curriculum. The underlying model is usually the same; the value is in the scaffold and the data the tutor accumulates about the learner.

For occasional use, free tiers are genuinely usable. If AI becomes a daily part of your work — and for most educators within a year it will — the paid tier (around $20/month) gives you the better model, longer context, file uploads, and fewer rate limits. Treat it like a professional tool, not a toy.

Banning is a posture, not a policy — it punishes the students who comply and rewards the ones who don't. The defensible move is to regulate where AI is allowed, require disclosure, and redesign the assignments where unsupervised AI use makes the grade meaningless. Bans look decisive in board meetings; in practice they delegate the problem to enforcement that doesn't exist.

No. False positives are common, they disadvantage non-native English writers, and the major detector vendors themselves quietly walk back their own accuracy claims. Using a detector score as evidence in an academic integrity case is how institutions end up in lawsuits and apology letters. Use them, if at all, as a prompt for a conversation — never as proof.

The serious liability surfaces are data privacy (student data going into a vendor's training pipeline), accessibility (an AI tool that doesn't work for students with disabilities), and discriminatory outcomes (a tool that systematically fails one group). Student misuse of AI is mostly an academic integrity matter, not a legal one. Talk to your counsel about the first three; talk to your faculty about the fourth.

Default consumer tiers — the free ChatGPT, Claude, or Gemini you sign up for with an email — generally reserve the right to use your inputs to improve their models. Education and enterprise tiers offer contractual data protection, but you have to read the DPA, not the marketing page. Never enter identifiable student information into a consumer tier; for anything more than anonymized examples, route through a tool your institution has actually contracted for.

It depends entirely on what the student is doing with it. AI used to skip the thinking — generate the answer, copy it down — atrophies the same muscles a calculator does for arithmetic, and that's a real concern for foundational skills. AI used as a tutor, a critic, or a sparring partner can deepen learning beyond what a single teacher can provide. The tool is neutral; the assignment design is what determines which version you get.

Cheating is the violation of a stated rule for the assignment in front of the student. If your syllabus says no AI on this essay and the student used AI, that's cheating. If your syllabus is silent and the student used AI to brainstorm and then wrote the essay themselves, that's tool use. The fix is not to litigate every case after the fact — it's to write assignments with explicit AI rules and teach students the distinction.

Shift weight from finished products you can't supervise to process you can — drafts with revision history, oral defenses, in-class writing, project work where the student has to explain their own decisions. Assume the take-home essay is dead as a high-stakes measure of writing ability; it was already wounded by paid essay mills. The good news is that process-based assessment measures more of what we actually want to measure.

On surface mechanics — grammar, structure, clarity — it's already as good as most teachers, available at 2 a.m., and infinitely patient. On substance — whether the argument is actually true, whether the analysis goes deep enough — it's mediocre and sometimes confidently wrong. Treat it as a first pass that frees you to focus your own feedback on the things that require your judgment.

Direct unsupervised use by under-13s is mostly a bad idea — both because of the terms of service of major tools and because the foundational skills (reading fluency, handwriting, arithmetic, attention) need to be built before they can be productively offloaded. Teacher-mediated use, where the adult drives the AI and the child sees the output, can be genuinely useful. The principle: build the muscle before you give them the machine.

Start with three sections: what's permitted by default, what requires disclosure, and what's prohibited. Then specify how those rules apply across student work, faculty work, and operational/administrative use — because those three contexts need different rules. Keep the document short, dated, and explicitly provisional; whatever you write this year will need rewriting next year, and pretending otherwise produces brittle policy.

Neither, alone. IT owns infrastructure, security, and vendor contracts; academic leadership owns pedagogy, assessment design, and what AI means for learning. If IT runs it alone you get a tools rollout with no pedagogical change. If academics run it alone you get pedagogical intent with no working infrastructure. The institutions getting this right have a small joint group with explicit decision rights between the two.

Four things, in roughly this order: a working policy, a contracted tool stack the institution actually pays for, faculty who have hands-on time with the tools, and assessment design that has been honestly revisited. Most institutions claim readiness when they have one of these — usually the policy. Real readiness is when all four are present and a new faculty hire would experience them on day one.

Don't try to convert the skeptics first; find the curious ones, give them time and a paid license, and let them become the in-house examples. Mandatory training for the resistant tends to harden the resistance. Once the curious group has a few months of real practice, the skeptics either join voluntarily or self-identify as a different conversation — which is also useful information.

Using AI to generate first-pass feedback that a human reviews before it reaches a student: fine, often an improvement. Using AI to assign a final grade that determines a student's standing without human review: not fine, and in some jurisdictions not legal. The line is whether a human with authority and accountability has actually looked at the result before it becomes consequential for the student.

The honest answer is that AI in education is currently widening gaps, not closing them. Students with paid tiers, supportive home environments, and English as a first language extract far more value than students without. The institutional response isn't to ban AI for the advantaged group; it's to make sure the disadvantaged group gets equivalent access through the institution itself — contracted licenses, supervised practice time, explicit instruction in how to use it well.

Yes — proactively, in plain language, and before they find out some other way. Parents are not the obstacle; the institutions that have run honest information sessions about how AI is used in teaching, grading, and operations have found parents broadly supportive once the actual practices are visible. The damage is done by discovery, not by disclosure.

Major models reflect the biases of their training data — Western, English-dominant, with documented blind spots around names, dialects, and cultures outside that center. In a classroom that shows up as feedback that's harsher on writing in non-standard English or examples that don't reflect your students' world. The mitigation is to read AI output as a draft to be edited, not a verdict to be trusted, and to teach students to do the same.

It's curated by Carlos Miranda Levy, a long-time educator, technologist, and social entrepreneur, as part of a small family of sister sites focused on AI's impact across different domains. The editorial point of view is one person's — informed by three decades of work in education, technology, and content — not a committee or a vendor. If you disagree with a position taken here, you're disagreeing with a specific human, which is the point.

Yes, eventually — consulting, training, and an AI suite for institutions are listed under Services, and that's how the work sustains itself. The vast majority of the site is free reference content with no gate. We don't run ads, we don't sell your data, and we don't take vendor money to recommend specific tools. If a service might be a fit for your institution, it will be obvious; otherwise the content stands on its own.

Honestly: both. Carlos sets the editorial direction, the positions taken, and the voice; AI tools help draft, translate, and structure; a human reviews everything before it goes live. That's the same workflow we recommend to educators in the rest of the site, and we'd be hypocrites to hide it. What you should expect is human judgment about what's worth saying, AI help producing the words, and human accountability for the result.