What should an AI literacy course cover? A checklist before you buy

A serious AI literacy course covers how the systems work, where they fail, how to evaluate an output, and who stays responsible for a decision. If a course is mostly tool tutorials, it will be obsolete before the year is out, because the tools change and the judgment does not.

The checklist below is drawn from the language two states have actually put in statute, which is a better specification than any vendor's marketing.

The twelve things to look for

1. What these systems are. That a language model predicts likely text from patterns rather than retrieving facts. Everything else follows from this one idea.

2. Capabilities and limitations, explicitly. Utah's statute leads with exactly these two words, and a course that only covers capabilities has taught half the subject.

3. Training data and its effects. Idaho asks students to analyze how "the variety and quality of training data can impact the functionality and outcomes of an AI system." Where the data came from, what is missing from it, and why that is a property of the method rather than an oversight.

4. Hallucination, taught as mechanism. Not "sometimes it gets things wrong" but why a fabricated citation looks exactly as good as a real one.

5. Evaluating an output. Checking a claim, spotting an invented source, and recognizing when you cannot check — and that your inability to check is itself the finding.

6. Human responsibility. Idaho: "why humans remain responsible for important decisions and outcomes." This is the spine of the ethics content.

7. Privacy and what you type in. That a prompt is not a private space, with the reason attached.

8. Fairness and harm. Evaluating tools "for fairness, accessibility, transparency, explainability, privacy, safety, intellectual property, and accuracy, including how they may help and harm different groups."

9. Academic integrity as a thought, not a rule. Maintaining ownership of your own learning and acknowledging what a tool contributed.

10. Machine versus human interaction. Recognizing "that AI lacks sentience, emotions, and personal beliefs" — the content that actually bears on companion products.

11. Societal and workforce impact. How AI creates, replaces and transforms jobs, and its effects on communities.

12. Deliberate stress-testing. Idaho's best standard: challenging models "by purposefully experimenting with complex prompts to discover their limitations, recognize potential errors, and understand when outputs are unreliable." Most courses skip this because it is hard to assess. It is the most valuable hour in the subject.

Five questions that expose a tool tutorial

What in this course would still be true if the leading model changed tomorrow? If the honest answer is "not much," it is training, not education.

Where does a student practice being wrong? A course with no activity where the learner catches a confident error has not taught evaluation.

Is ethics a module at the end? Six weeks of mechanism plus one week of "and now, the ethics" fails, because the consequence is the reason the mechanism matters. It should be braided through.

How much adult preparation does one lesson take? Ask for a real number. A course requiring ten hours a week of prep will be abandoned by November whatever it costs.

Where does student work live? A course whose work sits on the vendor's servers is a course you lose when the subscription lapses or the company pivots.

What a course cannot honestly claim

That it "meets" a state requirement. In most states there is no published standard to meet — Utah's are not due until 2027-28 and Idaho's approved standards still need legislative sign-off. What a publisher can honestly offer is a mapping onto the statutory language, and you should ask to see it rather than take the claim.

Be equally wary of "federally aligned." There is no federal AI curriculum standard.

The test at the end

Whatever you buy, judge it by one thing twelve weeks in. Hand the learner a confident, well-written, wrong paragraph in a subject they know. Can they find the error, explain why the system produced it, and say what they would check? That is the whole outcome. A course that gets a student there has done its job; one that cannot, has not, whatever it covered.