My PhD is in international education, and I have been publishing Wandering Educators for almost two decades. In all that time, I have never watched a single technology arrive in classrooms and kitchens this fast, or with this little instruction attached.
What do I mean?
A seventh grader opens a chatbot to ask about the Silk Road and gets back four paragraphs of fluent, organized, confident prose, two facts of which are wrong. A teacher asks for a rubric and gets one that is honestly, surprisingly good! A parent asks whether any of this is cheating and nobody at the school has an answer yet, because the policy meeting is scheduled for next month (if quorum is achieved). A graduate student submits a literature review with six citations in it...and three of those papers have never existed.
All four of those people are using the same tool. Only one of them got burned, and (this is the part that should worry us) it was not the one who used it the most.
That gap is what AI literacy actually is, and it is why we built this Compendium. Everything we have published about artificial intelligence can be found here, organized by what you are actually trying to do, with the reasoning underneath it rather than just the tips. ALL of it, in one place, finally!
What do my kids and my students actually need to understand about AI?
Let’s talk about it.
What AI literacy means (and what it does not mean)
It is not learning to use one tool...
Most of what gets sold as AI literacy is a tutorial, i.e., here is how to write a prompt, here are forty prompts for teachers, here is the shortcut that saves you an hour on lesson planning. Some of it is genuinely useful (I have used plenty of it myself!), and ALL of it expires. The tool you learn this autumn will have changed its interface, its pricing, and half its behavior before the school year ends.
Teaching the tool is like teaching phrases instead of grammar. Twenty memorized concepts will help ease a traveler through a train station and (perhaps) a restaurant. They will not help them through the first reply they did not understand, which in my experience arrives somewhere around the second sentence (yes, I know the Japanese word for bathroom, but not the words for how do I get there?).
But knowing what the tool is doing when it answers you
AI literacy is understanding the machine well enough to predict how it will fail. It transfers to every tool, every version, and every model that comes after this one. Tutorials age fast - learning how to learn doesn't.
A person who understands what a language model is doing when it produces an answer knows, without being told, that it will be excellent at structure and shaky on specifics. They know it will sound equally confident whether it is right or inventing. They know it has no idea which of those two things it is currently doing! And because they know that, they know exactly where to check.
A person who has only learned prompts knows none of this. They get speed, and no sense at all of when to be suspicious (critical thinking is difficult at speed, isn't it?).
How these tools actually work
Prediction, not knowledge
The most useful thing I can tell you about artificial intelligence takes exactly one paragraph. Ready?
A large language model does not look anything up. It is not consulting a database, searching an encyclopedia, or retrieving a stored fact. It is predicting what word is most likely to come next, over and over, based on patterns it absorbed from an enormous quantity of text. That is the entire engineering mechanism. When it writes “the capital of France is Paris,” it is not because it knows a fact about France. It is because, across everything it has read, the words that follow that phrase are overwhelmingly “Paris.”
Which works! It works ASTONISHINGLY well, most of the time, because the most statistically likely next word is very often the true one. The world’s text is mostly accurate, so prediction gets you accuracy for free...until it does not.
Why that explains almost every problem that follows
Once you have that one idea, the rest stops being mysterious.
Why does it make things up? A plausible-sounding citation is statistically likely to follow a request for a citation. The model is not lying, which would require knowing the truth in the first place! It is producing the shape of an answer, and sometimes the shape arrives without any substance inside it.
Why does it sound so sure? Confident prose is what most published writing looks like, so confident prose is what it learned to predict. Uncertainty is rare in its training material, so uncertainty is rare in its output. There is no internal signal that says this part I am sure about and this part I invented.
Why is the writing so flat? The most likely next word, chosen thousands of times in a row, is by definition the most average one. Averageness is not a bug in the output, but rather, the output working exactly as designed. (This is also why the fix is never a better prompt. It is a human being going back through and putting the very human specifics back in.)
Why is it better at some subjects than others? Because it saw more text about them. It is superb on topics with enormous written records...and unreliable on anything local, recent, or rarely written down. Ask it about the Roman Republic and it is terrific! Ask it which trailhead in your county has a working restroom in October and you should assume it is guessing. (Ask me how I know.)
Tip: when you want to know whether to trust an answer, ask yourself how much text probably exists about that exact question. The more sparse the written record, the more carefully you check.
the thing it is NOT doing
Using AI well: writing
This is where most people start using AI, and it is where we have done the most teaching on this topic.
The useful division of labor turns out to be stable. AI is good at the parts of writing that are structural, and bad at the parts that are yours. This is quite useful to know, because it shows you exactly where to spend your own labor! It can outline, summarize, reorganize, suggest what you have left out, and tell you that your third paragraph is doing the same job as your fifth. What it cannot do is supply the one detail that makes a sentence worth reading, because it was not there. It did not taste the thing, miss the train, smell the sea, or get rained on.
I wrote about this at length in How to Make AI Writing Sound Human (Without Losing Your Voice), which walks through a five-step edit pass, prompts that produce less generic drafts, and why the humanizer tools do not work.
Our companion piece, Top Signs Your Writing Was Generated by AI (and How to Fix It), catalogs the tells themselves, from transition crutches to what Bert memorably calls the emotional flatline.
I will add the thing I say most often to writers. After two decades of reading submissions, I can usually tell whether the writer was actually there by the second paragraph. What I read is rarely one big thing, but the rhythm being too even, the details being the kind found everywhere online, and the complete absence of a specific wrong turn or odd musing that a person who was actually there would have included. Your fix is to put the human back in!
Using AI well: research and verification
Almost nobody teaches this, and it is the part that does the real damage when it goes wrong.
A student who uses AI to draft an essay has a style problem that a teacher will probably catch. A student who uses AI to find sources has an accuracy problem that may survive all the way to a grade, a publication, or a courtroom. The second one is far more serious...and far less visible.
The failure has a name, and the name is unhelpfully whimsical. When a model produces a confident, well-formatted, entirely invented fact or citation, the field calls it a hallucination. The word makes it sound like a rare glitch. It is not a glitch. It is the prediction machine doing precisely what it always does, in a case where the pattern was available but the substance was not.
What this looks like in practice is ordinary and awful. You get a citation with a real journal, a plausible volume number, authors who genuinely work in that field, and a title that sounds exactly like a paper they would have written. You click through and the page does not exist. You search the journal’s archive and there is nothing. The feeling is specific, and I suspect a good many of you have had it. Instant nerves in your stomach, because you have already built an outline and three paragraphs around this specific citation, and it's not real.
So here is the working rule we recommend to any student, any teacher, and any writer. Use AI to find the shape of an answer, and use a human source to confirm the facts inside it. Every time. No exceptions! Ask it what the major debates in a field are, then go read the actual papers. Ask it to explain a concept four different ways until one fits you, then verify the specifics somewhere accountable. Treat every citation it hands you as a lead rather than a source, and go to a real database or to Google Scholar and read the actual article.
click the citation. always click the citation.
Using AI well: travel and planning
I did not expect travel to be such a clean teaching case for all of this, and yet it is, because the failure is so visible and impactful.
AI is genuinely useful for the bones of a trip. It is good at the sequencing problem, the what-order-should-we-do-this problem, the how-many-days-is-too-many problem. It will cheerfully build you a draft itinerary in about eleven seconds that would have taken you a whole evening (and it will be a decent one!).
And it will also tell you, with absolute assurance, that a restaurant is open on Tuesdays when it closed permanently two years ago. It will invent a bus route. It will give you museum hours that were true in 2022. Every one of those is the same prediction problem showing up somewhere you can actually catch it, because when you arrive at a locked door you find out IMMEDIATELY. Alas! It also lacks the depth of human experience.
Our article on AI Travel Planning: How to Use It Well on Your Next Journey covers the division of labor in detail, including the part I feel most strongly about, which is leaving deliberate room for the things an itinerary cannot contain. Our article on How AI Is Revolutionizing Travel Planning looks at the wider picture, from adaptive itineraries to virtual guides and translation.
The transferable lesson for a classroom is this. Travel is the one domain where students find out they were wrong within hours, which makes it an unusually good place to teach verification. Have them plan a trip with AI and then fact-check it against actual websites and travel writers. The errors they find will educate them more about how these tools fail than any lecture of mine ever could, and they will remember it, because they caught it themselves. (This is my favorite assignment in the whole guide!)
AI and kids: what parents and teachers are truly asking
The questions I hear from parents are almost never the ones in the think pieces. Nobody asks me about artificial general intelligence. They ask whether their nine-year-old should be talking to a chatbot at all, what it is doing with what she types, inherent harm, and whether the thing that talks back like a friend is going to become one.
Those are EXCELLENT questions and they deserve direct answers rather than think pieces! Our article on AI Chatbots: What Parents and Kids Should Know gives twelve concrete strategies, including the hard part, which is what to do when a child starts bringing emotional questions to a machine.
The one thing I would add, as a parent rather than a publisher, is that the conversation works far better than the restriction, and it is not close. A child who has been told simply not to use something will use it anyway, somewhere you cannot see, with no framework or critical thinking at all. A child who has been shown once, at the kitchen table, how confidently the thing can be wrong has learned something durable. Ask it about a topic your kid genuinely knows well: their sport, their game, your own town, and let them catch the mistake themselves. The look on their face when they find it (!!) is the entire lesson, and has an outsize educational impact.
For a longer and frankly more skeptical view, we also published Artificial Intelligence Will Impact Your Children’s Future Self-Sufficiency, which argues these tools do not belong in elementary classrooms at all. This argument deserves to be heard, reckoned with, and worked through for every parent.
let them catch it themselves
AI in the classroom
For teachers, the immediate relief of using AI is administrative. Teachers are drowning in work (both tedious tasks and creative ones). If a tool drafts your parent emails, builds a first-pass rubric, generates six differentiated versions of a worksheet, or writes the captions you never have time for, that is hours back in a week that has none of them to spare. Take the hours! Our article on The Digital Classroom: AI Tools Every Online Educator Should Use covers the practical ground.
A randomized trial at the University of Maryland, published this October, studied 2,379 undergraduates and found that the ones whose courses offered a built-in AI tutor ended up with lower grades, not higher, by roughly a third of a standard deviation in the most closely matched sections. First-generation students lost the most. This is a working paper rather than a peer-reviewed study, and only about fifteen percent of the students who had access used the tool at all. One study is not proof. It is still the kind of result that should slow down anyone who assumes adding AI to a course can only help.
The harder question to ponder is not what the teacher does with AI. It is what the students do with it, and what the teacher does about that.
The honesty question, and why we think disclosure beats detection
AI detectors do not work well enough to use for consequences. They produce false positives. They produce them unevenly, and research consistently finds they flag writing by non-native English speakers at higher rates. A tool that is wrong some of the time, in a biased direction, about something that can end a student’s academic standing, is not a tool. It is a direct liability. We said so in our humanizing article and we will repeat it here.
What works instead is boring yet effective. Ask students to disclose, make disclosure safe, and grade the thinking rather than the prose. That is it. That is the policy!
A policy that says tell me what you used and how reliably elicits honest answers, because there is no reason to lie. A policy that says you will be caught produces students who get better at not being caught, which is a genuinely worse outcome for everyone and a skill we did not mean to teach. Assignments that require a specific source, a personal observation, an in-class defense, or a visible draft history are not AI-proof, and nothing is, but they shift the work back toward the part we actually wanted to assess.
And the question worth putting to a class directly is not whether they used it. It is whether they could tell if it had been wrong. That question has a correct answer, it is teachable, and it is the one that will STILL matter in ten years!
An AI literacy reading list
If you want to go deeper than a single article, start with the book: Prompting Originality: The A.I. Handbook for Humans, which impressed me, as it does the rare thing of treating creative AI use seriously while keeping human judgment at the center of it rather than as an afterthought.
Beyond that, I would send you to the primary sources, rather than the commentary. Read your own district’s or institution’s AI policy, which a surprising number of educators and parents have never actually opened. Read what the model companies publish about their own systems’ limitations, which is more candid than you would expect. And read at least one piece by someone who thinks this is all a mistake, because a position you have never seen argued against is not a position you understand. You just have not met the objection yet.
Where we stand on all of this
We have published AI articles from several writers with genuinely different views, and I think readers deserve to know where Wandering Educators stands:
AI is a tool, and tools are worth learning! We are not going to pretend this is a passing enthusiasm, and we are not going to write the article about clutching pearls about technology. These systems are useful...and they are not leaving.
Disclosure beats detection, every single time. Build the honest path and then make it the EASY one. The surveillance approach does not work; it punishes the wrong students and it teaches evasion.
Verification is the skill that outlasts every one of the tools. Every model will change. The habit of asking how would I know if this were wrong will still be the thing that separates a person who uses AI well from a person it embarrasses. If a school teaches one single thing about AI, please let it be that one!
The judgment is the part you cannot hand over. A machine can draft your lesson, but it cannot know that this particular class needs twenty extra minutes on this particular day. It can summarize a culture, but it has never been anywhere! It has never been a guest in someone’s home, never been gently corrected by a person who grew up there, never sat at a table where it did not know the rules, never tried to converse in a new language. (This is a travel website, so forgive me the hobby horse, but it genuinely matters.) The part of your work that is irreplaceable is the part that comes from having been present. Honestly, that is a rather cheerful thing to know about your own profession in a year like this one.
this part does not automate
Frequently asked questions about AI literacy
What is AI literacy? Understanding what AI tools actually do, where they reliably fail, and how to check their output, rather than memorizing how to operate any one of them. It is closer to media literacy than to a software skill.
Why does AI literacy matter more than learning the tools? Because the tools change constantly and the underlying behavior does not. Someone who understands that these systems predict text rather than retrieve facts can pick up any new model and immediately know where to be careful.
How do you teach AI literacy? Show, rather than tell! Have students use a tool on a subject they already know well and catch its errors themselves. Then build verification into assignments as a graded step rather than an afterthought.
What should students know about AI before they use it for schoolwork? That it will sound equally confident whether it is right or inventing, that it can fabricate sources that look completely real, and that they remain responsible for every word they submit regardless of what produced it.
How do you talk to kids about AI? Directly, early, and with a demonstration, rather than a rule! Let them find a mistake themselves. A child who has personally caught the machine being wrong has a defense that no restriction can provide. Hands on learning will always imprint the most.
Are AI detectors reliable? No, not reliably enough to base consequences on. They generate false positives and the research repeatedly finds they do so unevenly, flagging non-native English speakers more often. Use disclosure policies instead.
We will keep adding to this Compendium as the ground shifts, which it will. If you are a teacher who has worked out something that actually functions in a real classroom with real students, we would LOVE to publish it!
What has AI actually changed about your work this year, and what has it not touched at all?
Everything we have published on AI and education
How to Make AI Writing Sound Human (Without Losing Your Voice)
Top Signs Your Writing Was Generated by AI (and How to Fix It)
AI Chatbots: What Parents and Kids Should Know
The Digital Classroom: AI Tools Every Online Educator Should Use
AI Travel Planning: How to Use It Well on Your Next Journey
How AI Is Revolutionizing Travel Planning
Read This: Prompting Originality: The A.I. Handbook for Humans
Artificial Intelligence Will Impact Your Children’s Future Self-Sufficiency
Jessie Voigts is the founder and publisher of Wandering Educators. She has a PhD in International Education, has lived and worked in Japan and London, and traveled around the world. She is constantly looking for ways to increase intercultural understanding, and is passionate about study abroad and international education.