A colleague mentioned at a department meeting that our district was committed to “building AI literacy across all content areas.” Everyone nodded. I nodded too.

Nobody asked what it meant.

This wasn’t unusual. “AI literacy” is a phrase that works the way “college-ready” and “21st-century skills” work: everyone uses it, everyone assumes everyone else understands it, and nobody means quite the same thing. That ambiguity was easy to live with when AI literacy was a niche concern. It’s not easy to live with now.

An AI literacy educator named Bette Ludwig published a piece this week that made the problem unusually visible. She’d written a short post on LinkedIn about assessment — specifically, that making AI use required doesn’t automatically make it educational. A reasonable point, and a specific one. What followed was a comment section that bore almost no relationship to her argument. Some readers debated homework. Others argued about cheating. Someone called her a Luddite. She was making a point about assessment design. Her readers were having at least five different conversations.

Her conclusion: we don’t have a shared baseline understanding of AI. Not even close.

She’s right, and the problem runs deeper than a noisy comment section.

Look at the vocabulary alone. “AI literacy” is the incumbent phrase, but “AI fluency,” “AI-aware,” and “AI-ready” have all entered the conversation in the last year. Each arrives carrying slightly different assumptions about what students are supposed to be able to do. Fluency suggests production — you can use AI tools effectively. Literacy suggests comprehension — you understand what AI is doing and what it’s not. Aware suggests basic familiarity. Ready suggests some kind of preparation for a world in which AI is ubiquitous.

Those aren’t the same skill sets. But in most policy documents and professional development sessions, the terms are used interchangeably, as though the important thing is that the phrase appears somewhere in the document. It’s not. The important thing is that everyone in a building means the same thing when they say it, and right now they don’t.

The consequences show up in policy.

Ludwig gives a pointed example: UC Berkeley Law School recently announced guidelines that ban AI use for a range of activities, including brainstorming. As she notes, there is no way to know whether a student brainstormed with AI unless they admit to it. The brainstorming ban reads like a clear rule. It is, in practice, unenforceable.

K–12 schools are writing the same kinds of policies. “Students may not use AI for original work” — what counts as original? “AI use is permitted with teacher guidance” — what constitutes guidance? “Students should demonstrate AI literacy” — by when, measured how, at what level? These are policy statements that produce a hundred different local interpretations, each one defensible, none of them coordinated.

The writing isn’t the problem — the concept is. You can’t write enforceable rules around something you haven’t defined, and the field hasn’t agreed on a definition yet.

There’s also the teaching problem.

If you don’t have a working definition of AI literacy, you can’t sequence it. You don’t know what an 8th grader should know that a 5th grader doesn’t. You can’t tell a parent what their student is learning or why it matters. You can’t design assessments that measure something real.

Ludwig flags a study where law professors were asked to choose which explanation better answered a student’s question — and chose the AI-generated explanation 75% of the time, without knowing which was which. She’s not using this to argue that AI is better at explaining the law. She’s using it to make the incoherence visible: here’s an institution banning AI use in one context, while in another context the professors themselves prefer AI outputs. The same technology, in the same professional ecosystem, producing opposite institutional responses — because nobody is working from the same frame.

K–12 teachers live in this incoherence daily. Your district may have one AI policy. Your building principal may have a different interpretation of it. Your department may have informal norms that don’t match either. The parent in your inbox is worried about something else entirely — usually job readiness or cheating, rarely the critical-thinking questions that tend to motivate teachers. And the student in front of you has probably already worked out what they can and can’t get away with, independent of all of the above.

You’re not navigating a disagreement. You’re navigating four different conversations happening under the same roof.

What you can do about it.

I’m not arguing for a national definition before anyone moves. That’s how nothing gets done for a decade.

What I am arguing is that the conversation in your building — and in your classroom — needs more precision than the phrase “AI literacy” provides. Not a formal definition, but something operational: what does an AI-literate student look like on a specific task? What can they do that a student without those skills can’t? What counts as a real signal of understanding, not just behavior that looks like the right behavior?

Those questions are harder. They also produce answers you can work with.

One version of this that I’ve found useful: stop asking what students know about AI and start asking what students can do when AI gets something wrong. Can they identify a plausible but incorrect answer? Can they explain why it failed? Can they use that failure to do better work than they would have done without the tool? That’s a skill set. It’s teachable, it’s assessable, and it doesn’t require anyone to agree on a definition of literacy first.

The definition vacuum gets filled by whoever happens to be in the room. If it’s a tech coordinator, “AI literacy” means tool proficiency. If it’s a department chair worried about academic integrity, it means detection and prevention. If it’s a parent, it might mean digital safety or job preparation. None of these are wrong. But they produce very different classroom priorities, and students experience that incoherence directly.

Ludwig’s pencil-sharpening post wasn’t supposed to generate a debate about five different things. It did anyway. That’s useful information — not just about LinkedIn comment sections, but about any conversation where the words being used carry different meanings for different people.

In education, those conversations happen in board meetings, in department meetings, in parent-teacher conferences, and in the first five minutes of September when you’re explaining your classroom AI policy to 30 students who are all bringing their own assumptions about what you mean.

Takeaway for Teachers

Before your next conversation about AI literacy — with a colleague, with admin, with a parent — ask them to describe what an AI-literate student looks like doing something specific. Not the definition. The behavior. “A student who uses AI responsibly” is a definition, and it raises more questions than it answers. “A student who can identify when an AI answer is plausible but wrong, and explain why” is a behavior you can teach toward and assess. You’ll find out immediately whether you’re in the same conversation.

If you’re thinking through questions like this — what AI literacy means for your students and what no algorithm can tell you about them — my book goes deeper. The AI Doesn’t Know Your Students is available on Amazon and at shouldiuse.ai/book.

Get the next piece before it’s published. One email, most weeks — what I’m seeing in the classroom, nothing else.

David Jacobson is a high school history teacher. He writes about AI, education, and the messy intersection of the two at shouldiuse.ai.